Advisories

Any user that has enabled the OTP 2FA can have their TOTP reused during the standard 30 second validity window.

The gin i18n middleware in nginx-ignition's API server runs in front of every HTTP request and calls golang.org/x/text/language.ParseAcceptLanguage on the raw Accept-Language header without imposing any size or shape filter. The underlying parser has quadratic-time behaviour on long lists of malformed language tags. The CVE-2022-32149 guard that golang.org/x/text added in v0.3.8 caps the number of - characters in the input at 1000, but it does not cap _ characters even …

The k8saudit plugin's per-container fields (ka.req.pod.containers.*) and the shipped k8s_audit_rules.yaml evaluated only requestObject.spec.containers. Security-relevant settings on a pod's initContainers or ephemeralContainers were not inspected, so the shipped Create Privileged Pod rule did not fire for a privileged container placed in either list.

The k8saudit plugin's per-container fields (ka.req.pod.containers.*) and the shipped k8s_audit_rules.yaml evaluated only requestObject.spec.containers. Security-relevant settings on a pod's initContainers or ephemeralContainers were not inspected, so the shipped Create Privileged Pod rule did not fire for a privileged container placed in either list.

The k8saudit plugin's per-container fields (ka.req.pod.containers.*) and the shipped k8s_audit_rules.yaml evaluated only requestObject.spec.containers. Security-relevant settings on a pod's initContainers or ephemeralContainers were not inspected, so the shipped Create Privileged Pod rule did not fire for a privileged container placed in either list.

The k8saudit plugin's per-container fields (ka.req.pod.containers.*) and the shipped k8s_audit_rules.yaml evaluated only requestObject.spec.containers. Security-relevant settings on a pod's initContainers or ephemeralContainers were not inspected, so the shipped Create Privileged Pod rule did not fire for a privileged container placed in either list.

The k8saudit plugin's per-container fields (ka.req.pod.containers.*) and the shipped k8s_audit_rules.yaml evaluated only requestObject.spec.containers. Security-relevant settings on a pod's initContainers or ephemeralContainers were not inspected, so the shipped Create Privileged Pod rule did not fire for a privileged container placed in either list.

POST /api/users/onboarding/finish is registered as anonymous (unauthenticated) and creates a user with full ReadWrite admin permissions. Because the handler uses a check-then-act (TOCTOU) pattern between the "onboarding already completed?" check and the user-creation write, with no atomic guard, a remote unauthenticated attacker who can reach an instance in its pre-onboarding state can create an administrator account for themselves — and concurrent requests can create multiple admin accounts in a single …

A bearer token with only pull and push scopes can successfully delete manifests and blobs from a zot registry. The bearer authentication handler maps all non-GET/HEAD HTTP methods, including DELETE, to the "push" action, and the DistSpecAuthzHandler middleware is bypassed entirely for bearer-authenticated requests. This allows any client holding a push-only bearer token to delete arbitrary manifests and blobs within the token's repository scope, in violation of the Docker Distribution …

A containerized MCP server running with the default network permission profile (insecure_allow_all: true) can reach host-local services via host.docker.internal. This includes the ToolHive API itself, other ToolHive-managed MCP server proxies, and any other service listening on the host's localhost. Combined with the unauthenticated ToolHive API and MCP proxy endpoints, this enables a compromised or malicious MCP server to perform lateral movement without any container escape.

Special:FacetedSearch cstate hidden inputs enable reflected XSS (residual of CVE-2025-10354)

The api.php?action=smwtask API module performs no authorization check. The equivalent maintenance interface in the web UI (Special:SMWAdmin) requires the smw-admin right, but the API module that backs several of the same operations enforces nothing. An unauthenticated visitor can therefore retrieve internal Semantic MediaWiki database statistics and reach state-changing maintenance operations that are intended to be administrator-only.

The SemanticMediaWiki extension inserts the unsanitized value of a data attribute into the DOM as HTML, allowing for stored XSS through wikitext.

Failure mode When headers=plain, table header text was emitted into <th> via a raw HTML path. User-controlled mainlabel content could therefore become executable HTML. Remediation TableResultPrinter now applies output-context escaping before passing plain headers to the table renderer. The fix is limited to the HTML/plain-header branch so safe rendering modes are unaffected. Why this is the right layer The header value is not a structural token; it is display data. …

Failure mode The value parameter was reflected back into rendered output and error messaging paths without enough output-context encoding. Remediation The form value is escaped before it is placed back into the input field. Derived error messages are also escaped before being rendered into HTML. Maintenance note Do not treat error text as trusted just because it originated from validation. Validation output can still be attacker-influenced.

Failure mode Special:URIResolver resolves its user-controlled subpage to a MediaWiki title and issues an HTTP 303 redirect to $title->getFullURL() without validating the resolved target. A crafted subpage can make that target point off-host: an interwiki prefix redirects to the foreign wiki (for example Special:URIResolver/mw-3AFoo, which decodes to mw:Foo, redirects to https://www.mediawiki.org/wiki/Foo), and where the resolved URL carries an authority the target can even embed user:pass@host credentials. The result is an …

Query debug output XSS Failure mode Semantic MediaWiki's query debug output (format=debug, or the debug request parameter on Special:Ask) is assembled by SMW\Query\DebugFormatter and emitted as raw HTML. Several of its sinks apply no output-context encoding, so attacker-controlled query input is reflected into the page without escaping: buildHTML() echoes the re-serialized ASK query string escaping only [; <, >, ", ' pass through. prettifySQL() returns the generated SQL verbatim. Query …

Failure mode sep was inserted verbatim into the HTML that joins a table cell's values. This made it possible to inject HTML through the separator value. The same unsanitised table HTML is produced both for the standard Special:Ask render and for its raw request output (request_type=raw), so the injection was reachable without authentication. Remediation In all non-wiki output modes (HTML, raw request, file), sep is escaped unless it is a …

Reflected XSS via a forged cursor pagination token Failure mode Special:Ask accepts a cursor query parameter for keyset pagination (added in 7.0.0). The token is decoded by CursorEncoder, which is an unsigned base64url-encoded JSON blob, so its contents are fully attacker-controlled. When the cursor's sort anchor does not match the request's sort= / order=, QueryCreator::applyCursorIfRequested() builds an error message by interpolating the attacker-controlled sort_prop / sort_order values into a raw …

A malicious website can use DNS rebinding to control a developer's local process-compose MCP SSE listener when MCP SSE is enabled. The vulnerable path accepts browser-origin requests before any Host validation, Origin validation, or caller-secret check, then dispatches the requests into process-compose MCP tools. This advisory covers https://github.com/F1bonacc1/process-compose, confirmed at commit d56aa59df04b72f8644811ac581a051bec05e485. The issue is in the MCP SSE transport, not the Gin REST API. The REST API token middleware …

When Perses is using the file system database, on the list endpoints, the project value is bound from the request into the resource Query struct and is never validated against directory-traversal characters (validation/Flatten only runs for Create/Update bodies, not list queries). The path is then used to retrieve files in the database directly. Attacker can read arbitrary YAML/JSON files from the server host and can bypass the security constraints to …

What kind of vulnerability is it? An authenticated user who is only a viewer on project team-a requests GET /api/v1/projects/team-a/dashboards?project=finance-secret (or simply GET /api/v1/datasources?project=finance-secret) and receives the full list of the finance-secret project's dashboards and datasource specifications, despite having no role on that project. This defeats Perses' project-level tenant isolation for all project-scoped read resources. Who is impacted? Any authenticated user reads every project's dashboards, datasources, variables across tenants.

The datasource proxy authorizes the caller on the Datasource scope, then resolves and decrypts any Secret named in the request body with no Secret-scope check. Datasource and Secret are distinct, independently grantable role scopes, so an operator can grant datasource access without secret access. The proxy and the service to create a datasource does not verify that the operator has the correct right. A user holding only GlobalDatasource:create can create …

The service downgrade implementation in app/Livewire/Services/Upgrade.php::doUpgrade() executes a proration calculation and a subsequent user credit refund without any transactional safety or database locks. The only concurrency check relies on an unisolated database query checking for existing pending upgrades. Because there is no active database transaction or pessimistic row-level lock spanning this validation and the subsequent credit update, concurrent downgrade requests can bypass the guard simultaneously. This allows a user to …

Any user who can edit their own user profile or any other document can execute arbitrary script macros including Groovy and Python macros that allow remote code execution including unrestricted read and write access to all wiki contents. The reason is that rendering output is included as content of HTML macros without further escaping and it is thus possible to close the HTML macro and inject script macros that are …

The Opencast Paella player renders caption cue text into innerHTML without escaping. The captions canvas clears _captionsContainer.innerHTML and then appends each active cue with _captionsContainer.innerHTML += cue, so HTML inside a WebVTT or DFXP cue becomes live DOM and executes in the Opencast origin. The caption track is read from any media package element with a captions/* flavor and is served, with the player manifest, to anonymous viewers through /search/episode.json. …

The Opencast Paella player renders caption cue text into innerHTML without escaping. The captions canvas clears _captionsContainer.innerHTML and then appends each active cue with _captionsContainer.innerHTML += cue, so HTML inside a WebVTT or DFXP cue becomes live DOM and executes in the Opencast origin. The caption track is read from any media package element with a captions/* flavor and is served, with the player manifest, to anonymous viewers through /search/episode.json. …

In affected versions, the URL of a remote MCP server is attacker-controlled at registration and is fetched server-side with no validation of the destination. There is no guard against loopback, link-local, RFC1918 private ranges, or the cloud metadata endpoint (169.254.169.254), so a use with the Power User, Power User Plus, or Admin role can coerce Obot into making requests to internal services and to the cloud instance metadata service, and …

In affected versions, an unauthenticated attacker could register an OAuth client with an arbitrary external redirect URI, and the authorization flow would auto-complete without a consent screen. If a logged-in victim visited a crafted authorization URL, an authorization code was delivered to the attacker's redirect URI and exchanged for an access token. That token was minted with the victim's full set of groups, so it could be used as a …

In affected versions, enabling registry authentication (OBOT_SERVER_ENABLE_REGISTRY_AUTH=true) does not actually protect the MCP Registry endpoints under /v0.1/* — they remain readable by unauthenticated callers.

The Mnemosyne sync server's authentication check decoded JWT bearer tokens but never verified their HMAC-SHA256 signatures. Any well-formed token was accepted, allowing an unauthenticated attacker to impersonate any user and read or modify their sync data. Severity: Critical CVSS 3.1: AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:N = 9.1 Assumes the sync server endpoint is network-reachable. If your deployment is localhost-only, the score drops substantially and severity becomes High or Medium depending on local exposure. Confirm …

MdPreview interpolates a fenced-code language into HTML attributes without escaping it. A crafted info string therefore executes JavaScript even when the shipped XSSPlugin is enabled.

lmdeploy <= latest contains a code injection vulnerability in lmdeploy/pytorch/config.py line 620 that allows an attacker to execute arbitrary Python code by publishing a malicious HuggingFace model with a crafted quantization_config.quant_dtype value. When a user loads the model with lmdeploy, the quant_dtype is passed to eval(f'torch.{quant_dtype}') without any validation.

LMDeploy's PyTorch DistServe/PD-disaggregation control plane used recv_pyobj() to deserialize messages received through a ZeroMQ PULL socket. PyZMQ implements recv_pyobj() using Python pickle deserialization, which can execute arbitrary code while reconstructing an object. The peer address used by the receiver was supplied through the POST /distserve/p2p_connect HTTP endpoint. An attacker who could reach an affected DistServe API server could cause the server to connect to an attacker-controlled ZeroMQ endpoint and deserialize …

The URL checking logic in lmdeploy has a logical flaw that could be bypassed by attackers, leading to SSRF attacks.

The kcp front-proxy fails to strip client-supplied identity headers before forwarding requests to shards. Any authenticated tenant can inject their own X-Remote-Group and X-Remote-Extra-* headers, which the shard trusts as a verified identity assertion — allowing a low-privilege user to escalate to cluster administrator (system:masters) and read, write, or delete resources in any workspace on the shard. This is a complete multi-tenant isolation and authorization bypass.

Moquette MQTT Broker fails to enforce ACL write permission checks when publishing Will (Last Will and Testament) messages on behalf of disconnected clients. All normal PUBLISH paths (receivedPublishQos0, receivedPublishQos1, receivedPublishQos2) correctly invoke authorizator.canWrite() before publishing, but the Will message publishing path (fireWill() → publishWill() → publish2Subscribers()) completely bypasses this authorization check. This allows an unauthenticated attacker (when allow_anonymous=true, which is the default) to inject arbitrary messages into any ACL-protected topic …

Before 2.3.1, the legacy .doc renderer emitted document hyperlink targets after HTML escaping but without a URL-scheme allowlist. A crafted .doc could therefore render a live javascript:, vbscript:, data:, or similarly unsafe link. Script could execute in the embedding origin if a viewer clicked it.

Before 2.3.1, the legacy .doc renderer emitted document hyperlink targets after HTML escaping but without a URL-scheme allowlist. A crafted .doc could therefore render a live javascript:, vbscript:, data:, or similarly unsafe link. Script could execute in the embedding origin if a viewer clicked it.

frain-dev/convoy (all versions up to and including v26.6.2, no patch available) lets any authenticated caller who is authorized on at least one project read ANY OTHER project's "Source" record by ID via GET /api/v1/projects/{projectID}/sources/{sourceID} – regardless of whether that Source actually belongs to the project named in the URL. The response includes the Source's full PubSub broker configuration in plaintext (AMQP/Kafka/SQS/Google credentials), with no redaction.

Capsule lets a cluster administrator forbid specific metadata keys that tenant owners must not place on their own resources: Tenant.spec.namespaceOptions.forbiddenLabels / forbiddenAnnotations (namespaces), Tenant.spec.serviceOptions.forbiddenLabels / forbiddenAnnotations (Services), and the cluster-wide forbidden worker-node labels/annotations. These lists are an isolation control — they exist to stop a tenant owner from setting metadata that other controllers or admission plugins key on (Pod Security Admission labels, kubernetes.io/metadata.name, LoadBalancer/externalIP service annotations, scheduler annotations, vendor labels …

A validation bug in internal/webhook/tenant/validation/forbidden_annotations_regex.go allows an invalid ForbiddenAnnotations.Regex value to bypass Tenant admission on update. The webhook compiles ForbiddenLabels.Regex for both labels and annotations, so a malformed annotations regex can be persisted. Once stored, namespace admission later evaluates the bad regex through pkg/api/forbidden_list.go, where regexp.MustCompile can panic and cause admission failure.

A parameter order bug in internal/webhook/tenant/validation/hostname_regex.go causes the hostnameRegexHandler.OnUpdate webhook to validate the old Tenant object's AllowedHostnames.Regex instead of the new one being submitted. This allows an invalid (malformed) regex to bypass admission validation and be persisted to etcd, causing a Denial of Service for all Ingress operations within the affected tenant.

Caddy v2.11.3 — Three vulnerabilities in handler/placeholder layer Tested against: caddy:2.11.3 (official Docker image, SHA verified at runtime) Reproduction environment: Docker Desktop 4.73.1 / Engine 29.4.3 on Windows 10 host, isolated containers, no network egress required for any of the exploits This advisory bundles three independent issues discovered together during a source review of the placeholder/replacer layer. Each issue has been reproduced end-to-end against the unmodified caddy:2.11.3 image with the …

Services using internationalized (non-ASCII) domain names are potentially vulnerable to TLS connections made from AnyIO's connect_tcp() or directly via TLSStream.wrap() where the connection has (through other means) been hijacked and redirected to a malicious server. The attacker would obtain a legitimate certificate using the IDNA 2003 encoded version of the domain name and offer it to the connecting client, making the certificate validate properly on the client's end.

AnyIO 4.14.0 accepts the POSIX extra_groups argument on anyio.run_process() and anyio.open_process(), but open_process() forwards the wrong variable to the backend: when extra_groups is not None, it assigns kwargs["extra_groups"] = group instead of extra_groups. As a result, callers cannot reliably clear or set supplementary groups for child processes. In a disposable Linux container, Python's subprocess.run(…, extra_groups=[]) clears a synthetic parent supplementary group list, while anyio.run_process(…, extra_groups=[]) preserves the parent groups. If …

AnyIO starts process-pool workers with stderr connected to a pipe but never drains that pipe. The worker redirects stdin and stdout to /dev/null to protect its protocol, but does not redirect stderr even though the documentation says all three standard streams are redirected. Worker code that writes enough attacker-influenced data to stderr can fill the pipe, block before returning the stdout protocol response, and wedge the awaiting process-pool call. Anyone …

The telemetry subsystem embeds a hardcoded auth token ("secret") in the public source and transmits raw CLI arguments—including –secret, –jwt_secret, and –http_rpc_secret values—to a third-party telemetry endpoint.

The Pusher-compatible REST API includes body_md5 in the HMAC signature string but never computes or verifies the MD5 of the received HTTP body, allowing anyone who observes a signed request to replay it with an entirely different body.

adm-zip allocates an entry's output buffer from the declared uncompressed size (central-directory size field) before validating it against the actual data. A tiny crafted ZIP that declares a huge uncompressed size forces a multi-gigabyte allocation from a few bytes.

Python's string format functionality allows someone controlling the format string to "read" objects accessible (recursively) via attribute access and subscription from accessible objects. Those attribute accesses and subscriptions use Python's full blown getattr and getitem, not the policy restricted AccessControl variants getattr and getitem. This can lead to critical information disclosure. The AccessControl package already guards against direct access to the formatting functions on string instances, but these mitigations did …

Wire's protobuf decoders did not consistently validate attacker-controlled length-delimited sizes against the current reader bounds before computing cursor, limit, or pointer positions. In the Kotlin runtime, ProtoAdapter.decode(ByteArray) and ProtoAdapter.decode(ByteString) use the ProtoReader32 fast path implemented by ByteArrayProtoReader32. In ByteArrayProtoReader32.internalNextLengthDelimited(), Wire read an untrusted varint length into an Int and rejected only negative values. A length such as 2147483647 is non-negative, so it passed that check, but pos + length overflowed …

Current vLLM main lets an inference request choose the PyNvVideoCodec GPU video decoder through media_io_kwargs.video.video_backend, but engine GPU memory reservation is computed only from static startup configuration and VLLM_VIDEO_LOADER_BACKEND. If the server starts with the default OpenCV/software backend and no –mm-ipc-gpu-memory-gb budget, a client can still route a video request into the PyNvVideoCodec path after startup, causing frontend CUDA-context, decoder-surface, and decoded-frame GPU allocations that were not carved out of …

[!IMPORTANT] Only instances running on the SQLite driver (better-sqlite3) are affected; SQLite is usually used in development/testing backend, so production deployments on PostgreSQL or MySQL/MariaDB are unaffected. The StringOperators.regex filter exposed on the public Shop GraphQL API is evaluated inside the Node.js event loop via a synchronous SQLite user-defined function (UDF). Supplying a catastrophically backtracking pattern blocks the entire event loop, causing a complete denial of service with no authentication …

The Shop API products, collections and facets queries inject a mandatory filter to restrict results to publicly-visible entities (Product.enabled = true, Collection.isPrivate = false, Facet.isPrivate = false). This injected guard was combined with the caller-supplied filter using the caller-controlled filterOperator. When a caller sets filterOperator: OR, the guard is OR-combined instead of AND-combined, so a caller can retrieve disabled products and private collections/facets by supplying a predicate that matches them.

The dashboard's RichTextDescriptionCell "strips HTML" from an entity's description by assigning it to a live element's innerHTML and reading back textContent. This pattern still executes active markup: a description containing <img src=x onerror=…> runs script when the element is parsed (image resource loads even on a detached node in Chromium/Firefox, firing onerror). Because description is an admin-settable field shown in multiple list views, a lower-privilege administrator can store a payload …

ExternalAuthenticationService.createCustomerAndUser() links a newly-presented external (OAuth/social) authentication method to a pre-existing User account selected purely by email-address match, and it does so without requiring config.verified === true. If any configured AuthenticationStrategy forwards an email that was not proven to belong to the external identity (the classic email_verified omission — common with custom OAuth providers, or providers/strategies that don't validate email ownership), an attacker can register at that provider using a …

The Content Delivery API enforces member / Public Access protection only at the controller layer, against the node that is directly requested. When a public (unprotected) node references a protected node through a Content Picker or Multi-Node Tree Picker (including those nested inside Block List, Block Grid, or Rich Text Editor blocks), the Delivery API expands and serializes the protected node with no access check applied. Similarly, configuration based restrictions …

@tinacms/auth's isAuthorized(req) decides authorization by validating the caller's bearer token against https://identity.tinajs.io/v2/apps/${req.query.clientID}/currentUser, where the clientID comes from the request and is never compared to the site's own configured TinaCloud app id. The function answers "is this token a verified user of whatever app the caller named?" instead of "is this token a verified user of THIS site?" Any TinaCloud user can create their own free app, get a valid token …

@tinacms/auth's isAuthorized(req) decides authorization by validating the caller's bearer token against https://identity.tinajs.io/v2/apps/${req.query.clientID}/currentUser, where the clientID comes from the request and is never compared to the site's own configured TinaCloud app id. The function answers "is this token a verified user of whatever app the caller named?" instead of "is this token a verified user of THIS site?" Any TinaCloud user can create their own free app, get a valid token …

devalue.parse prior to version 5.9.2 fails to reject out-of-bounds indices. Specially-crafted payloads can exploit this to cause devalue to alternate between different array representations, resulting in work that is quadratic with payload size. Applications are potentially affected if they call devalue.parse with untrusted data.

