@clear-capabilities/agentic-security-scanner 0.134.0 → 0.136.9

package/dist/compliance-frameworks/owasp-asvs-5.json

@@ -0,0 +1,79 @@
+{
+  "id": "owasp-asvs-5",
+  "name": "OWASP Application Security Verification Standard 5.0",
+  "publisher": "OWASP Foundation",
+  "license": "Creative Commons Attribution-ShareAlike 4.0",
+  "url": "https://owasp.org/www-project-application-security-verification-standard/",
+  "controls": [
+    {
+      "id": "V2.1",
+      "category": "Authentication",
+      "summary": "Verify that authentication is performed for protected functions.",
+      "evidence": ["Zero open critical findings in family auth-missing on the current scan."],
+      "mapsTo": ["family:auth-missing"]
+    },
+    {
+      "id": "V2.7",
+      "category": "Authentication",
+      "summary": "Verify that MFA / secondary authentication is used for high-value operations.",
+      "evidence": ["aws-no-mfa-condition zero findings (if AWS in scope)."],
+      "mapsTo": ["family:aws-no-mfa"]
+    },
+    {
+      "id": "V4.1",
+      "category": "Access Control",
+      "summary": "Verify that authorization is enforced for every resource access.",
+      "evidence": ["Zero idor / authz findings."],
+      "mapsTo": ["family:idor", "family:authz"]
+    },
+    {
+      "id": "V5.1",
+      "category": "Input Validation",
+      "summary": "Verify input validation for type, length, and content.",
+      "evidence": ["Zero open critical findings in sqli/xss/command-injection/ldap-injection/xpath-injection."],
+      "mapsTo": ["family:sqli", "family:xss", "family:command-injection", "family:ldap-injection", "family:xpath-injection", "family:nosql-injection"]
+    },
+    {
+      "id": "V6.1",
+      "category": "Cryptography",
+      "summary": "Verify use of strong cryptography per a documented policy.",
+      "evidence": ["Zero open findings in crypto-weak-cipher / crypto-weak-hash / crypto-ecb / crypto-static-iv."],
+      "mapsTo": ["family:crypto-weak-cipher", "family:crypto-weak-hash", "family:crypto-ecb", "family:crypto-static-iv", "family:crypto-kdf-weak"]
+    },
+    {
+      "id": "V6.2",
+      "category": "Cryptography",
+      "summary": "Verify all transit data uses TLS ≥ 1.2.",
+      "evidence": ["Zero findings in crypto-tls-version / crypto-tls-no-verify."],
+      "mapsTo": ["family:crypto-tls-version", "family:crypto-tls-no-verify"]
+    },
+    {
+      "id": "V7.1",
+      "category": "Error Handling & Logging",
+      "summary": "Verify that error handling does not disclose sensitive information.",
+      "evidence": ["No findings in family data-exposure."],
+      "mapsTo": ["family:data-exposure"]
+    },
+    {
+      "id": "V8.1",
+      "category": "Data Protection",
+      "summary": "Verify that personal data is protected at rest and in transit.",
+      "evidence": ["DPIA artifact at .agentic-security/dpia.md.", "Zero pii-exposure findings."],
+      "mapsTo": ["family:pii-exposure", "family:training-data-pii"]
+    },
+    {
+      "id": "V10.1",
+      "category": "Malicious Code",
+      "summary": "Verify the application does not include known-malicious or compromised dependencies.",
+      "evidence": [".agentic-security/sbom-history snapshots clean.", "No dependency-confusion or dependency-drift findings."],
+      "mapsTo": ["family:vulnerable-dep", "family:dependency-confusion", "family:dependency-drift"]
+    },
+    {
+      "id": "V14.1",
+      "category": "Configuration",
+      "summary": "Verify that build, deployment, and configuration are secure.",
+      "evidence": ["No findings in family iam-overpermissive / k8s-rbac-cluster-admin / k8s-pod-security-privileged."],
+      "mapsTo": ["family:iam-overpermissive", "family:k8s-rbac-cluster-admin", "family:k8s-pod-security-privileged"]
+    }
+  ]
+}