Steeltoe's /actuator/httpexchanges endpoint records and displays request URIs after passing them through MaskedUri. The masking only covers the UserInfo portion of the URI (inline user:password@host credentials) and does not inspect the query string. With IncludeQueryString enabled by default, any secrets carried in query strings (for example: OAuth tokens, password-reset tokens, signed-URL signatures, API keys) are returned verbatim in the uri field of the response and written to logs at DEBUG …

Steeltoe's Eureka discovery client deserializes the registry response as a single unit. If any registered instance contains a field value that cannot be parsed (for example, an unrecognized actionType, a non-boolean value for isCoordinatingDiscoveryServer, or a non-numeric timestamp), deserialization of the entire registry fails. All Steeltoe Eureka clients connected to that registry then receive an empty or stale instance list until the malformed registration is removed. This is a related …

Steeltoe's Consul discovery client parses the secure metadata field on each registered service instance using bool.Parse, which throws on any value other than true or false. A single service instance registered with a malformed secure value (for example yes or 1) aborts construction of the entire instance list for that service, making the service undiscoverable. When GetAllInstancesAsync is used, one malformed instance in any service can abort enumeration across all …

When Steeltoe's certificate-based authorization (UseCertificateAuthorization) is configured, the default configuration of the middleware relies on the X-Client-Cert HTTP header to identify the client certificate, without verifying private-key possession. This header is not stripped by common Cloud Foundry routers (like Gorouter or Envoy) on inbound requests.

Default SCP remote-path handling places caller-supplied paths into the command that runs scp on the server. On a shell-based server that command is interpreted by a shell, so an attacker-influenced path that is not quoted to suit that shell can execute as a command as the authenticated SSH user. SSH.NET provides ScpClient.RemotePathTransformation to control escaping behaviour (defaulting to RemotePathTransformation.DoubleQuote) but cannot guarantee safety for arbitrary remote command interpreters. This is …

Before tokenizing, selector_iter trims leading/trailing whitespace and comments by running two regexes over the whole raw selector with .search(). The trailing one, RE_WS_END = re.compile(r'{WSC}*$'), is anchored only at the end ($), not the start. Because .search() retries the pattern at every offset, a long run of whitespace or CSS comments that is not sitting exactly at the end of the string makes each retry greedily consume the run and …

soupsieve compiles CSS selector strings with a set of hand-written regular expressions. The shared IDENTIFIER sub-pattern (also embedded in VALUE, and therefore in attribute selectors) places two adjacent quantified groups over overlapping character classes: (?:[classA]|ESC)+(?:[classB]|ESC)*, where both classes match ordinary identifier characters such as a. When a selector contains a long identifier/value run that must ultimately fail to match (e.g. an attribute value with no closing ], or an identifier …

Skipper's opaAuthorizeRequestWithBody filter authorizes requests by handing the (bounded) request body to Open Policy Agent. When a body exceeds -open-policy-agent-max-request-body-size (default 1 MB), Skipper truncates it before OPA sees it. Advisory GHSA-8qqm-fp2q-v734 established that deny-on-presence / body-inspecting policies fail OPEN on oversized bodies, and its remediation instructs policy authors to guard on input.attributes.request.http.truncated_body (implemented as the top-level input.truncated_body): default allow := false allow if { input.truncated_body == false

Sanic's HTTP/1.1 chunked-body handling does not fully consume the trailer-part after the terminating 0\r\n chunk. Because of that, attacker-controlled bytes left in the connection buffer after the first chunked request can be interpreted as the start of a new HTTP request on the same keep-alive connection. In the attached verified proof, a single outer POST / request that correctly returns 405 Method Not Allowed is followed, within the same TCP …

The rmcp crate's StreamableHttpClientTransport forwards caller-supplied custom HTTP headers (such as X-API-Key, X-Auth-Token, Api-Key) to cross-origin redirect targets. The default_http_client() function builds a reqwest::Client without a redirect policy override, so the default limited(10) policy follows 307/308 redirects and forwards all per-request headers except Authorization, Cookie, and Proxy-Authorization. Custom auth headers injected via StreamableHttpClientTransportConfig.custom_headers are not classified as sensitive and are therefore forwarded verbatim to any redirect target — including an …

RestrictedPython could allow a sandbox escape when a policy exposes the standard library string module, or otherwise exposes string.Formatter, to restricted code. string.Formatter field resolution methods such as get_field can perform attribute and item traversal internally and return live object references. This can bypass RestrictedPython's normal attribute guards and may allow access to sensitive objects such as function globals, builtins, file access, or code execution primitives. Users are impacted if …

An OpenAPI or AsyncAPI description could make the split command write files outside the chosen output directory, on the machine of anyone who runs split against it. The write is constrained rather than a free file-write primitive: component data is emitted only as YAML/JSON, and code-sample files are named after the HTTP method, so an attacker can place or overwrite files at an unintended path but has limited control over …

A malformed HTTP message using Transfer-Encoding: chunked can drive React\Http\Io\ChunkedDecoder into an infinite loop, pegging a CPU core and freezing the event loop. Because ReactPHP is single-threaded, one such message stalls the entire process for every client until it is killed. Both directions are affected. ChunkedDecoder decodes chunked request bodies for React\Http\HttpServer and chunked response bodies for React\Http\Browser, so a server can be attacked by a malicious client and a …

Default configuration is vulnerable: Both requestedFrameMax (client) and legitimate servers' frameMax in Tune may be 0 Single-frame OOM: One malicious frame triggers up to ~2GB allocation (Integer.MAX_VALUE bytes) Bypasses existing protection: maxInboundMessageBodySize (introduced to cap allocations at 64MB) is entirely defeated at the frame level Different from ValueReader OOM: This is a frame-layer allocation in Frame.readFrom(), not a value-layer allocation in ValueReader.readBytes()

A data integrity and protocol corruption vulnerability exists in the AMQP client's property serialization logic. When encoding AMQP short string (shortstr) fields—such as identifiers, routing strings, and content metadata—the length of the string is explicitly cast to a fixed-size 8-bit unsigned integer (uint8). If an application provides a property string exceeding 255 bytes, the length counter silently wraps around (e.g., a length of 300 wraps to 44). As a result, …

Summary A flaw in the recvContent function allows a malicious AMQP server to trigger an Out-of-Memory (OOM) error, forcing the host operating system or container runtime to immediately terminate the client process. Vulnerability Details When receiving message content payloads, the client processes the expected size from the content header framework. The recvContent function attempts to optimize performance by pre-allocating memory for the message body based on the ch.header.Size field, which …

Summary A critical stream desynchronization vulnerability has been identified in the AMQP wire-protocol parser. When parsing a long string (readLongstr) within a table field, providing a length that exceeds the maximum signed 32-bit integer (2^31 - 1, or roughly 2.1 GiB) triggers an improper error-handling condition. The parser abruptly aborts the read and returns a success status ("",nil) without consuming the specified bytes from the underlying network buffer. This causes …

An information disclosure vulnerability exists in the AMQP client implementation's authentication handling configuration. Following a successful connection handshake, the Connection.Config.SASL field stores the Authentication implementation state used to establish the session. For standard PLAIN authentication, this state utilizes the PlainAuth struct, which defines both Username and Password as publicly exported, plaintext string fields. Because this sensitive data is retained permanently in-memory within an exported field structure, any peripheral code, internal …

A structural security weakness exists in the AMQP client's TLS configuration generator (tlsConfigFromURI). When constructing a *tls.Config object from an amqps:// connection URI, the library initializes the structure without explicitly defining the MinVersion field. While modern versions of the Go compiler toolchain (Go 1.18+) default the implicit minimum version to TLS 1.2, this security posture relies entirely on an implicit toolchain dependency. If the library is compiled using legacy Go …

A Denial of Service (DoS) vulnerability exists in the AMQP client's connection negotiation logic. The AMQP specification explicitly mandates a strict minimum frame size of 4096 bytes to prevent pathological packet fragmentation. While the library defines a frameMinSize = 4096 constant, the connection negotiation loop fails to enforce this boundary, blindly accepting whatever maximum frame size (FrameMax) the server advertises during the handshake. If a client connects to a malicious …

Summary A vulnerability in the readField function allows a malicious or compromised AMQP server to trigger an unhandled runtime panic in the client application, leading to an immediate crash of the entire process. Details When parsing incoming AMQP frames, the readField function processes byte-array fields (type tag 'x') by reading a 32-bit big-endian integer to determine the length of the data payload. // read.go:253-263 case 'x': var len int32 if …

A logic and resource exhaustion vulnerability exists in the AMQP client's Quality of Service (Qos) configuration method. The Qos function accepts signed integers (int) for the prefetchCount and prefetchSize parameters but casts them directly to unsigned integers (uint16 and uint32, respectively) when formatting the wire-level frame. If a developer passes a negative integer (such as -1) to these parameters—frequently intended as a sentinel value meaning "no change" or "no limit"—the …

A query parameter injection vulnerability exists in the AMQP client's connection URI formatting logic. When generating or parsing connection URIs, TLS-related filesystem paths (such as certificates or keys) are appended directly to the URI's query string using string concatenation rather than secure URL encoding via functions like url.QueryEscape. If an application handles a TLS file path containing special character delimiters (such as & or =), these characters are interpreted as …

PocketBase already has builtin panic-recover middleware for the regular requests handling but it doesn't cover panics in internal child/worker goroutines which in some situations could cause termination of the server process. To prevent this from hapenning all existing internal worker functions were wrapped with the new helper routine.SafeWrap(f) (auto recovers and returns any eventual panic as regular error). The vulnerability was reported by @gigioneggiando and it is strongly recommended to …

A flaw in the CORS plugin allowed the incoming request's Vary header to be reflected into the response, letting a client influence a header that should be controlled solely by the server.

oras-go's pagination helper parseLink() in registry/remote/utils.go follows the Link response header from a registry without validating the URL's host or scheme. When a malicious registry returns a Link header containing an absolute URL pointing to an arbitrary host (e.g., a cloud metadata endpoint), the client makes GET requests to that host from the victim's network. This affects all pagination-based listing operations: Tags, Referrers, and Repositories (catalog).

The content/file.Store in oras-go v2 unpacks OCI layer tarballs when a descriptor carries io.deis.oras.content.unpack=true. The extraction routine validates symlink targets purely lexically (filepath.Join) and, for regular files placed directly at the extraction root, skips the parent-symlink Lstat walk. A malicious tarball can plant a chain of symlinks whose lexical target stays inside the extraction root but whose kernel-resolved target is any absolute path, then write through it with a follow-up …

The OTLP log gRPC exporter loads TLS settings from environment variables but does not apply them when creating gRPC transport credentials. Operators who rely on OTEL_EXPORTER_OTLP_LOGS_CERTIFICATE, OTEL_EXPORTER_OTLP_CERTIFICATE, or related client certificate variables for CA pinning or mTLS get a connection that falls back to system roots and omits the env-supplied client certificate. A network attacker who can intercept or spoof the collector connection with a system-trusted certificate can read or …

OpenTelemetry Go versions 1.5.0 through 1.44.0 can include trace exporter endpoint configuration in an internal diagnostic log emitted when an SDK TracerProvider is created. The default OpenTelemetry logger does not emit this event. Exposure requires an application to install a logger that enables OpenTelemetry's internal Info-level diagnostics and for someone other than the intended audience to have access to those logs. The logged configuration can disclose the address of the …

OpenTelemetry Go versions 1.5.0 through 1.44.0 can include trace exporter endpoint configuration in an internal diagnostic log emitted when an SDK TracerProvider is created. The default OpenTelemetry logger does not emit this event. Exposure requires an application to install a logger that enables OpenTelemetry's internal Info-level diagnostics and for someone other than the intended audience to have access to those logs. The logged configuration can disclose the address of the …

OpenTelemetry Go versions 1.5.0 through 1.44.0 can include trace exporter endpoint configuration in an internal diagnostic log emitted when an SDK TracerProvider is created. The default OpenTelemetry logger does not emit this event. Exposure requires an application to install a logger that enables OpenTelemetry's internal Info-level diagnostics and for someone other than the intended audience to have access to those logs. The logged configuration can disclose the address of the …

OpenTelemetry Go versions 1.5.0 through 1.44.0 can include trace exporter endpoint configuration in an internal diagnostic log emitted when an SDK TracerProvider is created. The default OpenTelemetry logger does not emit this event. Exposure requires an application to install a logger that enables OpenTelemetry's internal Info-level diagnostics and for someone other than the intended audience to have access to those logs. The logged configuration can disclose the address of the …

OpenTelemetry Go versions 1.5.0 through 1.44.0 can include trace exporter endpoint configuration in an internal diagnostic log emitted when an SDK TracerProvider is created. The default OpenTelemetry logger does not emit this event. Exposure requires an application to install a logger that enables OpenTelemetry's internal Info-level diagnostics and for someone other than the intended audience to have access to those logs. The logged configuration can disclose the address of the …

nuxt-og-image exposes an unauthenticated HTTP endpoint at /_og/d/** that base64url-decodes and JSON.parses a fonts URL segment, then passes each fonts[i].path value directly into fetch() server-side without any URL validation (no scheme allowlist, no loopback/RFC1918 block, no host allowlist, no DNS rebinding mitigation). Under the module's documented default configuration (security.strict = false, security.secret = "", restrictRuntimeImagesToOrigin = false), any caller able to reach the deployed Nuxt site can force the Nuxt …

Several code paths in Marten's LINQ provider and tenant-management internals interpolated a runtime, potentially attacker-influenced value into generated SQL as a single-quoted string literal without escaping or parameterization. A value containing a single quote (') could break out of the literal and inject arbitrary SQL. The primary vector is a Dictionary<,> indexer key used in a Where filter — Where(x => x.Attributes[key] == v) — a common "filter by attribute …

MariaDB Connector/J does not enforce allowLocalInfile=false when processing server-initiated LOCAL INFILE requests (protocol packet type 0xfb). However, exploitation is constrained: the server can only request the exact filename the client already included in its LOAD DATA LOCAL INFILE query, it cannot redirect to arbitrary paths.

A server-side request forgery in LiteLLM Proxy lets an authenticated caller redirect the proxy's outbound request to a host of their choosing by smuggling an api_base inside the user_config request body, bypassing the existing parameter guard.

@libp2p/peer-store accepts a signed PeerRecord whose envelope is signed by one peer but whose payload claims a different peer ID. The vulnerable consumePeerRecord path verifies the envelope signature, but does not verify that the envelope signer is the same peer as the wrapped PeerRecord.peerId. As a result, an attacker can sign a record with their own key while placing a victim peer ID in the payload, causing attacker-controlled multiaddrs to …

@libp2p/gossipsub StrictSign validation does not bind a supplied message public key to the claimed from peer ID when from is an RSA-style peer ID that does not inline its public key. An attacker can set from to a victim RSA peer ID, sign the message with the attacker's own private key, include the attacker's public key in msg.key, and have the message accepted as a valid signed message from the …

Kestra's Micronaut management endpoints are served on port 8081 with no authentication, even when the main API (port 8080) has basic-auth enabled. Anyone who can reach :8081 can read GET /env (full resolved environment/configuration) and mutate runtime state via POST /loggers/{name} (change log levels), among the other management endpoints. Enabling basic-auth creates a false sense of protection because the auth filter only covers /api/v1/** on 8080 and never applies to …

The Pebble template engine's http() function in Kestra OSS accepts user-controlled URLs without any validation, allowing Server-Side Request Forgery (SSRF) attacks. An unauthenticated attacker can import a malicious Flow YAML and execute it to access internal services, cloud metadata endpoints (AWS 169.254.169.254), or localhost services. The vulnerability affects all Kestra OSS deployments with default configuration.

The Pebble template engine's http() function in Kestra OSS accepts user-controlled URLs without any validation, allowing Server-Side Request Forgery (SSRF) attacks. An unauthenticated attacker can import a malicious Flow YAML and execute it to access internal services, cloud metadata endpoints (AWS 169.254.169.254), or localhost services. The vulnerability affects all Kestra OSS deployments with default configuration.

When a request returns a 500, jupyter_server/log.py logs a small JSON block of request headers. The Referer header was copied into it as-is, so a token in the Referer URL ended up in the logs in plain text.

LocalFolderExtractor validates only the final canonical file path before extraction. However, makeFile() creates intermediate directories one path segment at a time without checking whether each created directory remains inside the destination folder. A malicious RAR entry can make the final file path resolve inside the extraction destination, while causing intermediate mkdir() calls to create attacker-chosen directories outside the extraction root. This is an extraction root escape via unchecked intermediate directory …

SHCParser inflates compressed Smart Health Card JWT payloads into memory without a decompressed-size limit. An attacker who can submit SHC content for validation can craft a small compressed JWT payload that expands to a very large byte array, causing memory exhaustion or severe garbage collection pressure.

SHCParser inflates compressed Smart Health Card JWT payloads into memory without a decompressed-size limit. An attacker who can submit SHC content for validation can craft a small compressed JWT payload that expands to a very large byte array, causing memory exhaustion or severe garbage collection pressure.

SHCParser inflates compressed Smart Health Card JWT payloads into memory without a decompressed-size limit. An attacker who can submit SHC content for validation can craft a small compressed JWT payload that expands to a very large byte array, causing memory exhaustion or severe garbage collection pressure.

A malformed Smart Health Card (SHC) JWT with zip: "DEF" and an empty or truncated DEFLATE payload causes SHCParser.inflate() to loop forever. This allows an attacker who can submit SHC content for validation to pin a JVM worker thread indefinitely, causing denial of service.

A malformed Smart Health Card (SHC) JWT with zip: "DEF" and an empty or truncated DEFLATE payload causes SHCParser.inflate() to loop forever. This allows an attacker who can submit SHC content for validation to pin a JVM worker thread indefinitely, causing denial of service.

A malformed Smart Health Card (SHC) JWT with zip: "DEF" and an empty or truncated DEFLATE payload causes SHCParser.inflate() to loop forever. This allows an attacker who can submit SHC content for validation to pin a JVM worker thread indefinitely, causing denial of service.

system/config/security.yaml's Twig sandbox policy allow-lists offsetget and offsetexists for Grav\Common\User\Interfaces\UserInterface. The concrete Grav\Common\User\DataUser\User class does not filter which fields offsetGet() returns, so any sandboxed template with access to a User object can read hashed_password, secret (2FA seed), and twofa_secret directly, bypassing the redaction Grav's own code applies everywhere else.

Verified against: getgrav/grav devel branch, GRAV_VERSION = "2.0.15", file `index.php

Grav\Common\Twig\Twig::init() unconditionally puts the raw system, site, and theme config arrays into $this->twig_vars. Twig::processPage() builds the variables for the sandboxed, editor-authored page-content render by copying that same base array ($sandbox_vars = $twig_vars;) and replacing only the config key with a filtered SandboxConfig facade. The system, site, and theme keys are carried into the sandboxed render completely untouched. Because these are plain PHP arrays, not objects, Twig's sandbox SecurityPolicy (the allowed_classes/allowed_methods/allowed_properties …

A page editor without admin.super can place an event handler after a > inside a quoted attribute. Grav accepts and stores the page, then executes the handler in the application origin when a visitor opens it.

A Markdown audio or video embed renders its <source> element as raw HTML with the media URL concatenated unescaped. The URL fragment is reflected without any encoding, so

Type: Stored XSS (CWE-79) Auth required: Page-edit ("publisher") account, not super-admin Consequence: Arbitrary JavaScript execution in any visitor's browser, including a super-admin who views the page — a cross-trust-boundary escalation from publisher to admin-equivalent action capability.

A path traversal vulnerability in MediaUploadTrait::deleteFile() allows an authenticated user with media management permissions to delete arbitrary files on the server. The method validates only the basename portion of the filename using Utils::checkFilename(), while the directory path (which may contain ../ sequences) is preserved and passed unvalidated to unlink(). This enables directory escape from the intended media storage path.

Reported by: Nihad Huseynli (@nihaddhuseynli (https://github.com/nihaddhuseynli)) — nihadd.huseynli@gmail.com ▎ Note: I attempted to report this via security@getgrav.org first, per SECURITY.md, but the email bounced with 550 5.1.1 Address does not exist. Filing directly here instead. Path Traversal in ImageMedium::watermark() leading to arbitrary file disclosure via publicly-served images Summary The watermark media action, documented and allow-listed for use in editor-authored Markdown image syntax, passes its $image argument unsanitized into UniformResourceLocator::findResource(). That …

Grav\Common\Uri::referrer() and Grav\Common\Page\Pages::referrerRoute() both check whether an incoming request's Referer header "came from our site" using str_starts_with($referrer, $base), where $base is the site's own absolute root URL (for example https://example.com, no trailing slash). Because the comparison has no boundary character after the prefix, any Referer value that merely starts with that string is accepted, including a Referer from a completely different host such as https://example.com.attacker.tld. This is the same class …

Grav\Common\Utils::verifyNonce(), the core function Grav and its plugins use to validate CSRF nonces, compares the submitted nonce to the expected value with PHP's === operator instead of hash_equals(). === on strings short circuits at the first differing byte, so the comparison time leaks how many leading bytes of a guess are correct. This is CWE-208, Observable Timing Discrepancy. The codebase already knows to avoid this pattern. hash_equals() is used for …

The core Flex group blueprint system/blueprints/user/group.yaml (access field, lines 48-55) omits the security@: admin.super field guard that its sibling account blueprint carries (account.yaml:131/138/150, added by the CVE-2026-42613 fix). A delegated non-super operator holding admin.users.update can therefore save a group whose access map contains admin.super: true, which UserGroupObject::authorize then grants to every member of that group, a full privilege escalation to super-admin (scheduler/cron RCE, Twig eval). This is a distinct file, …

The media_directory() Twig function is allow-listed for use in sandboxed, editor-authored page content (system/config/security.yaml). Its implementation, GravExtension::mediaDirFunc(), only treats the input as unsafe when it looks like a Grav stream (user://, theme://, etc). If the input is instead a plain filesystem path, absolute or relative, the stream check is skipped entirely and the raw string is handed straight to new Media($media_dir), which lists every file in that directory whose extension …

Grav CMS's blueprint dynamic-field callable guard can be bypassed with a fully-qualified Class::method string, letting an account with only page-editing rights (admin.pages, not super-admin) plant a directive in a page's form-field frontmatter that invokes an arbitrary public static PHP method with attacker-controlled arguments. Using built-in gadget methods this yields, at minimum, arbitrary reading of any server-readable file (disclosed to anonymous visitors of the crafted page) and arbitrary creation/copying of files …

A missing validation check in Grav's Flex framework lets an account holding nothing but an ordinary object-create permission on a single Flex directory execute arbitrary shell commands on the server. Any authenticated user with create or update rights on a Flex-based directory (Flex Users, Flex Pages, Flex Objects, or any custom Flex type) can trigger it the moment a blueprint field anywhere in that directory carries a data-*@: directive, since …

Affected versions and vulnerable location Confirmed on grav core at 78ebfc1 (tag 2.0.13). Detector: system/src/Grav/Common/Security.php:290, the on_events regex, run via patternMatches() (:315-330). The on_events pattern at HEAD: #<(?:"[^"]"|'[^']'|[^>"'])?(?:[\s\x00-\x20"'/]|"[^"]"|'[^']')on\s[a-z]+\s*=#iu Sole save-time guard for non-super content: Validation::checkSafety() (system/src/Grav/Common/Data/Validation.php:160 scalars, :165 arrays), invoked per field from BlueprintSchema::validate -> Validation::checkSafety (system/src/Grav/Common/Data/BlueprintSchema.php:248). security.xss_whitelist: [admin.super] exempts only super-admins (Validation.php:148). Root cause (distinct from GHSA-269c) GHSA-269c hardened the tag-body scan to be quote-aware so a > inside …

The decompression-bomb bound added in 2.0.1 (commit 1c1003c) sums ZipArchive::statIndex($i)['size'] and rejects an archive whose declared uncompressed total exceeds system.gpm.archive.max_uncompressed_size (default 1 GiB) before extracting (ZipArchiver.php:77-86; same logic in GPM\Installer::unZip at Installer.php:228-238). statIndex()['size'] is the uncompressed size declared in the ZIP central directory, which is attacker-forgeable and is not checked against the actual inflated stream. An archive declaring 1 byte per entry passes the cap while extractTo() writes the real …

system/config/security.yaml's default twig_sandbox.config_denied_paths list (plugins, streams, security, backups, scheduler) omits the system prefix. When an operator enables the documented, non-default twig_content.config_access: true setting (intended to safely expose low-sensitivity values like site.title to editor-authored Twig content), any real secret stored under system.* , for example system.cache.redis.password , is also exposed, both via config.get(…) and via config.toArray(), to any user with page-edit permission. This is a follow-up gap in the fix for …

Affected versions and vulnerable location Confirmed on grav core at 78ebfc1 (tag 2.0.13). Sinks: system/src/Grav/Common/Data/Blueprint.php:455-458 call_user_func_array($o, $params) (bare-function dynamic-data provider). Twin: system/src/Grav/Framework/Flex/FlexDirectory.php:936-938 call_user_func_array($function, $params). Validation gate: Blueprint::isSafeDynamicCall() at Blueprint.php:514-536. Class::method branch (:514-527) uses a strict positive allowlist self::$allowedDynamicCallables. Bare-function branch (:530-534) uses only a denylist: if (is_string($function) && Utils::isDangerousFunction($function)) return false; return !self::paramsContainDangerousCallable($params);. Denylist: Utils::isDangerousFunction() (system/src/Grav/Common/Utils.php, list around :2020-2270). Root cause GHSA-7pgq/CVE-2026-64850 hardened the Class::method half of the dynamic-callable validation …

A logged-in user can run any command on the server. A settings field can fill itself by calling one of Grav's built-in routines, and a safety check is supposed to allow only harmless ones. The check only recognises a routine when its name is written as one piece of text; named as a pair of values instead, it is not examined at all and is passed as safe. Pointing such …

fulgur converts untrusted HTML/CSS into PDF, commonly on a server that processes input supplied by many tenants. In versions prior to 0.19.0, a body-direct child whose CSS-resolved height greatly exceeds the page height was sliced into one fragment per page with no upper bound. The height is taken directly from attacker-controlled HTML/CSS (height, vh units), so a few bytes such as: <div style="height:99999999px"></div> forced on the order of 125,000 page …

fulgur converts untrusted HTML/CSS into PDF, commonly on a server that processes input supplied by many tenants. In versions prior to 0.26.0, a childless box that resolves to a pathologically tall height was amplified into thousands of blank PDF pages, even when it produces no visible output. The childless-collapse defense that would normally collapse such a box was gated by a tag-only "replaced content" check, so any non-painting replaced element …

ExifReader 4.41.0 is vulnerable to denial of service through a crafted HEIC or AVIF file with a malicious iloc box. When offsetSize, lengthSize, and baseOffsetSize are set to zero in the iloc header, the extent-parsing loop allocates an unbounded number of JavaScript objects - up to itemCount × extentCount (65535 × 65535 = 4.3 billion) - without advancing the buffer offset. A 652-byte file causes 400MB of heap growth; a …

Five independent defects in djust's template auto-escaping cause attacker-controlled input to be rendered as live markup where Django escapes it. All four are present in shipped 1.1.0 and are fixed in 1.1.1. They share one shape: a filter or grant that escapes nothing itself and relies on the render-time auto-escape, which something downstream then removes. They are grouped into a single advisory because the mitigation is identical — upgrade to …

A context safety grant was inherited by a template binding that never earned it, so rebinding a name the view had marked safe left the mark attached to the new, attacker-controlled value. djust's context safety channel is keyed by name, not by value. Every bind copied the value and left the grant in place: view marks p = mark_safe('<b>trusted</b>') template {% with p=user_input %}{{ p }}{% endwith %} djust '<img …

CoreDNS parses attacker-controlled DNS section counts before validating them on DNS-over-HTTPS (DoH and DoH3), DNS-over-QUIC (DoQ), and DNS-over-gRPC listeners. An unauthenticated client can use DNS name compression to make one 65,533-byte request allocate more than 10 MiB while it is unpacked. Concurrent requests can exhaust memory and terminate CoreDNS.

CoreDNS accepted RFC 2136 UPDATE messages over DoH, DoH3, DoQ, and DNS-over-gRPC, then allowed the proxy/forward plugin to send them unchanged to an upstream DNS server. UDP, TCP, and DoT rejected the same opcode before plugin dispatch. If an update-capable upstream trusts CoreDNS's source address or authenticated connection instead of requiring end-to-end TSIG, an unauthenticated client can use CoreDNS to add, replace, or delete DNS records.

Ability to run arbitrary code on the server without authentication.

The FunctionsBuilder::cast($field, $dataType), extract($part, $expr), datePart($part, $expr), dateAdd($expr, $value, $unit) methods are vulnerable to SQL injection if user controlled data is supplied to the ($dataType / $part / $unit) parameters.

The FunctionsBuilder::cast($field, $dataType), extract($part, $expr), datePart($part, $expr), dateAdd($expr, $value, $unit) methods are vulnerable to SQL injection if user controlled data is supplied to the ($dataType / $part / $unit) parameters.

A malicious SSH server can wedge an AsyncSSH client, and an authenticated client can wedge an AsyncSSH server, by sending a channel maximum packet size of 0 in SSH_MSG_CHANNEL_OPEN_CONFIRMATION (server→client) or SSH_MSG_CHANNEL_OPEN (client→server). AsyncSSH stores the peer-supplied value verbatim with no lower-bound check; the first time channel data is written, SSHChannel._flush_send_buf enters a synchronous infinite loop that cannot be interrupted by asyncio.wait_for or any timeout. The loop body has no …

With automatic response decompression enabled (the default), the HTTP/1.1 path decompresses response bodies with no limit on the total output size. A hostile or compromised server, or an attacker who can change a response in transit, can send a small compressed body that inflates without bound in memory, exhausting the client's heap and causing an OutOfMemoryError. gzip, deflate and snappy are always available as vectors; brotli and zstd apply only …

When a request uses an HTTP proxy to reach an HTTPS origin, the client opens the tunnel with a plaintext CONNECT sent to the proxy before any TLS exists. On affected versions the origin's preemptive credentials were added to that CONNECT. A Basic realm sent Authorization: Basic base64(user:pass) to the proxy in the clear, and NTLM, SPNEGO or Kerberos realms sent their token. The proxy, and anyone who can read …

A client configured with a client-wide realm (a Realm set on the config builder rather than on an individual request) and following redirects could re-send those credentials to a redirect target on a different origin. The redirect code strips the per-exchange realm, but when the target answered 401 the credentials were re-derived from the client config, handing Basic or Digest credentials, or a Negotiate or NTLM token, to an attacker …

For SCRAM, and for Digest with mutual authentication, the client computes the server's verification value (the SCRAM ServerSignature, or the Digest rspauth) but does not act on the result. If the value is present and does not verify, the client only logs it and still delivers the response to the application as a successful, authenticated result. A server that never proved knowledge of the shared secret is accepted, so the …

HTML strings passed to Plate's core deserialization APIs were parsed in the active document. Certain HTML attributes could therefore trigger browser behavior during parsing, before the content was converted into editor nodes. Applications that deserialize HTML from untrusted or cross-user sources may be affected.

A Windows-specific command injection vulnerability exists in @cyclonedx/cyclonedx-npm when the CLI is invoked with the –workspace <value> option. User-supplied –workspace values can be passed to a shell command without proper neutralization on the Windows fallback execution path, enabling attackers to inject arbitrary OS commands. The vulnerability was fixed in version [6.0.0][v6.0.0].

The audio decode-duration guard (max_duration_s, env VLLM_MAX_AUDIO_DECODE_DURATION_S, default 600s) that protects against audio decompression-bomb DoS is wired into only the speech-to-text path (/v1/audio/transcriptions). The chat audio path (/v1/chat/completions, input_audio content parts) calls the same decoder with no limit, so an unauthenticated client can submit a few-KB compressed audio file that expands to multiple GB of float32 PCM at decode time, OOM-killing the worker. This is a distinct sibling of CVE-2026-5497 …

An unauthenticated remote attacker can leak one entry per HTTP request out of the in-memory session table of LocalSessionManager by sending a well-formed JSON-RPC POST that is not an InitializeRequest. The Streamable HTTP server's handle_post allocates the session before it validates the body, then early-returns on the validation failure without calling close_session. The LocalSessionHandle (and the tokio mpsc channel internals it holds) is never released for the remainder of the …

The rmcp library does not validate the resource parameter in OAuth Protected Resource metadata (RFC 9728), allowing a malicious MCP server to redirect OAuth flows to a legitimate authorization server and steal the resulting access tokens.

The OpenTelemetry.Resources.Host NuGet package is affected by an untrusted search path vulnerability on macOS. The host.id resource attribute detector launches the sh and ioreg executables by bare name rather than by absolute path, so both are resolved through the PATH environment variable. A local attacker who is less privileged than the host application, and who can influence PATH or write to a directory that appears in PATH ahead of the …

In OpenFGA, the ListUsers API could incorrectly return a user who should have been excluded.

SUMMARY A combination of bugs in node-opcua causes unlimited TCP socket accumulation (FIN-WAIT-2 state) during automatic reconnection, leading to memory exhaustion and eventual container/process crash (OOM kill). The issue is triggered by the default configuration (keepSessionAlive: true) when the OPC UA server has clock skew relative to the client. Affected version: Tested on 2.169.0 (latest as of April 2026). ENVIRONMENT Node.js: v24.11.0 node-opcua: 2.169.0 OS: Linux (containerized via Podman, slirp4netns …

SUMMARY A combination of bugs in node-opcua causes unlimited TCP socket accumulation (FIN-WAIT-2 state) during automatic reconnection, leading to memory exhaustion and eventual container/process crash (OOM kill). The issue is triggered by the default configuration (keepSessionAlive: true) when the OPC UA server has clock skew relative to the client. Affected version: Tested on 2.169.0 (latest as of April 2026). ENVIRONMENT Node.js: v24.11.0 node-opcua: 2.169.0 OS: Linux (containerized via Podman, slirp4netns …

SUMMARY A combination of bugs in node-opcua causes unlimited TCP socket accumulation (FIN-WAIT-2 state) during automatic reconnection, leading to memory exhaustion and eventual container/process crash (OOM kill). The issue is triggered by the default configuration (keepSessionAlive: true) when the OPC UA server has clock skew relative to the client. Affected version: Tested on 2.169.0 (latest as of April 2026). ENVIRONMENT Node.js: v24.11.0 node-opcua: 2.169.0 OS: Linux (containerized via Podman, slirp4netns …

The LMdeploy implements an rpc server (AsyncRPCServer in zmq_rpc.py) for supporting the RPC communications. In its core functionality call_and_response(), I found it will directly use the pickles.loads() to deserialize the received messages without any sanitization, hence resulting in a remote code execution vulnerability by this RPC server.

The XSS blueprint validator (Security::detectXss()) runs on the raw page content before Twig processing. An attacker can use Twig's string concatenation operator (~) to dynamically construct an event handler name at render time. The validator sees {{ "on" ~ "error" }} - a harmless Twig expression - and allows the content. After Twig processes the template, the output contains <img src=1 onerror=alert(1)> which is rendered via {{ content|raw }} and …

Grav 2.0.0-rc.9 and the current 2.0 branch still allow stored CSS injection through Markdown image media actions. The prior media hardening rejects direct ?style= payloads and unsafe attribute() fallbacks, but the adjacent resize() action still writes caller-controlled values directly into styleAttributes. A publisher who can edit page Markdown can store a crafted image URL that renders additional CSS declarations in the final <img style=…> attribute. This crosses the same lower-privileged …

An authenticated admin.super user can crash Grav or fill the disk by uploading a specially crafted ZIP archive through the Direct Install tool. The method Installer::unZip() calls ZipArchive::extractTo() without any limit on uncompressed size, entry count, or directory depth, enabling Zip Bomb (CWE-409), stack overflow (CWE-674), and disk/inode exhaustion.

djust's observability endpoints expose live view/session state and a remote method-invocation surface (eval_handler). The localhost restriction was an opt-in middleware that the documented setup omits; the views themselves enforced only DEBUG. In the misconfigured-but-documented scenario (DEBUG on, middleware not installed) a non-localhost client could read live application state and invoke handlers remotely.

When a Django Model instance is assigned to a public view attribute, djust serialized it to the client with no sensitive-field denylist — sending fields such as password (the hash), privilege flags (e.g. is_staff / is_superuser), tokens, and other PII to the browser. Because exposing model objects to templates is a normal djust pattern, this could leak credentials/PII without the developer realizing the full object crossed the wire.

The WebSocket handle_mount and ViewRuntime._build_request rebuild an HttpRequest via RequestFactory().get(…) with no HTTP_HOST, so request.get_host() defaulted to "testserver" on the live path. Host/subdomain/domain TenantResolvers then misresolved the tenant — None on the live path while the HTTP path resolved correctly. With STRICT_MODE=False the tenant-scoped managers returned unscoped rows (cross-tenant disclosure); with the default they returned an empty queryset (broken tenancy).

For views that opt into state snapshots, the snapshot state_json embedded in the client page was restored on reconnect as trusted view state with no integrity check. A client could edit the unsigned state_json in their page and return it in the reconnect mount frame to inject arbitrary view attributes — e.g. flip is_admin to True, or change account_id / balance — escalating privilege or tampering with business state held …

SSE sessions were keyed solely by a client-chosen session_id with no binding to the authenticated user — a control the WebSocket transport has but that was dropped on SSE. An attacker who learns (or a victim who leaks) a session_id could connect to the message endpoint and dispatch event handlers that execute with the victim's identity and state.

djust.tenants isolation was enforced only on the HTTP path. The current tenant was stored in threading.local() and set exclusively by the HTTP-only TenantMiddleware, so on the live (WebSocket/SSE) path get_current_tenant() was always None during mount and every event handler — and the tenant-aware QuerySet manager failed OPEN (returned the unfiltered queryset, ignoring STRICT_MODE), disclosing every tenant's rows to whoever held the socket. threading.local was additionally shared across connections on the …

djust.mixins.model_binding.ModelBindingMixin provides a default update_model event handler and is part of the LiveView base MRO, so every LiveView exposes it. It setattrs a view attribute whose name is client-supplied (field), gated only by: reject _-prefixed names; reject a 14-entry denylist of framework internals (FORBIDDEN_MODEL_FIELDS); optional allowed_model_fields which defaults to None = allow all; and hasattr existence. Result: a client can set any public, existing view attribute — not just the …

Many djust built-in component template tags (djust.components.templatetags.*) render a developer/user-supplied URL into an href / action attribute, HTML-escaping it with conditional_escape but never validating the URL scheme. HTML escaping prevents attribute breakout but does not neutralize a javascript: URI (which needs no escapable characters), so a URL value of javascript:alert(document.cookie) lands verbatim in <a href="javascript:alert(document.cookie)"> and executes in the victim's session on click. Affected sinks (executing href/action contexts; not exhaustive): …

The SSE client→server POST endpoints are @csrf_exempt and the SSE GET stream endpoint had no Origin check, so a cross-origin page could drive a victim-cookie-authenticated SSE session: force the victim's browser to GET the stream URL (which creates and mounts a LiveView as the victim) and POST to the message endpoint with credentials: include to fire state-changing event handlers as the victim. The URL session_id is client-chosen (validated only for …

djust's per-object authorization (get_object + has_object_permission, ADR-017) was enforced on the WebSocket mount and event paths but not on three other render entry points: (a) the initial HTTP GET render, (b) SPA url_change navigation, and (c) {% live_render %} embedded child views. An authenticated user could therefore view (and on some paths act on) an object they are not authorized for by loading the page directly, navigating to it via …

The djust live transport resolves the LiveView to mount from a client-supplied dotted path by calling import(module_path, …). The module is imported — running its top-level code (import side effects) — before the framework checks that the resolved object is a LiveView subclass and before any per-view authentication. The LIVEVIEW_ALLOWED_MODULES allowlist that should contain this is fail-open (if allowed_modules: — skipped when the setting is unset, the framework default) and …

The live (WebSocket) transport authorizes a mount via check_view_auth, not Django's View.dispatch() chain. As a result, standard Django authorization — LoginRequiredMixin, PermissionRequiredMixin, UserPassesTestMixin, @method_decorator(login_required, name="dispatch"), and custom dispatch() guards — and the djust admin extension's staff gate (applied only in the HTTP as_view wrapper) were enforced on the initial HTTP GET but silently bypassed over WebSocket, where all events and state flow. An anonymous or under-privileged client could open a …

The SSE transport mode (SSE=true) exposes all MCP tools without any authentication. The upload_markdown tool reads arbitrary files from the server's local filesystem via an unsanitized file_path parameter and uploads them to a GitLab project. Combined, any unauthenticated network-reachable attacker can read /proc/self/environ to steal the server's GITLAB_PERSONAL_ACCESS_TOKEN and achieve full GitLab account takeover. This is the default configuration for Docker deployments.

@nuxtjs/mdc renders untrusted markdown (including raw HTML) to a Vue component tree. Across two prior advisories it added a URL/attribute sanitizer to block dangerous links in that HTML: validateProps / validateProp and an unsafeLinkPrefix deny-list (dist/runtime/parser/utils/props.js). The sanitizer runs at parse time (dist/runtime/parser/compiler.js) and parseMarkdown enables raw HTML by default (allowDangerousHtml: true, dist/runtime/parser/options.js), so the sanitizer is the only barrier and it applies with no configuration required. Two sibling vectors …

Nezha v2.2.3 regresses the GHSA-9rc6-8cjv-rcvx host header injection fix for deployments where the new dashboard_host setting is empty. In that configuration, /api/v1/oauth2/{provider} again reflects the request Host header into the OAuth2 redirect_uri sent to the identity provider, even if install_host is configured.

The sqliteDeleteRecord function in Netmaker's database layer constructs SQL DELETE statements using direct string concatenation of user-supplied input. This allows an authenticated attacker to perform boolean-based SQL injection.

libp2p-quic can panic on an inbound QUIC handshake if a malicious peer presents a valid, short lived libp2p TLS certificate and delays the final TLS 1.3 handshake fragment until the certificate expires. This is remotely reachable by a network peer and can crash applications exposing a libp2p QUIC listener.

The static content handlers ResourceService and WebjarService URL decode each path segment and then reject only segments that are exactly "", ".", or "..". A percent-encoded separator (%2F) lets an attacker smuggle a ../ segment past that filter and escape the configured base, reading resources that should not be public. On Windows, a similar attack exists with (%5C).

The static content handlers ResourceService and WebjarService URL decode each path segment and then reject only segments that are exactly "", ".", or "..". A percent-encoded separator (%2F) lets an attacker smuggle a ../ segment past that filter and escape the configured base, reading resources that should not be public. On Windows, a similar attack exists with (%5C).

The static content handlers ResourceService and WebjarService URL decode each path segment and then reject only segments that are exactly "", ".", or "..". A percent-encoded separator (%2F) lets an attacker smuggle a ../ segment past that filter and escape the configured base, reading resources that should not be public. On Windows, a similar attack exists with (%5C).

An unauthenticated peer can make Ember's HTTP/2 read loop hold 16 MiB of a single frame in memory on a connection where Ember advertised a 16 KiB limit. The declared length is readable from the frame's first 9 bytes, but it is not compared against SETTINGS_MAX_FRAME_SIZE until the whole payload has been read into a contiguous buffer. That is 1024x amplification per connection, bounded by the protocol's 24-bit length field …

An unauthenticated peer can make Ember's HTTP/2 read loop hold 16 MiB of a single frame in memory on a connection where Ember advertised a 16 KiB limit. The declared length is readable from the frame's first 9 bytes, but it is not compared against SETTINGS_MAX_FRAME_SIZE until the whole payload has been read into a contiguous buffer. That is 1024x amplification per connection, bounded by the protocol's 24-bit length field …

An unauthenticated peer can make Ember's HTTP/2 read loop hold 16 MiB of a single frame in memory on a connection where Ember advertised a 16 KiB limit. The declared length is readable from the frame's first 9 bytes, but it is not compared against SETTINGS_MAX_FRAME_SIZE until the whole payload has been read into a contiguous buffer. That is 1024x amplification per connection, bounded by the protocol's 24-bit length field …

Ember's chunk decoder parses the size token leniently: it strips leading and trailing whitespace and accepts a leading + or - sign. RFC9112 §7.1 defines chunk-size = 1*HEXDIG. An intermediary that parses the chunk boundary differently (or rejects it) will disagree with Ember on request framing, enabling HTTP request smuggling (TE.TE).

Ember's chunk decoder parses the size token leniently: it strips leading and trailing whitespace and accepts a leading + or - sign. RFC9112 §7.1 defines chunk-size = 1*HEXDIG. An intermediary that parses the chunk boundary differently (or rejects it) will disagree with Ember on request framing, enabling HTTP request smuggling (TE.TE).

Ember's chunk decoder parses the size token leniently: it strips leading and trailing whitespace and accepts a leading + or - sign. RFC9112 §7.1 defines chunk-size = 1*HEXDIG. An intermediary that parses the chunk boundary differently (or rejects it) will disagree with Ember on request framing, enabling HTTP request smuggling (TE.TE).

The DigestAuth server middleware's stale-nonce cleanup uses an inverted comparison: it removes fresh nonces and stops at the first stale one. Because a new nonce is created for every unauthenticated challenge, an attacker can drive the nonce map to grow without bound until the JVM runs out of heap.

The DigestAuth server middleware's stale-nonce cleanup uses an inverted comparison: it removes fresh nonces and stops at the first stale one. Because a new nonce is created for every unauthenticated challenge, an attacker can drive the nonce map to grow without bound until the JVM runs out of heap.

The DigestAuth server middleware's stale-nonce cleanup uses an inverted comparison: it removes fresh nonces and stops at the first stale one. Because a new nonce is created for every unauthenticated challenge, an attacker can drive the nonce map to grow without bound until the JVM runs out of heap.

The DigestAuth replay defence stores lastNc + 1 rather than the nonce-count (nc) value it just accepted. When a legitimate client sends non-contiguous nc values (parallel or retried requests, as browsers do), the server's counter lags behind the highest nc seen, and a captured Authorization header can be replayed multiple times.

The DigestAuth replay defence stores lastNc + 1 rather than the nonce-count (nc) value it just accepted. When a legitimate client sends non-contiguous nc values (parallel or retried requests, as browsers do), the server's counter lags behind the highest nc seen, and a captured Authorization header can be replayed multiple times.

The DigestAuth replay defence stores lastNc + 1 rather than the nonce-count (nc) value it just accepted. When a legitimate client sends non-contiguous nc values (parallel or retried requests, as browsers do), the server's counter lags behind the highest nc seen, and a captured Authorization header can be replayed multiple times.

The CookieJar client middleware decides whether to attach a cookie to an outgoing request using an unanchored substring test on the host and path, instead of the domain match specified by RFC6265 5.1.3. A cookie stored for example.com is therefore sent to any host whose name merely contains example.com (e.g. evilexample.com), leaking potentially sensitive cookies to an attacker-chosen host.

The CookieJar client middleware decides whether to attach a cookie to an outgoing request using an unanchored substring test on the host and path, instead of the domain match specified by RFC6265 5.1.3. A cookie stored for example.com is therefore sent to any host whose name merely contains example.com (e.g. evilexample.com), leaking potentially sensitive cookies to an attacker-chosen host.

The CookieJar client middleware decides whether to attach a cookie to an outgoing request using an unanchored substring test on the host and path, instead of the domain match specified by RFC6265 5.1.3. A cookie stored for example.com is therefore sent to any host whose name merely contains example.com (e.g. evilexample.com), leaking potentially sensitive cookies to an attacker-chosen host.

When processing a Set-Cookie from a response, the CookieJar client middleware trusts the server-supplied Domain attribute verbatim, with no check that it domain-matches the host that sent the cookie (RFC6265 §5.3 step 6) and no public suffix check. A malicious or compromised server can therefore plant a cookie for any domain in the cookie jar, which is subsequently set on the client's next request to that victim domain, enabling session …

When processing a Set-Cookie from a response, the CookieJar client middleware trusts the server-supplied Domain attribute verbatim, with no check that it domain-matches the host that sent the cookie (RFC6265 §5.3 step 6) and no public suffix check. A malicious or compromised server can therefore plant a cookie for any domain in the cookie jar, which is subsequently set on the client's next request to that victim domain, enabling session …

When processing a Set-Cookie from a response, the CookieJar client middleware trusts the server-supplied Domain attribute verbatim, with no check that it domain-matches the host that sent the cookie (RFC6265 §5.3 step 6) and no public suffix check. A malicious or compromised server can therefore plant a cookie for any domain in the cookie jar, which is subsequently set on the client's next request to that victim domain, enabling session …

Ember's HTTP/1.1 header parser matches the Transfer-Encoding header value with a case-sensitive substring test (hValue.contains("chunked")). RFC 9112 §7 requires transfer-coding names to be compared case-insensitively. A request carrying Transfer-Encoding: Chunked (capital C) is therefore not recognised as chunked, and Ember falls back to framing by Content-Length (or zero if absent) while a compliant intermediary frames the same bytes by chunked encoding. The two parsers then disagree on where the request …

Ember's HTTP/1.1 header parser matches the Transfer-Encoding header value with a case-sensitive substring test (hValue.contains("chunked")). RFC 9112 §7 requires transfer-coding names to be compared case-insensitively. A request carrying Transfer-Encoding: Chunked (capital C) is therefore not recognised as chunked, and Ember falls back to framing by Content-Length (or zero if absent) while a compliant intermediary frames the same bytes by chunked encoding. The two parsers then disagree on where the request …

Ember's HTTP/1.1 header parser matches the Transfer-Encoding header value with a case-sensitive substring test (hValue.contains("chunked")). RFC 9112 §7 requires transfer-coding names to be compared case-insensitively. A request carrying Transfer-Encoding: Chunked (capital C) is therefore not recognised as chunked, and Ember falls back to framing by Content-Length (or zero if absent) while a compliant intermediary frames the same bytes by chunked encoding. The two parsers then disagree on where the request …

Ember's HTTP/2 stack replenishes the inbound flow-control window based on bytes received off the wire, not bytes consumed by the application. Received DATA is buffered in an unbounded per-stream channel. Flow control therefore provides no backpressure: a peer can stream a large or unbounded body faster than the application drains it and the connection retains every payload in heap. This is the read-path mirror of the outbound queue issue. This …

Ember's HTTP/2 stack replenishes the inbound flow-control window based on bytes received off the wire, not bytes consumed by the application. Received DATA is buffered in an unbounded per-stream channel. Flow control therefore provides no backpressure: a peer can stream a large or unbounded body faster than the application drains it and the connection retains every payload in heap. This is the read-path mirror of the outbound queue issue. This …

Ember's HTTP/2 stack replenishes the inbound flow-control window based on bytes received off the wire, not bytes consumed by the application. Received DATA is buffered in an unbounded per-stream channel. Flow control therefore provides no backpressure: a peer can stream a large or unbounded body faster than the application drains it and the connection retains every payload in heap. This is the read-path mirror of the outbound queue issue. This …

When Ember receives an HTTP/2 HEADERS or PUSH_PROMISE frame without the END_HEADERS flag, it buffers the header block fragment and waits for subsequent CONTINUATION frames. These accumulate unbounded until the connection closes.

When Ember receives an HTTP/2 HEADERS or PUSH_PROMISE frame without the END_HEADERS flag, it buffers the header block fragment and waits for subsequent CONTINUATION frames. These accumulate unbounded until the connection closes.

When Ember receives an HTTP/2 HEADERS or PUSH_PROMISE frame without the END_HEADERS flag, it buffers the header block fragment and waits for subsequent CONTINUATION frames. These accumulate unbounded until the connection closes.

Ember's HTTP/2 connection serializes all outgoing frames through a single unbounded queue drained by one writer fiber (writeLoop). When the write side stalls, any frames the connection keeps producing accumulate in that queue without limit. The peer can drive this cheaply because the connection emits a control frame in response to inbound frames it does not flow-control: one PING ACK per PING, one SETTINGS ACK per SETTINGS, and a WINDOW_UPDATE …

Ember's HTTP/2 connection serializes all outgoing frames through a single unbounded queue drained by one writer fiber (writeLoop). When the write side stalls, any frames the connection keeps producing accumulate in that queue without limit. The peer can drive this cheaply because the connection emits a control frame in response to inbound frames it does not flow-control: one PING ACK per PING, one SETTINGS ACK per SETTINGS, and a WINDOW_UPDATE …

Ember's HTTP/2 connection serializes all outgoing frames through a single unbounded queue drained by one writer fiber (writeLoop). When the write side stalls, any frames the connection keeps producing accumulate in that queue without limit. The peer can drive this cheaply because the connection emits a control frame in response to inbound frames it does not flow-control: one PING ACK per PING, one SETTINGS ACK per SETTINGS, and a WINDOW_UPDATE …

An ember server with HTTP/2 enabled (.withHttp2) does not enforce SETTINGS_MAX_CONCURRENT_STREAMS on streams opened by the peer. A single unauthenticated connection can open an unbounded number of concurrent streams, each of which allocates per-stream server state that is never released, exhausting the heap.

An ember server with HTTP/2 enabled (.withHttp2) does not enforce SETTINGS_MAX_CONCURRENT_STREAMS on streams opened by the peer. A single unauthenticated connection can open an unbounded number of concurrent streams, each of which allocates per-stream server state that is never released, exhausting the heap.

An ember server with HTTP/2 enabled (.withHttp2) does not enforce SETTINGS_MAX_CONCURRENT_STREAMS on streams opened by the peer. A single unauthenticated connection can open an unbounded number of concurrent streams, each of which allocates per-stream server state that is never released, exhausting the heap.

Ember's HTTP/1.1 request parser does not reject a message that carries both a Transfer-Encoding and a Content-Length header. RFC 9112 §6.1 requires a server to treat such a message as a framing error and close the connection. An intermediary that follows the RFC's CL-strip-and-forward path (or that prioritises Content-Length) will frame the body differently from Ember, enabling HTTP request smuggling (CL.TE).

Ember's HTTP/1.1 request parser does not reject a message that carries both a Transfer-Encoding and a Content-Length header. RFC 9112 §6.1 requires a server to treat such a message as a framing error and close the connection. An intermediary that follows the RFC's CL-strip-and-forward path (or that prioritises Content-Length) will frame the body differently from Ember, enabling HTTP request smuggling (CL.TE).

Ember's HTTP/1.1 request parser does not reject a message that carries both a Transfer-Encoding and a Content-Length header. RFC 9112 §6.1 requires a server to treat such a message as a framing error and close the connection. An intermediary that follows the RFC's CL-strip-and-forward path (or that prioritises Content-Length) will frame the body differently from Ember, enabling HTTP request smuggling (CL.TE).

The http_poll C2 transport accepts attacker-controlled HTTP polling sessions before CBOR MsgAuth authentication is completed. A remote unauthenticated attacker can create arbitrary polling sessions and send request bodies that are forwarded into the C2 dispatch path. This can consume server resources and trigger pre-auth C2 processing.

@zereight/mcp-gitlab exposes its Streamable HTTP MCP endpoint without an effective Host or Origin allowlist. A malicious web page can use DNS rebinding to route browser requests to a victim's local MCP listener while preserving an attacker-controlled Host and Origin. The server accepts those headers and reaches the MCP initialization path instead of rejecting the request at the HTTP boundary. This is CWE-350, Reliance on Reverse DNS Resolution for a Security-Critical …

When the environment variable ENABLE_DYNAMIC_API_URL=true is set, the server reads the X-GitLab-API-URL HTTP request header and uses it as the base URL for all outbound GitLab API calls made within that request. The server validates that the value is a well-formed URL (new URL(dynamicApiUrl)) but applies no allowlist or hostname restriction. The server then attaches the victim's Private-Token to every outbound fetch that uses the redirected URL. Any caller who …

@zereight/mcp-gitlab exposes GitLab to an LLM agent while relying on read-only mode, a project allow-list, and transport auth as its safety controls. Five defects defeat those controls. Under the MCP threat model, tool-call arguments/content can be shaped by untrusted input (prompt injection) or a malicious client. Reviewed commit: 60adcc0de5b0e96c4c2029f7a25d2775946421d8 (package version 2.1.28). Source review only; PoCs are local/offline. F1 (HIGH) execute_graphql defeats BOTH read-only mode and GITLAB_ALLOWED_PROJECT_IDS. F2 (HIGH, deployment-conditional) …

A vulnerability in ZITADEL’s OAuth2 Token Exchange endpoint allows an authenticated user or client to exchange a low-privilege access token for a token with elevated permissions at a completely different application. This bypasses the intended authorization and separation policies configured within ZITADEL.

A bug in how ZITADEL updates permissions when multiple project roles are deleted at the same time can cause some user permissions to be missed. This issue specifically affects User Grants on Granted Projects (projects shared between different organizations), potentially allowing users to keep access rights that were supposed to be completely removed.

A vulnerability was identified in the Twig sandbox security policy that allowed a chained bypass when cms.safe_mode is enabled. The Laravel session store was exposed to Twig with unrestricted method access, and raw SQL methods reachable through Eloquent's __call forwarding were not blocked across the full builder chain. When combined, a backend user with CMS markup editing access could read arbitrary database values via raw SQL expressions and write to …

The backend SessionMaker trait stored widget session state as base64(serialize(…)) and consumed it with unserialize() without an allowed_classes restriction. Any code path that could write to a widget.* session key with attacker-controlled bytes could trigger PHP object injection the next time the widget read its session state, allowing instantiation of arbitrary classes and reachable PHP gadget chains. This issue only affects installations running with cms.safe_mode enabled. Safe Mode is a …

The image resizer classified external sources by testing whether the source string began with the substring http, and the string-source branch in ResizeImageItem::fromObject() accepted any value containing :// as a URL. As a result, non-http(s) PHP stream wrappers such as phar://, file:// and ftp:// could be stored in the resizer cache and later passed to the underlying image library. On a phar:// path this can lead to metadata deserialization during …

The dashboard reads its authentication credentials from $ESPHOME_USERNAME and $ESPHOME_PASSWORD. Earlier versions, and the legacy esphome dashboard, read the bare $USERNAME and $PASSWORD instead. When the env vars were renamed the bare names were dropped with no fallback, so an operator who had protected their dashboard with USERNAME / PASSWORD (as the older getting started guide documented) loses authentication on upgrade and the dashboard starts open to anyone who can …

A token lifecycle validation vulnerability was discovered in ZITADEL's external JWT Identity Provider (IdP) implementation Specifically, within the validation pipeline, if an incoming JWT omits the exp claim entirely, the expiration block is silently skipped rather than rejected. The token is treated as valid forever. Per the OIDC Core 1.0 specification, identity token validation pipelines must strictly handle and enforce session expiration. ZITADEL's silent acceptance of tokens missing these temporal …

Store/LegacyStore key internal lookup tables by the type, id, and relationship names from a JSON:API document. Because these were plain objects, a document with type: "proto" writes onto Object.prototype, polluting every object in the process.

Four Livewire components in the Settings area expose destructive Filament actions (delete / edit) that perform no server-side authorization. Any authenticated user who can reach the Settings pages — i.e. holding only the coarse access_setting permission, without being an admin and without any delete_/edit_ permission — can delete tax zones, tax rates, shipping zones, and carrier (shipping-rate) options by invoking the component action directly over the Livewire endpoint. These records …

A lack of authorization control was discovered in the stockAction() method in packages/admin/src/Livewire/Components/Products/VariantStock.php. The component exposes a public $variant property without the #[Locked] attribute, so the variant ID is client-mutable via the Livewire wire payload. The stockAction() returns an Action with no ->authorize(…) chain, meaning any authenticated admin-panel session, including browse-only staff who hold zero edit permissions, can call this action to adjust inventory levels for any product variant. The …

Three Livewire admin components in shopper/framework (latest master at commit fcd0c59, released as v2.8.0) gate state-mutating actions on the read-only view_users permission. This is the same class as the issue Shopper fixed in v2.8.0 / PR #511 / GHSA-f946-9qp6-vgch — the PR moved most write actions from view_users to access_setting, but three were missed (one of them is a brand-new file added by the security commit itself). A staff user …

The Shopper Framework discount management functionality accepts negative discount values without server-side validation. It was confirmed that negative fixed-amount discounts can be created through the administrative interface, persisted to the database, and subsequently processed by the cart/order calculation pipeline. The application appears to assume that discount values are always positive but does not enforce this assumption during creation, storage, or calculation. As a result, malformed discount records can influence financial …

A lack of authorization control was discovered on both the per-record delete action and the bulk delete action inside packages/admin/src/Livewire/Components/Collection/CollectionProducts.php. Neither the Action::make('delete') at line 73 nor the DeleteBulkAction::make() at line 91 carries an ->authorize(…) chain. The component also exposes public Collection $collection without #[Locked], so the collection ID is mutable in the Livewire wire payload. Any authenticated admin-panel session, including staff who hold only browse_collections, can detach individual products …

A lack of authorization control on the store() method was found in packages/admin/src/Livewire/Components/Products/Form/Media.php. The security fix released for GHSA-h4mp-g9c6-xwph added #[Locked] to the $product property in this file but did not add an authorize() call to store(). The commit message for that fix (fcd0c59) explicitly names the five repaired sub-form components: Edit, Inventory, Seo, Shipping, Files. Media is absent from that list and absent from the published advisory. As a …

Five Filament groupedBulkActions blocks across the Shopper admin Livewire pages omit the ->authorize(…) permission gate, while their per-record sibling actions (and other Shopper Index pages such as Pages/Settings/Currencies.php, Pages/Reviews/Index.php, Pages/Collection/Index.php, and Pages/Discount/Index.php) correctly chain ->authorize(…). Each affected page's mount() only requires the read-only browse_* permission, so a low-privilege staff user holding only the read permission can drive the bulk endpoint via the standard Livewire callTableBulkAction flow and execute state-mutating operations …

Prowler's SAML authentication flow trusted the email domain asserted in a SAMLResponse when deciding which tenant should receive the final token. A malicious tenant with its own SAML configuration and a self-controlled IdP could complete a valid SAML flow for its own configured domain, while asserting an email address from another configured domain. In the vulnerable flow, the ACS finish logic later derived the tenant from the asserted email domain …

In SSE/HTTP transport mode, mysql_mcp_server constructs SseServerTransport without passing security_settings. As a result, the MCP Python SDK's DNS-rebinding protection (Origin/Host header validation) is disabled; the Starlette application has no CORS or TrustedHost middleware; and the service binds to 0.0.0.0 by default with no authentication on any route. Trigger condition: MCP_TRANSPORT=sse. The default stdio mode is not affected.

The published fix for GHSA-v6ph-xcq9-qxxj / CVE-2026-39885 added a direct hostname denylist for OpenAPI external $ref dereferencing, but the latest patched dependency mcp-from-openapi 2.3.0 still makes backend-origin requests to loopback when the target is reached through hostname resolution, redirects, or IPv4-mapped IPv6 syntax. FrontMCP latest release v1.2.1 and current main still call OpenAPIToolGenerator.fromURL() and OpenAPIToolGenerator.fromJSON() from mcp-from-openapi 2.3.0 when loading OpenAPI adapters. An attacker who can cause a hosted or …

The published fix for GHSA-v6ph-xcq9-qxxj / CVE-2026-39885 added a direct hostname denylist for OpenAPI external $ref dereferencing, but the latest patched dependency mcp-from-openapi 2.3.0 still makes backend-origin requests to loopback when the target is reached through hostname resolution, redirects, or IPv4-mapped IPv6 syntax. FrontMCP latest release v1.2.1 and current main still call OpenAPIToolGenerator.fromURL() and OpenAPIToolGenerator.fromJSON() from mcp-from-openapi 2.3.0 when loading OpenAPI adapters. An attacker who can cause a hosted or …

The published fix for GHSA-v6ph-xcq9-qxxj / CVE-2026-39885 added a direct hostname denylist for OpenAPI external $ref dereferencing, but the latest patched dependency mcp-from-openapi 2.3.0 still makes backend-origin requests to loopback when the target is reached through hostname resolution, redirects, or IPv4-mapped IPv6 syntax. FrontMCP latest release v1.2.1 and current main still call OpenAPIToolGenerator.fromURL() and OpenAPIToolGenerator.fromJSON() from mcp-from-openapi 2.3.0 when loading OpenAPI adapters. An attacker who can cause a hosted or …

Threat model: An on-path attacker on the corporate network — ARP spoofing on the LAN, internal DNS poisoning, malicious internal DNS overriding github.com or the configured internal git hostname, BGP hijack, sidecar/CNI compromise in Kubernetes, or any process able to answer TCP on the resolved IP. No Central Dogma account required; only network position. Direction LOCAL_TO_REMOTE: the attacker impersonating the remote git server receives the entire mirrored repository contents over …

Threat model: any unauthenticated network client that can reach the Central Dogma login page. No prior account, no MITM position required — the attack is performed during a normal login request. Authentication confusion — In AD environments that select this realm (the canonical LINE corporate scenario), typing username * causes the realm to look up the first directory entry (typically Administrator, admin, or a service account in alphabetical AD ordering) …

Central Dogma: Hard-coded ZooKeeper replication secret 'ch4n63m3' with silent fallback enables cluster takeover

Mockoon's admin API (commons-server/src/libs/server/admin-api.ts) is mounted on the same Express listener as the user-defined mock routes, enabled by default in every shipped runtime (commons-server, CLI, serverless), serves Access-Control-Allow-Origin: * on every endpoint with all HTTP methods allowed including PUT/POST/PATCH/DELETE/PURGE and Content-Type in Access-Control-Allow-Headers, and has zero authentication of any kind (no token, no shared secret, no MOCKOON_ADMIN_TOKEN env var — searched the repo, returns zero hits). Any unauthenticated caller who …

Mockoon's admin API (commons-server/src/libs/server/admin-api.ts) is mounted on the same Express listener as the user-defined mock routes, enabled by default in every shipped runtime (commons-server, CLI, serverless), serves Access-Control-Allow-Origin: * on every endpoint with all HTTP methods allowed including PUT/POST/PATCH/DELETE/PURGE and Content-Type in Access-Control-Allow-Headers, and has zero authentication of any kind (no token, no shared secret, no MOCKOON_ADMIN_TOKEN env var — searched the repo, returns zero hits). Any unauthenticated caller who …

A FILE response whose filePath embeds request data (e.g. "/srv/public/{{queryParam 'name'}}", the documented way to let the client pick a file) is confined by getSafeFilePath with resolvedPath.startsWith(staticBaseDir). That prefix test has no path-separator boundary, so a ../-escaped path whose absolute form string-prefixes the base directory passes. An unauthenticated client reads files from sibling paths outside the served directory.

A FILE response whose filePath embeds request data (e.g. "/srv/public/{{queryParam 'name'}}", the documented way to let the client pick a file) is confined by getSafeFilePath with resolvedPath.startsWith(staticBaseDir). That prefix test has no path-separator boundary, so a ../-escaped path whose absolute form string-prefixes the base directory passes. An unauthenticated client reads files from sibling paths outside the served directory.

A flaw in the external identity provider handler allows unauthorized account linking to occur under specific administrative configurations. When auto-linking by email is enabled, ZITADEL checks that the local user's email is verified, but does not explicitly cross-check whether the incoming external identity provider (IdP) also verified that email address upstream.

Traefik accepts an HTTP/1.x request whose request-target is in rootless / opaque form (for example GET http:http://internal-vhost/admin HTTP/1.1). Go parses this into URL.Opaque with an empty URL.Path, so Traefik evaluates all routing, path-sanitization, middleware and access-log decisions against a path that normalizes to /, while the proxy forwards the attacker's original target byte-for-byte to the backend. Router path/prefix guards, forwardAuth path-scoped policies and the encodedCharacters hardening never see the real …

Traefik accepts an HTTP/1.x request whose request-target is in rootless / opaque form (for example GET http:http://internal-vhost/admin HTTP/1.1). Go parses this into URL.Opaque with an empty URL.Path, so Traefik evaluates all routing, path-sanitization, middleware and access-log decisions against a path that normalizes to /, while the proxy forwards the attacker's original target byte-for-byte to the backend. Router path/prefix guards, forwardAuth path-scoped policies and the encodedCharacters hardening never see the real …

There is a medium severity vulnerability in Traefik's HTTP/3 entry points: the respondingTimeouts settings were not applied to the HTTP/3 request path. readTimeout in particular is on by default at 60s and is documented as bounding the time to read the entire request including its body, but it is enforced as a deadline on the TCP connection, which cannot reach a QUIC stream, and Traefik's HTTP/3 server was constructed with …

There is a medium severity vulnerability in Traefik's HTTP/3 entry points: the respondingTimeouts settings were not applied to the HTTP/3 request path. readTimeout in particular is on by default at 60s and is documented as bounding the time to read the entire request including its body, but it is enforced as a deadline on the TCP connection, which cannot reach a QUIC stream, and Traefik's HTTP/3 server was constructed with …

There is a high-severity request-smuggling vulnerability in Traefik's handling of the HTTP/1.1 Upgrade mechanism. Since Traefik moved to unencrypted HTTP/2 with prior knowledge (Go 1.24), a client-initiated Upgrade: h2c request header and its connection-specific HTTP2-Settings header were forwarded to the backend. A backend that honours the h2c upgrade and answers 101 Switching Protocols puts Traefik into a raw byte tunnel that bypasses the router and the entire middleware chain (authentication, …

There is a high-severity request-smuggling vulnerability in Traefik's handling of the HTTP/1.1 Upgrade mechanism. Since Traefik moved to unencrypted HTTP/2 with prior knowledge (Go 1.24), a client-initiated Upgrade: h2c request header and its connection-specific HTTP2-Settings header were forwarded to the backend. A backend that honours the h2c upgrade and answers 101 Switching Protocols puts Traefik into a raw byte tunnel that bypasses the router and the entire middleware chain (authentication, …

There is a medium severity vulnerability in Traefik's handling of request headers whose name aliases another header name. Go canonicalizes header names on dashes only, so X-Auth-User, X_Auth_User and X.Auth.User are three distinct headers to Traefik, while backends that derive variable names from header names (CGI, WSGI, PHP, NGINX, and others) collapse all of them into the same variable. A client can therefore smuggle an alias of a header that …

There is a medium severity vulnerability in Traefik's handling of request headers whose name aliases another header name. Go canonicalizes header names on dashes only, so X-Auth-User, X_Auth_User and X.Auth.User are three distinct headers to Traefik, while backends that derive variable names from header names (CGI, WSGI, PHP, NGINX, and others) collapse all of them into the same variable. A client can therefore smuggle an alias of a header that …

Traefik's HTTP/3 request path did not initialize the connection-scoped backend transport holder that isolates connection-bound NTLM and Negotiate (Kerberos) authentication on the HTTP/1.1 and HTTP/2 paths. The HTTP/3 entrypoint reuses the HTTPS handler chain and reaches the same backend round-tripper, but its ConnContext never called service.AddTransportOnContext, so kerberosRoundTripper fell back to the shared backend transport instead of a per-frontend-connection pool. On a route served over HTTP/3 to a backend that …

Traefik's HTTP/3 request path did not initialize the connection-scoped backend transport holder that isolates connection-bound NTLM and Negotiate (Kerberos) authentication on the HTTP/1.1 and HTTP/2 paths. The HTTP/3 entrypoint reuses the HTTPS handler chain and reaches the same backend round-tripper, but its ConnContext never called service.AddTransportOnContext, so kerberosRoundTripper fell back to the shared backend transport instead of a per-frontend-connection pool. On a route served over HTTP/3 to a backend that …

Traefik's entrypoint defenses against spoofed trusted header names — aliasHeadersStrategy / underscoreHeadersStrategy in delete or reject mode, and the default forwardedHeaders stripping of client-supplied X-Forwarded-* — scan req.Header only and never req.Trailer. An unauthenticated client can therefore smuggle a sanitized name (an aliasing spelling such as X_Auth_User, or a trusted name such as X-Forwarded-Prefix) as an HTTP/1.1 chunked trailer or an HTTP/2 trailer: reject does not return its documented 400, …

Multiple backends, when given a specially crafted object to copy, can escape the backend confinement. | Backend | Keep/Close | Per-backend severity | |—|—|—| | sftp | Medium | Real filesystem escape, fires under default encoding. | | smb | Low-Medium | Escapes to a different SMB share the credential can reach. | | ftp | Low | Real, leading-.. overshoot PoC is partly neutralized by encoding; escape bounded to …

In streamed multipart mode, serve s3 passes the request's declared part length to multipart.NewRW().Reserve(contentLength) before reading any part data. Reserve immediately obtains enough 1 MiB pool pages for the entire declared length. The request handler therefore allocates attacker-selected memory based only on Content-Length or X-Amz-Decoded-Content-Length; the client does not need to transmit the corresponding body. –multipart-streaming-buffer-limit does not stop the allocation for the current expected part or for one oversized …

serve/start accepts protocol options in a per-server proxyOpt object. The FTP and S3 RC adapters parse that object and pass it to their server constructors, but the constructors decide whether proxy authentication is enabled by checking the process-global proxy.Opt.AuthProxy instead of the supplied proxyOpt.AuthProxy. When the process-global option is empty—the normal case when only the RC request configures the server—the supplied authentication proxy is silently ignored. FTP falls back to …

Exfiltration of API keys / bearer tokens / session cookies configured for one host, to any host the (trusted-at-configuration-time) remote later redirects to. All operations on the http backend (list, stat, download, mount, serve) are affected. No special rclone privileges or unusual user interaction are needed beyond a normal sync/list/copy once the redirect exists.

The FTP auth-proxy driver stores one obscured password per username in a server-wide map. It does not bind the credential or returned VFS to the authenticated FTP session. If two accepted credentials use the same username but resolve to different proxy backends, the later login overwrites the map entry. Subsequent operations on the first, still-authenticated session are re-authorized with the later session's password and execute against the later session's backend. …

With -l/–links, rclone's local backend recreates a source .rclonelink object as a real symlink at the destination verbatim (preserved by design for faithful backups). Directory-metadata application, however, does not go through the os.Root sandbox and does not use NOFOLLOW syscalls. A local Directory always has translatedLink=false, so when the destination path already exists as a planted symlink, rclone applies chmod/chown/chtimes through that symlink to a target outside the destination tree. …

rclone serve s3's handler chain, when –auth-proxy is configured, is (outermost first): authPairMiddleware -> proxyAuthMiddleware -> gofakes3's own SigV4-verifying handler. authPairMiddleware parses the accessKeyID straight out of the incoming request's own Authorization header (entirely client-controlled) and registers {accessKey: ws.s3Secret} into gofakes3's shared credential store via AddAuthKeys, for EVERY access key any client presents - not just ones previously known to the server. ws.s3Secret defaults to "" whenever –auth-key is not …

When backend/local is used with –links/-l (or the links=true config option), each symlink is exposed as an rclone object whose content is the target path string, suffixed .rclonelink. Object.Open() decodes an incoming fs.RangeOption via Decode(o.Size()), then for a translated-symlink object passes the decoded offset straight into openTranslatedLink, which indexes the target string directly: linkdst[offset:]. RangeOption.Decode's Start >= 0 branch (an ordinary Range: bytes=X- request) sets offset = o.Start with no …

backend/archive mounts a zip file as a browsable, syncable rclone Fs (e.g. rclone lsf :zip:downloaded.zip or rclone copy :zip:downloaded.zip dest:). Go's archive/zip package does not sanitize file.Name - it is taken verbatim from the untrusted zip's central directory. readZip() in backend/archive/zip/zip.go applies path.Clean to the entry name, but this alone cannot fully neutralize a name with more .. components than real segments preceding them (e.g. "../../etc/cron.d/evil" stays exactly as-is after …

Pimcore: SQL Injection in Custom Reports via Malicious Report Configuration

Open WebUI's OAuth token exchange endpoint issues a session for a provider access token without running the OAuth role management that the normal OAuth login callback runs. A user whose provider roles the login callback would refuse, or would demote, could still obtain a working session at their existing role through this endpoint.

Open WebUI's OAuth token exchange endpoint issues a session for a provider access token without applying the email domain allowlist that the normal OAuth login callback enforces. An account whose email domain the login callback would refuse could still obtain a working session through this endpoint.

The OIDC back-channel logout endpoint is unauthenticated by design, because the identity provider calls it without a browser session. Before checking whether the submitted logout token was genuine, the handler fetched the provider's discovery document and its signing keys over the network, and repeated both fetches on every request because nothing was cached. The signing-key fetch also ran as a blocking call inside the async event loop. A small number …

With the Playwright web loader enabled, Open WebUI checks the address behind a user-submitted URL before allowing the request, then handed the request to the browser to perform. The browser resolved the hostname a second time, on its own, and that answer was never checked. An attacker who controls the authoritative DNS for a hostname they submit can answer the first lookup with a public address and the second with …

On SQLite deployments, the lookup that maps an external identity to a local account does a substring match instead of an exact match. A subject value containing SQL wildcard characters therefore matches accounts the value was never issued for, and the sign-in binds to whichever account the database returns first, which can be an administrator. The same defect affects SCIM external-ID resolution. PostgreSQL deployments are not affected, because they take …

Any authenticated user with access to a shared terminal server could get script of their choosing to run in the Open WebUI origin itself. The in-app port preview rendered the content of a previewed port in an iframe whose sandbox always granted allow-same-origin alongside allow-scripts, and that content is served from a path on the application's own origin, so the sandbox provided no isolation at all. Script served on a …

External knowledge connections are created and owned by administrators, and are shared by every external knowledge base bound to them. Deleting an external knowledge base also removed that connection from the instance configuration, with no check on the caller's role and no check for other knowledge bases still using it. Any authenticated user holding a write grant on a single external knowledge base could therefore wipe a shared connection that …

The built-in knowledge search tool returns knowledge bases the calling user has no access to. The tool works out which knowledge bases the caller may read and hands that set to the vector store as a search filter, and that filter is the only access control on the path. Most of the shipped vector backends accept the filter argument on their search method and never apply it, so the search …

Any member of a channel who can post to it could also replace the text of a message written by a different member. The channel branch of the chat completions endpoint checked that the caller may write to the channel, and that the targeted message belongs to that channel, but never checked that the caller actually wrote the message being edited. The message kept its original author, so the replacement …

Calendar events carry a free-form meta object that is stored exactly as submitted, with no validation of the values inside it. The scheduler reads the per-event alert offset out of that object in a single pass that covers every user's upcoming events, and compares it numerically without checking that it is a number. Any verified user could store a text value there, which made the comparison raise and abort the …

Any authenticated user can move one of their own folders under itself, leaving a loop in their folder tree. The re-parent endpoint performed no check that the new parent was not the folder itself or one of its own subfolders, and the folder tree walks did not track which folders they had already visited. A single request against a folder in a loop therefore never finishes.

Open WebUI fetches user-supplied URLs on the server for RAG URL ingestion and web search, and screens the resolved addresses so internal destinations cannot be reached. That screen decided whether a destination was external by asking Python's standard library whether the address is globally routable. Several addresses reserved for internal use answer yes to that question, including 168.63.129.16, the Azure platform channel every Azure virtual machine can reach. Any authenticated …

The chat-completions endpoint reads a folder id out of the request body and saves the newly created chat into that folder without checking that the caller is allowed to write there. Any authenticated user who knows a folder's id can put a chat of their own into another user's folder, including a shared folder where they hold read-only access and a folder they have no access to at all. The …

A user who is demoted from admin by an identity provider keeps admin-level read and write access to every user's notes, over any Socket.IO connection that was already open when the demotion happened. Open WebUI caches the user's role on the socket at connection time, and the two SSO role-sync paths, the reverse-proxy trusted role header and OAuth role mapping, changed the role in the database without tearing that cached …

When more than one external tool server is reachable in the same request, a tool call to a server configured for bearer authentication can arrive carrying the calling user's Open WebUI session cookies alongside that server's own key. The cookie jar is built per connection, but the callable that performs the request reads it late instead of per connection, so every tool callable built in the same pass sends the …

The OAuth Dynamic Client Registration endpoints validated field sizes only for redirect_uris, leaving client_name and grant_types bounded by presence checks alone. An unauthenticated remote caller could submit arbitrarily large values in either field and have them persisted to the database. Repeated registration therefore grew persistent storage without limit and without any account on the instance. The patch caps client_name at its column length and limits grant_types to the grant types …

The Git node's clone operation matched the destination path against the default N8N_BLOCK_FILE_PATTERNS expression, which was written so that a crafted path caused catastrophic backtracking. Evaluation runs synchronously in the main n8n process, so an authenticated user could freeze the instance for everyone with a single workflow execution and no unusual configuration. The patch rewrites the pattern so that matching is linear in the length of the path.

The Instance AI workflow summary built its result with get-or-create-then-nested-write idioms keyed by node names and connection keys taken from the stored workflow. Those are arbitrary strings: the restricted-name guard n8n shows in the editor is client-side and is bypassed by posting the workflow to the REST API directly. So a reserved key such as proto resolved up the prototype chain instead of creating an own key, and the write …

The OAuth token endpoint bound an authorization code's first access token to the consented resource, but not its refresh token. Refreshing only checked that the requested resource was registered, not that it matched the original grant. An OAuth client approved for one workflow could refresh with a different workflow's URL and get a valid, unapproved token for it. The patch binds refresh tokens to the granted resource and rejects mismatches. …

The Elasticsearch and ElasticSecurity nodes built REST endpoints by interpolating user-provided identifiers straight into the request path. A value containing path separators or dot segments changed which endpoint the request actually reached, so an operation intended for one document could hit another index or a cluster administration endpoint instead, under the stored Elasticsearch credential. The patch encodes each identifier as a single URL path segment and rejects values that normalise …

A log streaming event destination may reference a generic HTTP credential, and the destination resolved and decrypted whichever credential it named without checking that the caller had access to it. A user holding a custom global role scoped to Log Streaming operations could therefore name a credential belonging to another project and have its decrypted secret sent to an endpoint they controlled. The patch applies the standard credential access check …

Instance AI credential setup accepted a credential test/verification URL without checking it matched the workflow node's origin. Exfiltration required the user to actively inject an attacker-controlled URL into the setup flow. The patch derives the credential destination from the node's own URL and confines authenticated requests, redirects and probes to that origin.

The GitHub Trigger generated a signing secret when it registered a webhook and verified X-Hub-Signature-256 on each delivery. When GitHub answered the create request with a 422 because a webhook for the URL already existed, the node adopted the remote hook but kept only its id and events, discarding the secret it had generated. Deliveries to that workflow's webhook URL were then accepted from anyone. The patch re-fetches the existing …

The Git node validated the repository that a fetch or pull targeted, but setUpstream wrote a branch.<name>.remote value into the repository's own configuration without validating it. A subsequent fetch or pull resolved the remote from that configuration rather than from the checked parameter, so an authenticated user with workflow-edit permission could point it at any local repository the n8n process could read and receive its contents. The patch validates the …

Two stages of expression code generation built source text by calling the global JSON.stringify at generation time: the compiler when printing synthetic string literals, and the isolate bridge when interpolating a timezone value into its per-evaluation wrapper. An expression that replaced that global therefore changed the code that was subsequently generated and executed, turning literal data into executable source. The patch renders both code-generation stages through a reference captured at …

The expression compiler's sanitizer resolved through a dynamically-scoped this, so a class field named __sanitize rebound it and reached the Function constructor. On the backend, any expression author could run code in the n8n process; in the editor preview, a member's expression could run as JavaScript in the session of whoever opened the workflow. The patch rejects reserved class-member names.

The OpenAI Chat Model node checked a custom base URL against the credential's allowed-domains configuration before sending a request, but the model-search dropdown did not. A request setting options.baseURL on that path reached an arbitrary host with the credential attached, so the domain restriction an administrator configured did not hold for every route the node offered. The patch routes every OpenAI call site, including model listing, through one shared allowed-domains …

The public OIDC login and callback endpoints ran the full flow whether or not OIDC was the instance's active, enabled authentication method, so turning OIDC off in Settings did not stop it issuing sessions. An administrator who disabled the provider still had a working login route. The patch requires OIDC to be the enabled, active authentication method before either endpoint starts the flow or issues a session. The issue affects …

The endpoint /rest/active-workflows returned every active workflow ID on the instance to any member, regardless of sharing, and workflow activation, deactivation and publication push events were broadcast to every connected client, carrying workflow IDs, version IDs and activation-error detail across users. The patch scopes both the listing and the push events through the sharing service, so a user only sees workflows they can access.

The endpoints /rest/roles/:slug/assignments and /rest/roles/:slug/assignments/:projectId/members checked only that the caller could manage the role type, not that they could see the project named in the request. A user holding role-management permission could therefore name any project on the instance and read back its members' names and email addresses. The patch adds a project-access check to both routes, hiding projects the caller cannot see and returning not-found for a project it …

The /chat WebSocket route resumed a paused execution from a resume token without checking that the node being resumed was a chat node. n8n hands that token to anonymous form submitters, so a party with no account could present it on the chat route and release an execution waiting at an approval gate, the control an author puts in front of an action considered too consequential to automate. The patch …

mistral.rs fetches any request-supplied image/audio URL with no host or IP validation, and opens arbitrary local files (a file:// URL, or any existing relative/absolute path). A remote, unauthenticated client of any vision/audio deployment can cause the server to issue requests to internal or cloud-metadata addresses (SSRF) and to open arbitrary local files, via the standard OpenAI image_url / audio_url message content. The server is unauthenticated by default.

Excelize's prior row-bound fix for GHSA-h69g / CVE-2026-54063 protects the checked worksheet parser, but the streaming worksheet reader used by Rows and GetRows does not enforce the same TotalRows bound on the row r attribute. A small XLSX file can set a row number above Excelize's maximum row (1048576) and omit the cell coordinate. GetRows then appends empty rows up to the attacker-controlled row index and returns success. This was …

Excelize's prior row-bound fix for GHSA-h69g / CVE-2026-54063 protects the checked worksheet parser, but the streaming worksheet reader used by Rows and GetRows does not enforce the same TotalRows bound on the row r attribute. A small XLSX file can set a row number above Excelize's maximum row (1048576) and omit the cell coordinate. GetRows then appends empty rows up to the attacker-controlled row index and returns success. This was …

Excelize parses shared-string cell values with strconv.Atoi and checks only the upper bound before indexing the shared string slice. If an XLSX file contains a shared-string cell with <v>-1</v>, the parsed index is negative. The upper-bound check still passes (len(sharedStrings) > -1), and Excelize indexes sharedStrings[-1], causing a runtime panic. This was reproduced on the current default branch commit 1213a8bd7c5ab360554603ac5c995ccaf6eb4314 and the latest release tag v2.10.1 (5ad5ab3af0054c55bdce09f1530085600e9f2e45). The issue is …

Excelize parses shared-string cell values with strconv.Atoi and checks only the upper bound before indexing the shared string slice. If an XLSX file contains a shared-string cell with <v>-1</v>, the parsed index is negative. The upper-bound check still passes (len(sharedStrings) > -1), and Excelize indexes sharedStrings[-1], causing a runtime panic. This was reproduced on the current default branch commit 1213a8bd7c5ab360554603ac5c995ccaf6eb4314 and the latest release tag v2.10.1 (5ad5ab3af0054c55bdce09f1530085600e9f2e45). The issue is …

A discrepancy between WHATWG URL parsing and Angular SSR's URL resolution allows attackers to bypass same-origin checks and cause Server-Side Request Forgery (SSRF), potentially leaking sensitive server-side credentials.

An XSS vulnerability exists in @angular/platform-server during server-side rendering (SSR) HTML serialization when traversing ancestor tags across <template> element boundaries. When an application renders untrusted user input within raw-text tags (<xmp>, <style>, <script>), comments, or text nodes inside a <template> that is nested within a fallback raw-content element (<noscript>, <iframe>, <noembed>, <noframes>), matching closing tags (e.g., </noscript>) are not escaped during HTML serialization. When rendered in a browser, this unescaped …

Angular automatically sanitizes untrusted values bound to security-sensitive DOM sinks (such as href, src, action, xlink:href, and data) to protect against Cross-Site Scripting (XSS). Prior to the fix, the Angular compiler determined the SecurityContext for directive host bindings (host: {'[attr.href]': 'value'} or @HostBinding('attr.href')) based solely on the declaring directive or component selector at compile time, rather than the concrete host element that the directive was applied to. When a directive …

Angular automatically sanitizes untrusted values bound to security-sensitive DOM sinks (such as href, src, action, xlink:href, and data) to protect against Cross-Site Scripting (XSS). Prior to the fix, the Angular compiler determined the SecurityContext for directive host bindings (host: {'[attr.href]': 'value'} or @HostBinding('attr.href')) based solely on the declaring directive or component selector at compile time, rather than the concrete host element that the directive was applied to. When a directive …

A security bypass vulnerability was discovered in @angular/common when Server-Side Rendering (SSR) and hydration are enabled in applications using a hierarchical HttpClient configuration with withRequestsMadeViaParent(). The HttpTransferCache utility optimizes hydration by caching outgoing HTTP requests performed during SSR and transferring the cached state to the client-side application via TransferState (serialized as JSON in <script id="ng-state">). Following the remediation of CVE-2026-50170, HttpTransferCache automatically skips caching requests that contain authentication headers or …

@jhb.software/payload-alt-text-plugin v0.7.0 exposes custom Payload CMS endpoints (POST /api/alt-text-plugin/generate and /bulk) that call the Payload Local API (findByID and update) without setting overrideAccess: false. Because Payload's internal logic evaluates shouldOverrideAccess = overrideAccess !== false, omitting the parameter causes it to default to true, silently bypassing all collection-level access control functions. Any authenticated user — regardless of role — can read and overwrite the alt and keywords fields of arbitrary upload …

Embedded font-family names (word/fontTable.xml) were interpolated unescaped into an injected @font-face <style> and into the print window's document.write(). A crafted name injects page-wide CSS on open, and breaks out of <style> into executable HTML on Print.

Embedded font-family names (word/fontTable.xml) were interpolated unescaped into an injected @font-face <style> and into the print window's document.write(). A crafted name injects page-wide CSS on open, and breaks out of <style> into executable HTML on Print.

@argos-ci/core@6.2.0 passes attacker-controlled CI branch/ref strings directly into an execSync() template literal in packages/core/src/ci-environment/git.ts:89. When a CI project has hasRemoteContentAccess: false, the Argos upload flow calls getMergeBaseCommitSha(), which invokes gitFetch() with the unsanitized branch name. Because execSync() passes the command string to /bin/sh -c, shell metacharacters such as $() command substitution are evaluated before git runs, enabling an attacker who can influence the branch name (e.g., via a pull request) …

POST /api/acp/agents registers a custom ACP agent. The endpoint accepts user-controlled binary and versionCommand values. After saving the custom agent, the same request calls refreshAgentCache(), which triggers agent version detection. The version probe eventually runs: execFileSync(probe.command, probe.args, …) The only validation is resolveVersionProbe(binary, versionCommand, true), which checks that the first token of versionCommand matches the request-provided binary. Because binary is also attacker-controlled, an attacker can submit: { "binary": "node", "versionCommand": …

The POST /v1/chat/completions endpoint in mistral.rs fetches attacker-supplied media URLs (image, audio, video) into server memory with no byte limit, and extracts every frame of a supplied video when num_frames is None. An unauthenticated remote attacker can exhaust server memory, disk space, and CPU by pointing the endpoint at an infinite-streaming HTTP server or a long high-framerate video, causing a complete denial of service. No credentials or special configuration are …

Before 1.22, if the Basic Auth (htpasswd) middleware was not configured, all incoming HTTP headers were blindly forwarded to the webhook script execution environment as shell variables. While the Basic Auth middleware correctly strips the authentication header (X-WebAuthn-User) from the incoming request before conditionally re-injecting it on successful authentication, disabling Basic Auth left the system vulnerable if deployed behind an unhardened reverse proxy.

url_fetcher is WeasyPrint's documented mechanism for restricting resource loading - applications use it to block file://, internal hosts, etc. when rendering untrusted input. Two write_pdf() channels ignore the document's url_fetcher and build a fresh default URLFetcher() instead. A restrictive fetcher set on HTML() is silently bypassed for: xmp_metadata=[url] - the URL is fetched and the bytes are embedded verbatim in the output PDF. This is an arbitrary local file read …

ModelView.sort_query() uses the attacker-controlled sortBy list-view query parameter without checking it against the configured column_sortable_list allow-list. The value is resolved with getattr(model, …) and fed into relationship joins and order_by(), so a request can sort by any column of the model — including ones hidden from column_list — and, via a dotted path, by columns of related models. Because row order then reflects the value of an unexposed column, this …

parse() can be forced into an infinite loop when a value inside an array or inline table is followed by a comment that has no trailing newline (i.e. the comment "ends" the document). The library fails to exit an internal loop when attempting to find the end of the structure, resetting its cursor to the beginning of the string instead. This causes the parser to indefinitely hang, compromising the availability …

Open WebUI protects server-side web fetches with two controls: the operator's list of excluded hosts, and a check that refuses private and internal addresses. Neither control was applied to the destination of an HTTP redirect. On a deployment where redirect following is enabled, any authenticated user who can cause the server to fetch a URL could submit a page that redirects, and the server would fetch the redirect destination without …

Chat histories are stored as an unvalidated JSON object. After a message is deleted, the code that picks the chat's new current message walked down the childrenIds links without recording where it had already been. Any account with the default user role could store a chat whose messages list each other as children, then delete a message from it, and the walk would run forever. That walk runs on the …

Chat histories are stored as an unvalidated JSON object. The walk that reconstructs a chat's message chain detected repeats using each message's own id field while moving through the history by map key, so a message that simply omitted id was never recorded as visited. A history whose messages referenced each other in a parent cycle therefore made the walk run forever. Any account with the default user role could …

nuxt-ollama@1.2.26 unconditionally merges all module options — including api_key — into Nuxt's public runtime config (runtimeConfig.public.ollama). Nuxt serializes runtimeConfig.public into the SSR HTML response inside a <script> payload block (window.NUXT), making the API key visible in plaintext to any unauthenticated HTTP client that fetches the page. An attacker with no credentials can steal the Ollama cloud API key with a single HTTP GET request, then use it to make arbitrary …

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.DiaSymReader.Native. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Microsoft.DiaSymReader.Native can perform an out of bounds write while processing MSFZ PDB files.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.DiaSymReader.Native. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. An out of bounds write can occur when processing PDB files.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.DiaSymReader.Native. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. An out of bounds write can occur when parsing a Portable PDB file.

Microsoft is releasing this security advisory to provide information about a vulnerability in ASP.NET Core IIS out-of-process hosting and request decompression. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. The IIS Middleware for ASP.NET Core did not properly constrain decompression of certain types of request, leading to excess memory consumption and a Denial of Service.

Server-side request forgery (SSRF) vulnerability in eKuiper allows an attacker with permissions to register external services or create rules to induce the eKuiper server to make requests to unintended network locations, such as internal services, loopback interfaces (localhost), or cloud metadata endpoints.

A Cross-Site Scripting (XSS) vulnerability in external service creation allows an authenticated attacker to inject HTML/script payloads into external service names, which may execute in a user's browser when rendered by administrative web interfaces.

A path traversal vulnerability in eKuiper's administrative management endpoints allows privileged users or attackers with access to management APIs to delete arbitrary files or directories on the host system.

Vulnerability Overview The session_token cookie is set without the SameSite or Secure attributes (login.go:68). All /api/admin/ management endpoints rely solely on this cookie for authentication, with no CSRF token or Origin validation. The server-side vulnerability is confirmed to exist; however, exploitation via cross-site requests is mitigated in modern browsers by the default SameSite=Lax behavior. Root Cause // komari-main/api/public/login.go:68 c.SetCookie("session_token", session, 2592000, "/", "", false, true) // Secure=false, SameSite not explicitly …

In Joker versions before 1.8.2, joker –lint <file> located a .jokerd/ directory by walking up from the linted file and executed matching linter.* files from that directory before linting. Because these files are executable Joker/Clojure code, linting a file inside an untrusted repository could execute code supplied by that repository. This could be triggered by editor integrations or CI jobs that automatically run joker –lint on checked-out source code.

identrail's GitHub App connection-completion endpoint binds a fully client-supplied installation_id to the caller's workspace without verifying that the installation belongs to, or was installed by, the workspace that initiated the connect flow. identrail then mints a GitHub App installation access token for the supplied installation_id using the app's own JWT, so an authenticated tenant can link any other identrail customer's GitHub App installation into their own workspace and read that …

In a process run with full administrative rights on Windows, gix-sec wrongly treats all locations as trusted, leading to the execution of commands configured in repositories controlled by limited user accounts.

GitHacker through 1.1.7 did not validate path segments parsed from attacker-controlled .git/HEAD before joining them onto its output directory. A malicious server could coerce GitHacker into reading arbitrary local files. Contents do not stream back wholesale, but the recovery loop turns any 40-character hex substring into an outbound HTTP GET — an existence oracle for arbitrary paths plus hex-fragment exfiltration of file contents.

An unauthenticated server-side request forgery vulnerability lets any anonymous user make the GeoNetwork server issue arbitrary outbound HTTP requests. This gives an external attacker a position inside the server's network, making it possible to make internal requests no matter if the response is XML-type or not. The SLD tooling endpoint POST /api/tools/ogc/sld takes a caller-supplied WMS server URL and performs a server-side HTTP GET to it, with no validation. The …

The set_functype_version MCP tool in functype-mcp-server accepts an unconstrained version string, interpolates it directly into an npm package specifier (functype@<version>), and installs it via pnpm add without any validation. Because npm/pnpm package specifiers support file:, npm:, and other alias syntaxes, an attacker who can send an MCP tools/call request to this tool can cause the server to install an arbitrary local or remote package as functype. Immediately after installation, the …

On the Home Assistant add-on, the dashboard serves a trusted ingress site that skips authentication because the supervisor authenticates the request upstream. That site was binding 0.0.0.0. The add-on runs in host network mode for mDNS, so binding all interfaces also bound the host's LAN interface, and any device on the local network could reach http://<ha-ip>:<ingress_port>/ and get the full dashboard with no credentials.

A low-privilege process-scoped admin who can manage elections can store arbitrary HTML in the question statement/body without sanitization, and the public elections UI renders that value unsafely.

A bug in containerd's CRI ExecSync implementation allows exec probes and lifecycle hooks with background child processes to keep containerd's stdio-drain goroutines indefinitely blocked. Because the I/O drain phase lacks a default timeout or context cancellation handling, repeated ExecSync invocations (like probes) that include long-lived background processes against a container can cause containerd to leak goroutines and host memory. Over time, this resource exhaustion can cause the containerd daemon to …

A bug in containerd's CRI ExecSync implementation allows exec probes and lifecycle hooks with background child processes to keep containerd's stdio-drain goroutines indefinitely blocked. Because the I/O drain phase lacks a default timeout or context cancellation handling, repeated ExecSync invocations (like probes) that include long-lived background processes against a container can cause containerd to leak goroutines and host memory. Over time, this resource exhaustion can cause the containerd daemon to …

@yeger/turbo-graph starts its embedded Next.js server without binding to the loopback interface, causing it to listen on all network interfaces (0.0.0.0:29312 by default). The /api/run HTTP endpoint exposed by this server performs no authentication, authorization, CSRF protection, or task allowlist check before executing attacker-supplied Turborepo task names via spawn(). Any adjacent-network attacker can send an unauthenticated GET request to trigger arbitrary tasks defined in the victim's repository, resulting in code …

@openhop/server passes unsanitized HTTP route parameters directly to path.join() when constructing filesystem paths for flow YAML files. An unauthenticated attacker who can reach the server can read arbitrary .yaml files accessible to the OpenHop process outside the configured flow directory, and can delete arbitrary .yaml files at any path reachable by the process. Because CORS is set to origin: true (allow all origins), a victim's browser can be used to …

An embedded line terminator bypasses the requireWellFormed serializer check for element and attribute names. The check was added to fix GHSA-w2rr-34g9-rvrj and GHSA-4w3w-2rp5-g8jm; a name whose first line is well-formed slips past it and is serialized verbatim, so the characters after the line terminator break out of the start/end tag or attribute. Callers who enabled requireWellFormed specifically to neutralize those name-injection issues remain exposed.

An embedded line terminator bypasses the requireWellFormed serializer check for a DocumentType's publicId and systemId. The check was added to fix GHSA-f6ww-3ggp-fr8h; an id whose first line is a valid literal slips past it and is emitted verbatim into the <!DOCTYPE …> declaration, so the markup after the line terminator breaks out into the surrounding document. Callers who enabled requireWellFormed to neutralize DocumentType injection remain exposed.

xmldom's malformed-input error-recovery path has two quadratic-time (O(n²)) behaviors that a single crafted input triggers together, so a tiny, highly compressible document (tens of KB) stalls the Node.js event loop for multiple seconds. It is reachable from DOMParser.parseFromString under default options — i.e. from unauthenticated, network-delivered XML — making this an unauthenticated denial of service. One of the two behaviors, the normalize() adjacent-text merge, is additionally reachable programmatically — via …

xmldom's malformed-input error-recovery path has two quadratic-time (O(n²)) behaviors that a single crafted input triggers together, so a tiny, highly compressible document (tens of KB) stalls the Node.js event loop for multiple seconds. It is reachable from DOMParser.parseFromString under default options — i.e. from unauthenticated, network-delivered XML — making this an unauthenticated denial of service. One of the two behaviors, the normalize() adjacent-text merge, is additionally reachable programmatically — via …

xmldom builds the attribute collection of every parsed element by inserting attributes one at a time into a DOM NamedNodeMap. Each insertion first performs a linear scan of all already-inserted attributes to enforce the DOM uniqueness rule (no two attributes with the same qualified name / namespace+local-name). Parsing an element that carries M distinct attributes therefore costs 1 + 2 + … + M = O(M²) comparisons. Because the trigger …

xmldom builds the attribute collection of every parsed element by inserting attributes one at a time into a DOM NamedNodeMap. Each insertion first performs a linear scan of all already-inserted attributes to enforce the DOM uniqueness rule (no two attributes with the same qualified name / namespace+local-name). Parsing an element that carries M distinct attributes therefore costs 1 + 2 + … + M = O(M²) comparisons. Because the trigger …

When an element declares a namespace prefix, xmldom copies the entire in-scope namespace map into a fresh object and keeps that copy on the element while it is open on the parse stack. A crafted document that nests N elements, each declaring one unique prefix, therefore drives the parser to hold on the order of N(N+1)/2 = O(N²) namespace-map entries at its peak, so a small, highly compressible input exhausts …

When an element declares a namespace prefix, xmldom copies the entire in-scope namespace map into a fresh object and keeps that copy on the element while it is open on the parse stack. A crafted document that nests N elements, each declaring one unique prefix, therefore drives the parser to hold on the order of N(N+1)/2 = O(N²) namespace-map entries at its peak, so a small, highly compressible input exhausts …

Document.createProcessingInstruction() in @xmldom/xmldom performs no validation on the target parameter. The requireWellFormed: true serializer option validates only for : in the target and a case-insensitive xml prefix, but does not check for > characters. A > in the target breaks the processing instruction boundary (<?…?>), allowing injection of arbitrary content into the serialized XML output.

Document.createProcessingInstruction() in @xmldom/xmldom performs no validation on the target parameter. The requireWellFormed: true serializer option validates only for : in the target and a case-insensitive xml prefix, but does not check for > characters. A > in the target breaks the processing instruction boundary (<?…?>), allowing injection of arbitrary content into the serialized XML output.

xmldom's parser silently accepts a not-well-formed end tag whose valid name is followed by trailing content — e.g. </a⏎junk>. The element is closed, the trailing content is discarded, and no error is reported, even though the XML end-tag production allows only optional whitespace after the name and both Chromium and Firefox reject such input as application/xml. An application that relies on xmldom to reject not-well-formed input therefore receives a false …

xmldom's parser silently accepts a not-well-formed end tag whose valid name is followed by trailing content — e.g. </a⏎junk>. The element is closed, the trailing content is discarded, and no error is reported, even though the XML end-tag production allows only optional whitespace after the name and both Chromium and Firefox reject such input as application/xml. An application that relies on xmldom to reject not-well-formed input therefore receives a false …

In HTML mode (text/html), a raw-text element (script, style, textarea, title) whose closing tag differs in case from its opening tag (e.g. </ScRiPt> for <script>) is mishandled by the parser, producing quadratic (O(n²)) output growth — a small crafted document parses and serializes into output orders of magnitude larger, exhausting CPU and memory. A modest input of tens of KB can therefore cause a denial of service in any service …

On the @xmldom/xmldom 0.8.x line, parsing an XML end tag whose name is followed by a long run of whitespace and then a non-whitespace character triggers quadratic-time regular-expression backtracking (ReDoS), so a single small crafted end tag stalls the Node.js event loop. It is reachable from DOMParser.parseFromString under default options, unauthenticated, before any validity check — an availability-only denial of service. The 0.9.x line is not affected.

Document.createElement() in @xmldom/xmldom accepts arbitrary strings as the tagName parameter with zero validation. The serializer emits the tag name verbatim into XML/HTML output. Critically, the requireWellFormed: true serializer option — the recommended mitigation from CVE-2026-41672, CVE-2026-41674, and CVE-2026-34601 — did NOT catch this, making it a bypass of the existing security controls. An attacker who controls the element name string can inject arbitrary attributes (including event handlers) into the serialized …

Document.createElement() in @xmldom/xmldom accepts arbitrary strings as the tagName parameter with zero validation. The serializer emits the tag name verbatim into XML/HTML output. Critically, the requireWellFormed: true serializer option — the recommended mitigation from CVE-2026-41672, CVE-2026-41674, and CVE-2026-34601 — did NOT catch this, making it a bypass of the existing security controls. An attacker who controls the element name string can inject arbitrary attributes (including event handlers) into the serialized …

The @xmldom/xmldom serializer emits DocumentType.name verbatim into the <!DOCTYPE …> declaration with no well-formedness guard. GHSA-f6ww-3ggp-fr8h (CVE-2026-41674) hardened the serializer's requireWellFormed path for a DocumentType's sibling fields — publicId, systemId, and internalSubset — but it did not add any check for name. A > (or whitespace) in the name terminates the doctype declaration early, letting the remaining characters become sibling markup in the serialized output. Because requireWellFormed: true — the …

The @xmldom/xmldom serializer emits DocumentType.name verbatim into the <!DOCTYPE …> declaration with no well-formedness guard. GHSA-f6ww-3ggp-fr8h (CVE-2026-41674) hardened the serializer's requireWellFormed path for a DocumentType's sibling fields — publicId, systemId, and internalSubset — but it did not add any check for name. A > (or whitespace) in the name terminates the doctype declaration early, letting the remaining characters become sibling markup in the serialized output. Because requireWellFormed: true — the …

An embedded line terminator bypasses xmldom's always-on, WHATWG-mandated creation-time name validation. createElementNS, createAttributeNS, createDocumentType, and createAttribute should reject a malformed qualified name with InvalidCharacterError, but a name whose first line is well-formed slips through and enters the DOM. On serialization it is emitted verbatim, so the characters after the line terminator inject markup into the output. The injection reaches the default serialization path, and enabling requireWellFormed does not prevent it.

Element.setAttribute() in @xmldom/xmldom bypasses attribute name validation by calling the private _createAttribute(name) method, which performs no validation. The public createAttribute() method correctly validates names against an anchored QName pattern, but setAttribute() never uses it. The serializer escapes attribute values but trusts attribute names, allowing an attacker to inject additional attributes (including event handlers) into serialized output. The requireWellFormed: true option did not catch this.

Element.setAttribute() in @xmldom/xmldom bypasses attribute name validation by calling the private _createAttribute(name) method, which performs no validation. The public createAttribute() method correctly validates names against an anchored QName pattern, but setAttribute() never uses it. The serializer escapes attribute values but trusts attribute names, allowing an attacker to inject additional attributes (including event handlers) into serialized output. The requireWellFormed: true option did not catch this.

@xmldom/xmldom's processing-instruction (PI) grammar regex exhibits quadratic-time backtracking (ReDoS) when parsing an unterminated processing instruction. A single small XML document containing <? + a target + a long run of whitespace and no closing ?> forces the regular expression engine into O(n²) work, stalling the Node.js event loop. The input is parsed with DOMParser.parseFromString under default options, so it is reachable from unauthenticated, network-delivered XML (SOAP/SAML, webhooks, uploads, XML APIs).

Case Description: MSRC Notes: Attachments: 1 file(s) attached (1 mp4) Summary: The vulnerability lies in the 'serve/cli_api.py' component of the 'winml-cli' project, which exposes all winml CLI commands over HTTP without authentication. Although it binds to localhost by default, it sets 'allow_origins' to a wildcard, allowing any website to interact with the endpoint. This, combined with the '–trust-remote-code' flag in 'build' and 'config' commands, enables an attacker to execute arbitrary …

vllm/transformers_utils/processors/mimo_v2_omni.py — the multimodal processor for MiMoV2OmniForCausalLM — issues requests.get(…) directly on user-supplied image and audio URL strings and Image.open(…) on user-supplied local paths, without the SSRF / allowed_local_media_path checks that vllm.multimodal.utils.MediaConnector was hardened with in GHSA-qh4c-xf7m-gxfc, GHSA-v359-jj2v-j536, and GHSA-pf3h-qjgv-vcpr. This is the same bug class as those three published advisories, in a code path the patches missed. When a user passes a URL or local-file string through multi_modal_data (e.g. …

An integer overflow in the act_and_mul_kernel kernel can cause the output of one user request to be incorporated into the response of another request within the same inference batch. Under certain conditions, the last request in a batch can receive a partial or complete copy of the first user's inference result, resulting in cross-user data leakage.

@vitest/mocker registers a redirect mock's target path without validating it against the dev server's file-serving allowlist. An attacker who can reach the dev server's WebSocket can register a redirect mock pointing outside the project root; when the mocked module is requested, the plugin's load hook returns readFile(<attacker path>) as the module source, disclosing local files. This is exploitable without authentication only through the public mockerPlugin / standalone interceptorPlugin exports (used …

@vitest/mocker registers a redirect mock's target path without validating it against the dev server's file-serving allowlist. An attacker who can reach the dev server's WebSocket can register a redirect mock pointing outside the project root; when the mocked module is requested, the plugin's load hook returns readFile(<attacker path>) as the module source, disclosing local files. This is exploitable without authentication only through the public mockerPlugin / standalone interceptorPlugin exports (used …

@tiptap/core contains two quadratic regular-expression denial-of-service paths in its default Markdown attribute parsers. Pandoc-style block attributes use two unanchored greedy expressions that rescan repeated __QUOTED_0 prefixes. Inline shortcode attributes use another unanchored greedy key expression that rescans a long word-character run when no equals sign follows. The public createAtomBlockMarkdownSpec and createBlockMarkdownSpec helpers call the vulnerable Pandoc-style parser; createInlineMarkdownSpec calls the separately vulnerable shortcode parser. Using unmodified npm 3.29.2, a complete …

@swc/html minifies JSON contained in script elements such as application/json and application/ld+json by parsing and serializing the JSON value. Before the patched versions, JSON serialization could convert escaped less-than signs such as \u003C into literal < characters. If the JSON contained an escaped </script> sequence, the generated HTML could terminate the containing script element early because HTML tokenization occurs before the JSON is consumed. Applications that minify HTML containing attacker-controlled …

@swc/html minifies JSON contained in script elements such as application/json and application/ld+json by parsing and serializing the JSON value. Before the patched versions, JSON serialization could convert escaped less-than signs such as \u003C into literal < characters. If the JSON contained an escaped </script> sequence, the generated HTML could terminate the containing script element early because HTML tokenization occurs before the JSON is consumed. Applications that minify HTML containing attacker-controlled …

SVGO's opt-in removeScripts plugin did not inspect executable HTML content inside SVG <foreignObject> elements. Applications that used this plugin as their only protection for untrusted SVG input could produce SVGs containing active HTML and expose users to cross-site scripting (XSS). SVGO is an optimizer rather than a comprehensive sanitization library, but removeScripts is maintained for consumers that already rely on it to remove common script execution paths.

SVGO's opt-in removeScripts plugin failed to remove some executable links. Namespace-prefixed SVG anchors and URL schemes containing ASCII tabs or newlines could bypass its checks. Applications that used this plugin as their only protection for untrusted SVG input could expose users to cross-site scripting (XSS). SVGO is an optimizer rather than a comprehensive sanitization library, but removeScripts is maintained for consumers that already rely on it to remove common script …

publishAccess.json is an opt-out list. The publish gate returns accessible for anything not explicitly listed in it. Encrypted notebooks are never written into that file, because only the administrator-gated setPublishAccess writes it and no part of the encryption subsystem does. Consequently every encrypted notebook is publish-accessible as far as the gate is concerned. While an encrypted notebook is unlocked, an anonymous reader in publish mode can list it, enumerate its …

/api/notebook/getNotebookInfo is registered with CheckAuth only and applies no authorization check of any kind. Given a notebook identifier, it returns that notebook's author-written name, document count, total size and creation and modification timestamps. It does so for every notebook the server has loaded, including notebooks that lsNotebooks deliberately withholds from readers because they are closed or not publish-visible.

/api/av/getAttributeViewFieldViews is registered with CheckAuth only and applies no publish-access filtering. Given a database avID, it returns the complete view structure of that database: every view's name, icon, layout type and the per-field Hidden flag, regardless of whether the caller is authorized to see that database or those views. The route registered immediately before it, getAttributeViewKeys, does gate reader roles.

A number of vulnerabilities, two rated as "Critical" severity using CVSSv3, have been discovered and fixed in the upstream libheif dependency. These can lead to possible remote code execution (RCE) on glibc-based Linux when run under certain conditions. The attack vector for these claims to be "network" however sharp does not provide any networking features so this vulnerability is down-rated to a "High" severity using CVSSv4 but please note its …

An OS command injection in repository git_url handling lets any user holding the Manager or Owner role on any project (the normal project-collaborator roles) achieve remote code execution on the Semaphore server host. Using git's –upload-pack= option, an attacker runs arbitrary commands. The command executes inside the main Semaphore server process (via the schedule commit-hash poller), so it runs even when jobs are configured for remote runners — bypassing runner …

Prowler's HTML output formatter inserts finding.resource_tags into the generated report without HTML escaping. A cloud principal who can create or edit a resource tag in an account that is later scanned can store HTML or JavaScript in that tag. When another user opens the generated Prowler HTML report, the payload executes in the report page.

Prowler's HTML output formatter inserts finding.resource_tags into the generated report without HTML escaping. A cloud principal who can create or edit a resource tag in an account that is later scanned can store HTML or JavaScript in that tag. When another user opens the generated Prowler HTML report, the payload executes in the report page.

An improper CRLF neutralization flaw in Predis' pipeline handling on aggregate connections lets an unauthenticated attacker who can influence any pipelined argument — a value or a key, e.g. a URL slug used as a cache key — smuggle arbitrary Redis commands into the connection. On cluster connections (cluster option, incl. client-side sharding) this is remote command injection: shard-wide FLUSHDB, targeted DEL/SET, same-slot key theft via GET, cache poisoning, and …

The pure-PHP X25519 scalar multiplication in phpseclib is not constant-time. Field addition and subtraction each perform a data-dependent conditional modular reduction, so the cost of each Montgomery-ladder step is a linear function of that step's reduction count which is a quantity determined by the secret scalar's prefix. An observer with per-ladder-step resolution recovers the 251-bit clamped private scalar. This is a per-step leak, not an aggregate one: an instrumented code …

The @typespec/openapi3 emitter retains the value of a @versioned enum member and interpolates it into the output filename as {version} without sanitizing path separators or traversal components. The completed path reaches the compiler's emitFile(), which creates the parent directory and writes the file without verifying containment under emitterOutputDir. A crafted declarative .tsp input can therefore create or overwrite an OpenAPI-formatted .yaml or .json file outside the configured output tree, subject …

The @typespec/openapi3 emitter retains the value of a @versioned enum member and interpolates it into the output filename as {version} without sanitizing path separators or traversal components. The completed path reaches the compiler's emitFile(), which creates the parent directory and writes the file without verifying containment under emitterOutputDir. A crafted declarative .tsp input can therefore create or overwrite an OpenAPI-formatted .yaml or .json file outside the configured output tree, subject …

Nodemailer's disableFileAccess / disableUrlAccess options are a security sandbox that lets an application forbid untrusted message content (html/text/attachment path/href) from reading local files or making outbound HTTP(S) requests. The fix for GHSA-wqvq-jvpq-h66f (commit 5f69497) threaded these flags through the library's internal resolution paths (MailMessage.resolveAll() and _convertDataImages()), but the public plugin API MailMessage.resolveContent(…args) (lib/mailer/mail-message.js:41-43) remains a raw passthrough to shared.resolveContent(). When called with the documented legacy signature mail.resolveContent(data, key, callback), shared.resolveContent …

Nodemailer's email-address parser treats an RFC 5322 comment ( … ) inside the domain as a point to concatenate the surrounding text, rather than as folding whitespace (CFWS) that terminates the domain. Consequently a recipient address such as user@good-corp.com(x)evil.com is parsed and delivered to good-corp.comevil.com (registrable domain comevil.com, attacker‑controlled), while a conformant RFC 5322 parser terminates the domain at the comment and reads good-corp.com. An application that decides whether it …

Nodemailer's address parser (lib/addressparser/index.js) parses a list of comma‑separated addresses in quadratic time — O(n²) in the number of addresses. A single crafted address string (e.g. a To, Cc, Bcc, From, or Reply‑To value, or any value passed to the exported addressparser) therefore consumes CPU proportional to the square of its length and blocks Node's single‑threaded event loop for the entire duration, denying service to every other request in the …

Nodemailer resolves an international (IDN / non-ASCII) recipient domain to a different Punycode xn– label than every UTS‑46‑conformant parser (web browsers, the WHATWG URL Standard, Node's url.domainToASCII, Python's idna). Its address normalizer (_normalizeAddress in lib/mime-node/index.js) uses the bundled raw RFC‑3492 Punycode codec with no UTS‑46 mapping/normalization, so a domain that a standards‑compliant validator maps to a trusted domain is delivered by Nodemailer to a different, attacker‑registrable domain. An application that …

With columns: true and group_columns_by_name: true, a duplicated proto header causes the duplicate-column branch to assign an array to obj['proto'], invoking the proto setter and replacing the parsed record object's prototype with attacker-controlled data. Fixed in 7.0.2 (Object.hasOwn duplicate check + Object.defineProperty assignment).

The NLTK library's TransitionParser.parse() method deserializes model files using pickle_load() with the default restricted=False parameter, allowing arbitrary Python code execution when loading a malicious model file. The library provides a RestrictedUnpickler class for safe deserialization, but it is never used by production code paths, leaving the vulnerability unpatched.

The original advisory was fixed (PR #3581) by adding _reject_unsafe_path_component(), which blocks literal /, , .., and Windows drive prefixes in caller-/corpus-supplied names. It never resolves symlinks. All three call sites that use this guard still resolve the resulting path through self.abspath() (nltk/corpus/reader/api.py, self._root.join(fileid)), which is a plain lexical join, not the symlink-resolving, required_root-scoped check that CorpusReader.open() (and NKJPCorpusReader's own fix for its sibling advisory) correctly use elsewhere in this …

IPIPANCorpusReader (nltk/corpus/reader/ipipan.py) exposes public methods, channels(), domains(), categories(), and fileids(channels=…), that accept a caller supplied fileids list and read a file via a completely unprotected builtin open() call, with no nltk.pathsec involvement at all. A symlink placed inside the corpus root, with a name containing no separators or .., passes NLTK's existing traversal checks and is opened directly, reading a file from anywhere on the filesystem the process can access.

nltk.corpus.reader.api.CorpusReader.open() can be used to read files outside the intended corpus root via a symlink placed inside that root. Although NLTK blocks absolute paths and .. traversal, the current boundary check is only lexical and does not account for symlink resolution. This leads to an arbitrary local file read / filesystem sandbox bypass for applications that rely on CorpusReader or FileSystemPathPointer to restrict file access.

Setting nltk.pathsec.ENFORCE = True is documented to sandbox all file access to allowed NLTK data directories and raise PermissionError on unauthorized access. However, StreamBackedCorpusView opens files via builtins.open() directly, bypassing pathsec.validate_path() entirely. An attacker who can influence the fileid argument can read arbitrary local files regardless of the ENFORCE setting.

Published nltk==3.9.4 still contains several XML-reader entrypoints that build parser paths from caller-controlled selectors or trusted-looking index state without preserving the corpus-root boundary.

The NLTK tgrep module accepts user-supplied regular expressions and passes them to the Python re engine without a timeout or validation, enabling catastrophic backtracking (ReDoS). Applications that expose the tgrep API to external input are vulnerable to a single-request denial of service that blocks the Python process indefinitely.

Pl196xCorpusReader still parses whole TEI blocks with multiple lazy regexes over attacker-controlled text. A malformed file with many opening tags and no matching closing tags forces repeated rescans and produces quadratic CPU growth in public reader APIs.

Current NLTK source reopens SSRF in proxied environments. pathsec.urlopen() validates the requested hostname locally, but once proxy inheritance is enabled the real fetch is performed by the proxy rather than by the validated direct-connect socket path.

NLTK's package downloader in nltk/downloader.py does not verify file integrity after download and before extraction. The download flow at lines 789-825: File is downloaded to a temp path via HTTP os.replace(tmp_filepath, filepath) moves it to the final location (line 799) Extraction begins via _unzip_iter() (line 825) Between steps 2 and 3, there is no SHA-256 verification. The checksum logic exists in _pkg_status() (lines 982-1015) but it is only used BEFORE …

There's a logic bug in FileSystemPathPointer.open() inside nltk/data.py that makes the sandbox check permanently inert. The guard condition is always False — meaning any file the process can read is accessible by passing a file:// URL to nltk.data.load().

Several XML parsing sites in NLTK still used xml.etree.ElementTree directly, which honours <!ENTITY> declarations in a document's internal DTD subset. A crafted document a few hundred bytes long can expand to megabytes in memory (each nesting level multiplies by ten), a denial-of-service. Affected call sites (<= 3.10.2): nltk.chunk.named_entity.load_ace_file — parses ACE annotation XML nltk.internals.ElementWrapper — converts any given string to an Element nltk.downloader — Package.fromxml, Collection.fromxml, _find_collections, _find_packages libexpat 2.6.0 …

Several corpus readers still step outside NLTK's symlink-aware trusted-root model. They derive in-root paths from trusted corpus state, convert those paths back into plain strings, and reopen them with built-in open() rather than nltk.pathsec.open().

NLTK corpus-reader constructors can still reach outside-root file and database reads before the nltk.pathsec sandbox boundary is enforced. The PoC shows the safe path blocked by pathsec.open, then LinThesaurusCorpusReader and PanLexLiteCorpusReader succeeding in the same process.

The current source tree still allows arbitrary code execution during supposedly safer allowlisted pickle loading. The allowlist trusts whole module namespaces instead of exact safe globals, so crafted pickles can invoke dangerous in-namespace callables through pickle REDUCE.

A vulnerability in applications using Pages and App router without Cache Component can lead to remote code execution when the server is hosted on machines using a Windows filesystem.

A vulnerability in the underlying libheif library used by sharp which Next.js uses for image optimization can lead to remote code execution when AVIF files are optimized. Until a fix has propagated, optimization of AVIF files is disabled.

A fragmented TLS ClientHello whose handshake header spans multiple records makes Netty silently fall back to the default SslContext; where per-SNI selection is the sole mTLS gate, an unauthenticated attacker can bypass the route's mTLS requirement.

Netty's default SNI entrypoint reparses and recopies previously received ClientHello fragments on every additional TLS handshake record. A remote peer can send a small first record that advertises a large ClientHello length and then drip the body in many tiny records, causing superlinear (quadratic) CPU work before the handshake completes. With 4095 one-byte fragments, the handler recopies 8,386,560 bytes from only 24,579 bytes on the wire — a 341× amplification …

A workflow's "This workflow can be called by" setting was enforced by the Execute Workflow node but was not consulted when the same workflow was attached to an Agent as a tool. A user who could build an Agent could therefore call a workflow that its owner had restricted, and read back what it returned. The patch applies the sub-workflow caller policy on the Agent tool path.

When multer is configured with an asynchronous fileFilter, the limits.fileSize limit can be bypassed. The 'limit' event is registered inside the async fileFilter callback, so if a file exceeds limits.fileSize before that callback runs, the event is missed and the oversized upload is accepted instead of being rejected with a LIMIT_FILE_SIZE error. Applications that rely on limits.fileSize to reject oversized uploads are affected on all upload methods (.single(), .array(), .fields(), …

multer is vulnerable to a Denial of Service (DoS) via a crafted array index in multipart field names. The append-field dependency parses bracket notation in field names, and a large numeric index such as items[4294967294] forces allocation of a maximum-length sparse array. A following field with a non-numeric key on the same base then converts that array to an object by iterating its full length, which consumes CPU synchronously and …

A vulnerability in multer 2.2.0 allows an attacker to trigger a Denial of Service (DoS) by aborting or truncating multipart uploads. When using diskStorage, the destination write stream is not closed if the upload is aborted before it finishes, so each failed request leaks an open file descriptor and retains its disk blocks until the process exits. Repeated failed uploads can exhaust the available file descriptors. All applications using multer's …

A vulnerability in multer allows a remote, unauthenticated attacker to crash the Node.js process with a single multipart/form-data request. Two specially crafted text field names cause an uncaught RangeError: Invalid array length inside multer's field parsing, which is not routed to the application error handler and terminates the process. All applications using multer to parse multipart requests are affected.

Morgan writes attacker-controlled request data to the access log through its tokens. The 1.11.0 fix neutralizes C0 control characters, DEL, and backslash, but it does not escape the Unicode line separators U+0085 (NEL), U+2028 (LINE SEPARATOR), or U+2029 (PARAGRAPH SEPARATOR). These code points are reachable through several tokens, including the request URL (:url), request headers such as :user-agent and :referrer, arbitrary :req[header] values, and the Basic auth username (:remote-user). NEL …

Passing untrusted input as part of a database or collection name may result in targeting a different database or collection than specified.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.Build.Tasks.Git and Microsoft.SourceLink.AzureRepos.Git. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Improper removal of sensitive information before storage or transfer in .NET allows an unauthorized attacker to disclose information over a network.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.Build.Tasks.Git and Microsoft.SourceLink.AzureRepos.Git. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Improper removal of sensitive information before storage or transfer in .NET allows an unauthorized attacker to disclose information over a network.

Microsoft is releasing this security advisory to provide information about a vulnerability in Windows Presentation Foundation (WPF). This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. A remote code execution vulnerability exists in Windows Presentation Foundation (WPF) in .NET 8, .NET 9, and .NET 10 when parsing specially crafted XAML input. An attacker who successfully exploits this vulnerability could execute arbitrary …

Microsoft is releasing this security advisory to provide information about a vulnerability in Windows Presentation Foundation (WPF). This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. A remote code execution vulnerability exists in Windows Presentation Foundation (WPF) in .NET 8, .NET 9, and .NET 10 when parsing specially crafted XAML input. An attacker who successfully exploits this vulnerability could execute arbitrary …

Microsoft is releasing this security advisory to provide information about a vulnerability in Windows Presentation Foundation (WPF). This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. A remote code execution vulnerability exists in Windows Presentation Foundation (WPF) in .NET 8, .NET 9, and .NET 10 when parsing specially crafted XAML input. An attacker who successfully exploits this vulnerability could execute arbitrary …

Use after free in Microsoft QUIC allows an unauthorized attacker to execute code over a network.

Use after free in Microsoft QUIC allows an unauthorized attacker to execute code over a network.

DOM.sanitize() in src/util/dom.ts iterated elem.attributes (a live NamedNodeMap) while calling elem.removeAttribute() in the same loop. Removing an attribute shifts subsequent attributes down by one index, causing the iterator to skip the adjacent attribute. An attacker can provide an HTML payload with consecutive dangerous attributes (such as <details open onload="1" ontoggle="…">). The first attribute is stripped while the second survives and executes upon insertion into innerHTML via the attribution control without …

The join filter (src/filters/array.ts:8-13) charges memoryLimit by array element count, not by the string length it produces, letting a template bypass a configured memoryLimit and allocate strings far past budget — bounded only by V8/process limits, not by memoryLimit.

Excel::store() resolved the destination path against the process working directory rather than the configured filesystem disk. When that path resolved to an existing file, the export was written straight to it with fopen(), bypassing the disk entirely. An application that passes a user-controlled value as the export path could therefore be made to overwrite an arbitrary existing file that the PHP process can write to, with content the user controls.

maxTotalMergeKeys does not count empty mappings. An attacker can repeatedly merge a large sequence of them and consume significant CPU without reaching the configured limit.

An application that passes attacker-controlled data into joi's custom message configuration (messages(), message(), prefs({ messages }), Joi.extend({ messages }) or rule({ message })) lets the attacker write properties onto Object.prototype, where every object in the process then inherits them. A key named proto was treated as a language name, and the code reused the object it found at that key, which resolves to the prototype rather than to a new …

An application that passes attacker-controlled data into joi's custom message configuration (messages(), message(), prefs({ messages }), Joi.extend({ messages }) or rule({ message })) lets the attacker write properties onto Object.prototype, where every object in the process then inherits them. A key named proto was treated as a language name, and the code reused the object it found at that key, which resolves to the prototype rather than to a new …

Applications are affected only if a schema renames keys with a regular-expression source and a Joi.expression() / Joi.x() target that interpolates the pattern's own match data, combined with { multiple: true }, for example .rename(/^x-(.+)$/, Joi.x('{#1}'), { multiple: true }). Because the target is rendered from the matched input key, an attacker who controls input keys can send x-proto with an object value and make the rename target render as …

Sensitive information exposure in the Infracost Terraform Cloud/Enterprise integration. Several code paths attached a configured secret token to a request whose host was taken straight from the scanned Terraform, with no check that the host was the trusted endpoint. The condition that matters is whether a Terraform Cloud/Enterprise (or registry) token is present in the run: if one is, the injected hostname can disclose it. Impact therefore depends on the …

Path traversal via link following in the Infracost config-template parser. The readFile, pathExists, isDir, and matchPaths template functions confined repo-supplied paths with a lexical filepath.Rel check plus a leaf-only os.Lstat: {{ readFile "evil/file" }} With an intermediate directory symlink committed in the repo (evil -> /), the path is lexically clean and the leaf is a regular file, so both checks pass, but os.ReadFile follows the symlink and reads outside …

When decoding a compressed response body (gzip, deflate, br, or zstd), HTTPX2 fully decompressed each network read before yielding content to the application. A small compressed input could therefore cause a large intermediate memory allocation, even when the application streamed the response to keep memory usage bounded.

httpcore2 does not start TLS for wss:// connections routed through a SOCKS5 proxy. The WebSocket opening handshake and all subsequent frames are sent in plaintext through the proxy path, despite the caller selecting the secure wss scheme. The transport flaw affects httpcore2 releases before 2.10.0. HTTPX2 exposed this behavior through its public Client.websocket() and AsyncClient.websocket() APIs from 2.6.0 through 2.9.1.

httpcore2 does not start TLS for wss:// connections routed through a SOCKS5 proxy. The WebSocket opening handshake and all subsequent frames are sent in plaintext through the proxy path, despite the caller selecting the secure wss scheme. The transport flaw affects httpcore2 releases before 2.10.0. HTTPX2 exposed this behavior through its public Client.websocket() and AsyncClient.websocket() APIs from 2.6.0 through 2.9.1.

HTTPX2's Server-Sent Events (SSE) parser repeatedly copied and rescanned buffered text when a server split one unterminated line across many response chunks. The total work grows quadratically with the length of the line. An attacker-controlled or compromised SSE endpoint can exploit this behavior to consume excessive client CPU.

HTTPX2 serializes the per-file Content-Type and custom headers supplied through the files= tuple API directly into the multipart/form-data body without validating custom header names or values. An attacker who can influence upload metadata passed to HTTPX2 can use CR or LF characters to terminate a multipart part header and inject additional part headers or end the part header block early.

HTTPX2 can automatically add a Content-Length header to a request that already contains a caller-supplied Transfer-Encoding header. The resulting HTTP/1.1 request contains both framing headers, which can create an ambiguous message boundary and enable request smuggling or connection desynchronization when processed by intermediaries that disagree about which header takes precedence.

When parseBody() expands dot-separated form field names into nested objects, it does not limit the nesting depth or the total number of objects created. A request body well within a normal size limit can therefore allocate an object graph far larger than the request itself, and concurrent requests can exhaust the heap and terminate the process.

Hono's query parsing does not stop at the URL fragment: a ? appearing after a # is treated as the start of a query string. As a result, the application can read request parameters that no other component involved in handling the request can see.

The fix released for CVE-2026-39408 does not cover every traversal sequence. toSSG() can still write files outside the configured output directory when a route parameter contains consecutive parent-directory segments.

A vulnerability in the xDS RBAC HTTP filter implementation in grpc-go allows remote attackers to bypass authorization policies (specifically DENY rules) by using mixed-case or canonical-case header matchers (e.g., X-Role instead of x-role). Additionally, the safety guards introduced by gRFC A41 to block grpc- prefixed headers can be evaded via variations in casing (e.g., Grpc-Status).

A vulnerability exists in gRPC-Go servers configured with xds.NewGRPCServer() where a crafted request missing both :authority and Host headers can cause a server panic, resulting in a Denial of Service (DoS). Servers built with xds.NewGRPCServer install an xDS routing interceptor on every RPC. This interceptor looks up the request’s :authority header to pick a virtual host. The HTTP/2 server transport previously accepted requests that had neither :authority nor Host. When …

TagReference.create() forwards a caller-influenced positional reference value into git tag without it ever being inspected by the unsafe-option guard, allowing an arbitrary file read (the file's contents are returned in-band as the annotated tag message). This is an incomplete-fix bypass of commit 3af0c251 (the fix for GHSA-3f7w-8rr8-f37f's tag instance).

Repo.blame() / Repo.blame_incremental() guard forwarded revision options against unsafe_git_revision_options, but that denylist only contains the file-WRITE options –output/-o. git blame also honors –contents <file> and -S <file>, which cause the file's lines to be echoed into the blame result — an arbitrary file READ. Neither option is in the denylist, so a caller-influenced revision value of –contents=<path> passes the guard and leaks file contents. This is a distinct sink-option and …

CWE: CWE-88 (Argument Injection) / CWE-94 (Code Injection) — via a read-then-corrupt-on-rewrite config round trip, not a direct setter argument Affected component: git/config.py — GitConfigParser._read() (multi-line value decoding, lines 444-541, esp. string_decode() at line 460 and its call sites at 519/541) and GitConfigParser._write()/write_section() (serialization, lines ~694-712, esp. line 708) Affected version: GitPython at HEAD (9729ed3b948f2bde09f1f188c5311e172212b67e, 2026-08-05, VERSION 3.1.58)

CWE: CWE-73 (External Control of File Name or Path) / CWE-22 (Path Traversal, in the "escapes intended base directory" sense) Affected component: git/repo/base.py, Repo.unsafe_git_clone_options (class attribute, lines 153-165) and Repo._clone() (lines 1477-1520), reached via the public Repo.clone_from() (line 1626) and Repo.clone() (line 1567) APIs. Affected version: GitPython at HEAD (9729ed3b948f2bde09f1f188c5311e172212b67e, 2026-08-05, VERSION 3.1.58)

Non-blind local file content disclosure (first line) of any file readable by the victim process, triggered purely by attacker-controlled repository content and one routine, read-only GitPython call. Bounded to one line per triggering file (parsing aborts at the first MissingSectionHeaderError), but that line very often is the secret — .env files (DATABASE_URL=…, API_KEY=…), single-line credential/token files, /etc/passwd's root entry for host fingerprinting. The primitive additionally serves as a generic error-based …

Gitea's diffpatch endpoint can be abused to install and execute a Git hook from repository-controlled content. An attacker with ordinary write access to a repository can execute arbitrary shell commands as the Gitea OS user. With default open registration, an unauthenticated visitor can obtain the required write access by registering an account and creating a repository.

The isBlockedIP SSRF guard in Dozzle's webhook notification dispatcher blocks loopback, link-local, multicast, and unspecified addresses but does not recognize IPv6 transition mechanism addresses (RFC 3056 6to4, RFC 6052 NAT64, RFC 4380 Teredo) that embed arbitrary IPv4 addresses. An authenticated user can bypass the guard to reach loopback services, cloud metadata endpoints (169.254.169.254), and other blocked ranges via webhook notification URLs.

If the p4 Perforce CLI client is installed, a malicious dependency package from a package repository allowing arbitrary perforce source URLs (packagist.org is safe) could execute arbitrary commands when running composer install or composer update. Composer passed a package's Perforce source address to the p4 CLI client without validating it. The p4 CLI client accepts addresses that mean "run this local command" instead of "connect to this server". This is …

colord's CSS color string matchers described a number as ([+-]?\d*.?\d+). In that form \d* and \d+ can match the same digits, so a run of n digits can be divided between them in O(n²) ways, and rejecting an input retries every division. Parsing is synchronous and uninterruptible, so a long malformed color string blocks the thread: | input | time to reject | | — | — | | 16 …

Custom mail headers added with Message::setHeaders() or addHeaders() do not have CRLF replaced allowing header injection if user controlled data is added to message headers without stripping CRLF bytes.

The FunctionsBuilder::jsonValue($field, $jsonPath) methods with the Postgres driver is vulnerable to SQL injection if user controlled data is supplied to the $jsonPath parameter.

The FunctionsBuilder::jsonValue($field, $jsonPath) methods with the Postgres driver is vulnerable to SQL injection if user controlled data is supplied to the $jsonPath parameter.

A vulnerability in libheif, used by the default Sharp image service in Astro, can lead to remote code execution when a malicious AVIF image is optimized. Projects are affected when an attacker can cause Astro to process an untrusted AVIF image. The fix was released in Astro 7.2.8, which requires Sharp 0.35.4.

Astro stripped a configured base path from request pathnames using a string-prefix check that did not verify a path-segment boundary. With base: "/app", a request to /appX/admin was treated as being under the base and resolved internally to the /admin route, while middleware still observed the public pathname /appX/admin. Middleware that authorizes routes by inspecting context.url.pathname could therefore be bypassed.

@sap/cds-mtxs NPM library does not perform sufficient checks on certain functionality used in multitenant CAP applications with extensibility enabled. An unauthenticated attacker could send specially crafted requests to obtain sensitive credentials and abuse them to replace or delete tenant data. Successful exploitation can result in a high impact on availability and integrity of the application. There may also be partial impact to the confidentiality of business data.

NLTK's Text.findall() and TokenSearcher.findall() methods accept user-supplied regular expressions and pass them to the Python re engine without timeout or validation, enabling catastrophic backtracking (ReDoS). This issue is isolated to the nltk.text module and was resolved in a prior commit.

When the vLLM API receives a malformed request (e.g., invalid JSON or missing required fields), FastAPI raises a Pydantic RequestValidationError. The validation_exception_handler in vllm/entrypoints/openai/server_utils.py converts this exception to a string via str(exc), which includes the internal file path and line number of the handler function. The existing sanitize_message() function in vllm/entrypoints/utils.py strips memory addresses (e.g., 0x7f…) but does not strip File "…", line X patterns. The result is a user-facing …

The fix for GHSA-rwxx-mrjm-wc2m ("ReDoS via structured_outputs.regex compiled without timeout") wrapped the regex compile in the xgrammar and outlines backends with compile_regex_with_timeout (and, for outlines, validate_regex_is_buildable). The lm-format-enforcer backend was left unguarded: it compiles the attacker-supplied regex with no timeout and no buildability check. A single request with a catastrophic regex hangs the structured-output compile step and stalls the engine worker (denial of service).

The /v1/completions/derender and /v1/chat/completions/derender endpoints accept caller-supplied GenerateResponse objects and postprocess every nested choices[*].token_ids list directly. Unlike the normal render/generate path, derender does not enforce model context length, resolved max_tokens, max_num_seqs, choice-count, or response-size bounds before detokenizing and returning the supplied token IDs. An authenticated API client can therefore make the CPU-only render frontend, or any server exposing these /v1 derender routes, spend CPU and memory proportional to attacker-chosen generated-output-shaped …

@typespec/spector registers a POST /.admin/stop HTTP route with no authentication, authorization token, Origin check, or IP-source restriction. Any network-reachable client can send a single unauthenticated POST request to terminate the mock server process. Because the server binds to 0.0.0.0 by default (all interfaces), this endpoint is exposed to any host that can reach the server's port—not just localhost—making a complete denial-of-service trivially achievable with one HTTP request. Severity is High …

A PERMISSIONS … WHERE clause is evaluated with permission enforcement disabled, so it can't recurse into its own checks. But the clause could also contain data-modifying statements, and these ran with enforcement still off — so evaluating a permission check could write to tables the caller cannot write. For example: DEFINE TABLE post PERMISSIONS FOR update WHERE (CREATE log SET at = time::now()) OR true; Any user allowed to update …

An authenticated user scoped to one namespace/database could invoke a custom API (DEFINE API) belonging to a different namespace/database, reaching another tenant's endpoint. The route /api/{namespace}/{database}/{endpoint} took the namespace and database from the URL and applied them to the caller's session before the endpoint was looked up or run, without checking that the caller's authenticated scope covered them. Because a custom API handler runs with permissions disabled (definer's rights), the …

SurrealDB offers http functions that can access external network endpoints. A typical, albeit not recommended configuration would be to start SurrealDB with all network connections allowed with the exception of a deny list. For example, surreal start –allow-net –deny-net 10.0.0.0/8 will allow all network connections except to the 10.0.0.0/8 block. An authenticated user of SurrealDB can use bypass this restriction, using http::<fn>(<url>) functions where the hostname resolves to an IP …

/api/system/getConf returns Conf.CookieKey, the key used to sign the server's session cookies in its response body. The endpoint is registered with CheckAuth only, so the field reaches the publish RoleReader token and the anonymous account when Publish.Auth.Enable is false. The configuration-export endpoint in the same file strips this exact field before returning config, so the project already treats it as secret. The reader-facing masking path does not.

/api/tag/getTag filters its results for reader roles through FilterTagsByPublishIgnore, which checks only the visible publish tier. Documents that are published but password-protected pass that check, so a reader who has never entered a document's publish password receives every tag label used inside it, together with occurrence counts. The project has already treated this exact tier mismatch as a vulnerability on a sibling path: commit 82e9ded42 ("Enforce publish access for graph …

Several static-file routes in the server mux (kernel/server/serve.go) are registered with CheckAuth only and serve directories directly, without the publish-access checks, sensitive-path blocklist, or refuseToAccess rules that the REST API applies to the same data. They are therefore reachable by the publish RoleReader token and by the anonymous account when Publish.Auth.Enable is false. Most notably, /templates/ serves data/templates a directory the REST file API explicitly refuses to serve to non-administrators.

renderAttributeView correctly applies the reader publish-access filter, but the filter's row-accessibility decision is keyed solely to the row's first cell, and it never inspects the remaining cells' values. Relation and Rollup cells carry mirrored content from a different database, so a row belonging to a published database can hand an anonymous reader the contents of a related database whose host document is hidden, publish-forbidden, or password-protected. Separately, when the first …

/api/system/getConf is registered with CheckAuth only and is reachable by the publish RoleReader token, and anonymously when Publish.Auth.Enable is false. Its non-administrator masking chain is a blocklist that enumerates fields individually. Three fields that the configuration-export endpoint in the same file deliberately clears are absent from that blocklist and are returned to readers: | Field | JSON | What it is | Cleared by exportConf at | |—|—|—|—| | Conf.CookieKey …

Five filetree endpoints resolve arbitrary document IDs and paths with no publish-access check of any kind. All are CheckAuth-only, so they are reachable by the publish RoleReader token and by the anonymous account when Publish.Auth.Enable is false. An anonymous reader can map the complete private document tree every notebook, folder, and document title, and which notebook holds each document for documents marked hidden, password-protected, or publish-forbidden, and can resolve titles …

POST /api/notebook/getEncryptedNotebookStatus returns the identifier, name, and current lock state of every encrypted notebook, with no publish-access filtering. The route is registered CheckAuth only, no CheckReadonly, no CheckAdminRole so it is reachable by the publish RoleReader token and by the anonymous account when Publish.Auth.Enable is false. Encrypted notebooks are private by design; their names frequently reveal the sensitive topic that motivated encrypting them.

/api/block/getBlockDOMWithEmbed and /api/block/getBlockDOMsWithEmbed gate only the requested block against publish access. The blocks pulled in by that block's embed (transclusion) query are inlined into the returned DOM with no publish-access check at all, so a reader who requests a legitimately-published block containing an embed query receives the content of every block that query matched including blocks in hidden, forbidden and password-protected documents. The dedicated embed endpoint getEmbedBlock filters these same …

validateCertificatePath() does not verify that an attestation's certificate chain actually terminates at a configured trust anchor. When walking the chain it stops at the first self-signed certificate it finds (which could be user-supplied), and exits early. This happens before the configured Apple/Google/etc trust anchor (which is concatenated to the end of the chain) is reached. A user can therefore register a credential and have the server accept it as if …

The OpenChoreo control-plane cluster-gateway exposes internal management APIs (/api/proxy/, /api/exec/, /api/wirelogs/) that tunnel requests through to connected data planes' Kubernetes APIs, but the internal listener authenticates no caller. Its request validator permits mutating HTTP methods and reads of Secrets in tenant namespaces (only kube-system Secrets are blocked), so although the client library documents these requests as "read-only," the server enforces no such restriction. Any party able to reach the internal …

DNS-pinning failure allows natural failure of code, however with a custom DNS server that fails the initial requests and allows the secondary requests, it's possible to bypass the logic.

DNS-pinning failure allows natural failure of code, however with a custom DNS server that fails the initial requests and allows the secondary requests, it's possible to bypass the logic.

DNS-pinning failure allows natural failure of code, however with a custom DNS server that fails the initial requests and allows the secondary requests, it's possible to bypass the logic.

DNS-pinning failure allows natural failure of code, however with a custom DNS server that fails the initial requests and allows the secondary requests, it's possible to bypass the logic.

The rlm_eval tool runs an arbitrary Python string chosen by the model in a real python3 interpreter. Its approval_requirement() returns ApprovalRequirement::Auto, which the engine treats as "never prompt," regardless of the user's configured –approval-policy. A single tool call — which prompt injection from any untrusted content the agent reads (a web page, a fetched URL, a repo file, an MCP tool result) can induce — runs code on the user's …

The rlm_eval tool runs an arbitrary Python string chosen by the model in a real python3 interpreter. Its approval_requirement() returns ApprovalRequirement::Auto, which the engine treats as "never prompt," regardless of the user's configured –approval-policy. A single tool call — which prompt injection from any untrusted content the agent reads (a web page, a fetched URL, a repo file, an MCP tool result) can induce — runs code on the user's …

The rlm_eval tool runs an arbitrary Python string chosen by the model in a real python3 interpreter. Its approval_requirement() returns ApprovalRequirement::Auto, which the engine treats as "never prompt," regardless of the user's configured –approval-policy. A single tool call — which prompt injection from any untrusted content the agent reads (a web page, a fetched URL, a repo file, an MCP tool result) can induce — runs code on the user's …

The rlm_eval tool runs an arbitrary Python string chosen by the model in a real python3 interpreter. Its approval_requirement() returns ApprovalRequirement::Auto, which the engine treats as "never prompt," regardless of the user's configured –approval-policy. A single tool call — which prompt injection from any untrusted content the agent reads (a web page, a fetched URL, a repo file, an MCP tool result) can induce — runs code on the user's …

A malicious .codewhale/config.toml or .deepseek/config.toml committed to a repository can set instructions to an array of arbitrary file paths (including paths outside the workspace like ~/.ssh/id_rsa or ~/.aws/credentials) that are read from disk and injected into the AI model's system prompt. There is no path validation, workspace boundary check, or tightening guard on the instructions field. This enables a malicious repository to exfiltrate the contents of sensitive files on the …

A malicious .codewhale/config.toml or .deepseek/config.toml committed to a repository can set instructions to an array of arbitrary file paths (including paths outside the workspace like ~/.ssh/id_rsa or ~/.aws/credentials) that are read from disk and injected into the AI model's system prompt. There is no path validation, workspace boundary check, or tightening guard on the instructions field. This enables a malicious repository to exfiltrate the contents of sensitive files on the …

A malicious .codewhale/config.toml or .deepseek/config.toml committed to a repository can set instructions to an array of arbitrary file paths (including paths outside the workspace like ~/.ssh/id_rsa or ~/.aws/credentials) that are read from disk and injected into the AI model's system prompt. There is no path validation, workspace boundary check, or tightening guard on the instructions field. This enables a malicious repository to exfiltrate the contents of sensitive files on the …

A malicious .codewhale/config.toml or .deepseek/config.toml committed to a repository can set instructions to an array of arbitrary file paths (including paths outside the workspace like ~/.ssh/id_rsa or ~/.aws/credentials) that are read from disk and injected into the AI model's system prompt. There is no path validation, workspace boundary check, or tightening guard on the instructions field. This enables a malicious repository to exfiltrate the contents of sensitive files on the …

A malicious .codewhale/config.toml or .deepseek/config.toml committed to a repository can silently set allow_shell = true for any user who clones and opens the repository in CodeWhale. This enables the AI model's exec_shell tool, granting arbitrary shell command execution on the victim's machine without the user's explicit opt-in. The approval_policy and sandbox_mode fields correctly enforce tightening-only semantics from project config, but allow_shell has no such guard, contradicting the intent of GHSA-72w5-pf8h-xfp4 …

A malicious .codewhale/config.toml or .deepseek/config.toml committed to a repository can silently set allow_shell = true for any user who clones and opens the repository in CodeWhale. This enables the AI model's exec_shell tool, granting arbitrary shell command execution on the victim's machine without the user's explicit opt-in. The approval_policy and sandbox_mode fields correctly enforce tightening-only semantics from project config, but allow_shell has no such guard, contradicting the intent of GHSA-72w5-pf8h-xfp4 …

A malicious .codewhale/config.toml or .deepseek/config.toml committed to a repository can silently set allow_shell = true for any user who clones and opens the repository in CodeWhale. This enables the AI model's exec_shell tool, granting arbitrary shell command execution on the victim's machine without the user's explicit opt-in. The approval_policy and sandbox_mode fields correctly enforce tightening-only semantics from project config, but allow_shell has no such guard, contradicting the intent of GHSA-72w5-pf8h-xfp4 …

A malicious .codewhale/config.toml or .deepseek/config.toml committed to a repository can silently set allow_shell = true for any user who clones and opens the repository in CodeWhale. This enables the AI model's exec_shell tool, granting arbitrary shell command execution on the victim's machine without the user's explicit opt-in. The approval_policy and sandbox_mode fields correctly enforce tightening-only semantics from project config, but allow_shell has no such guard, contradicting the intent of GHSA-72w5-pf8h-xfp4 …

js_execution exposes parent process environment to model-provided JavaScript The js_execution tool spawns Node with tokio::process::Command::new without calling the child_env scrubber that exec_shell, the Python REPL, and the MCP launcher all use. Model-provided JavaScript reads process.env and the values flow back to the parent transcript as the tool's stdout, exposing API keys, cloud credentials, and forge tokens to the next model turn.

js_execution exposes parent process environment to model-provided JavaScript The js_execution tool spawns Node with tokio::process::Command::new without calling the child_env scrubber that exec_shell, the Python REPL, and the MCP launcher all use. Model-provided JavaScript reads process.env and the values flow back to the parent transcript as the tool's stdout, exposing API keys, cloud credentials, and forge tokens to the next model turn.

js_execution exposes parent process environment to model-provided JavaScript The js_execution tool spawns Node with tokio::process::Command::new without calling the child_env scrubber that exec_shell, the Python REPL, and the MCP launcher all use. Model-provided JavaScript reads process.env and the values flow back to the parent transcript as the tool's stdout, exposing API keys, cloud credentials, and forge tokens to the next model turn.

js_execution exposes parent process environment to model-provided JavaScript The js_execution tool spawns Node with tokio::process::Command::new without calling the child_env scrubber that exec_shell, the Python REPL, and the MCP launcher all use. Model-provided JavaScript reads process.env and the values flow back to the parent transcript as the tool's stdout, exposing API keys, cloud credentials, and forge tokens to the next model turn.

image_analyze follows workspace symlinks and leaks outside-workspace file bytes to the vision endpoint The image_analyze tool resolves its image_path with a bare context.workspace.join instead of routing through ToolContext::resolve_path. The pre-join lexical check rejects absolute paths, Windows prefixes, and parent-dir components but never canonicalizes, so a symlink inside the workspace whose name ends in an image extension and whose target sits outside the workspace is read transparently. The tool has ReadOnly …

image_analyze follows workspace symlinks and leaks outside-workspace file bytes to the vision endpoint The image_analyze tool resolves its image_path with a bare context.workspace.join instead of routing through ToolContext::resolve_path. The pre-join lexical check rejects absolute paths, Windows prefixes, and parent-dir components but never canonicalizes, so a symlink inside the workspace whose name ends in an image extension and whose target sits outside the workspace is read transparently. The tool has ReadOnly …

image_analyze follows workspace symlinks and leaks outside-workspace file bytes to the vision endpoint The image_analyze tool resolves its image_path with a bare context.workspace.join instead of routing through ToolContext::resolve_path. The pre-join lexical check rejects absolute paths, Windows prefixes, and parent-dir components but never canonicalizes, so a symlink inside the workspace whose name ends in an image extension and whose target sits outside the workspace is read transparently. The tool has ReadOnly …