package/dist/compliance-frameworks/owasp-llm-top-10.json

@@ -0,0 +1,69 @@
+{
+  "id": "owasp-llm-top-10",
+  "name": "OWASP Top 10 for LLM Applications 2025",
+  "publisher": "OWASP Foundation",
+  "license": "Creative Commons Attribution-ShareAlike 4.0",
+  "url": "https://owasp.org/www-project-top-10-for-large-language-model-applications/",
+  "controls": [
+    {
+      "id": "LLM01",
+      "summary": "Prompt Injection — controls preventing injected instructions from overriding the system prompt.",
+      "evidence": ["Zero open findings in family prompt-injection / llm-app-security.", "Prompt-template integrity check passing."],
+      "mapsTo": ["family:prompt-injection", "family:llm-app-security", "family:prompt-integrity"]
+    },
+    {
+      "id": "LLM02",
+      "summary": "Insecure Output Handling — LLM output is not directly piped to dangerous sinks (eval/shell/SQL/HTML).",
+      "evidence": ["Zero open findings in subfamily llm-output-untrusted-sink."],
+      "mapsTo": ["family:llm-app-security:llm-output-untrusted-sink"]
+    },
+    {
+      "id": "LLM03",
+      "summary": "Training Data Poisoning — provenance and integrity controls on training/fine-tuning data.",
+      "evidence": ["No streaming-dataset-url findings.", "datasets.load_dataset uses pinned revisions."],
+      "mapsTo": ["family:streaming-dataset-url", "family:hf-datasets-rce"]
+    },
+    {
+      "id": "LLM04",
+      "summary": "Model Denial of Service — rate-limit / token-cap controls.",
+      "evidence": ["LLM API calls in source carry an explicit max_tokens cap (bodyguard rule no-max-tokens)."],
+      "mapsTo": ["rule:no-max-tokens"]
+    },
+    {
+      "id": "LLM05",
+      "summary": "Supply Chain — model & dataset supply-chain integrity.",
+      "evidence": ["No mlflow-untrusted-uri / model-format / hf-endpoint-override findings.", "Sigstore provenance verification configured if opt-in."],
+      "mapsTo": ["family:mlflow-untrusted-uri", "family:model-format", "family:hf-endpoint-override"]
+    },
+    {
+      "id": "LLM06",
+      "summary": "Sensitive Information Disclosure — output filters prevent leakage.",
+      "evidence": ["Zero findings in subfamily llm-credential-in-prompt.", "Redact module covers prompt + response."],
+      "mapsTo": ["family:llm-app-security:llm-credential-in-prompt"]
+    },
+    {
+      "id": "LLM07",
+      "summary": "Insecure Plugin / Tool Design — tools the LLM can invoke are narrowly scoped.",
+      "evidence": ["Zero llm-app-security:llm-tool-exec findings.", "Agent tool definitions reviewed for excessive agency."],
+      "mapsTo": ["family:agent-tool-exec", "family:llm-app-security:llm-tool-exec"]
+    },
+    {
+      "id": "LLM08",
+      "summary": "Excessive Agency — agent autonomy is bounded by explicit guards.",
+      "evidence": ["MCP server config limits write-tools to confirm:true + reserved-write-path refusal.", "Apply-fix path requires HMAC-verified scan."],
+      "mapsTo": ["module:mcp-tools", "module:apply-fix"]
+    },
+    {
+      "id": "LLM09",
+      "summary": "Overreliance — human review checkpoints exist.",
+      "evidence": ["security-fixer agent has Stop-on-verify-fail loop.", "PreToolUse bodyguard exists."],
+      "mapsTo": ["module:security-fixer", "module:pre-edit-bodyguard"]
+    },
+    {
+      "id": "LLM10",
+      "summary": "Model Theft — model access controls.",
+      "evidence": ["No private-key-in-frontend or rpc-key-inline findings in client code."],
+      "mapsTo": ["family:private-key-in-frontend", "family:rpc-key-inline"]
+    }
+  ]
+}

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@clear-capabilities/agentic-security-scanner",
-  "version": "0.134.0",
-  "description": "Scanner engine for the agentic-security Claude Code plugin — SAST, SCA (function-level reachability + CISA KEV), secrets, IaC, prompt-injection, MCP/agent-tool audit, auth/authZ deep analysis, attack chains, PoC generation, business logic, toxic-combinations scoring, SBOM, SARIF ingest, pipeline integrity, compliance attestation, and more.",
+  "version": "0.136.9",
+  "description": "Scanner engine for the agentic-security Claude Code plugin \u2014 SAST, SCA (function-level reachability + CISA KEV), secrets, IaC, prompt-injection, MCP/agent-tool audit, auth/authZ deep analysis, attack chains, PoC generation, business logic, toxic-combinations scoring, SBOM, pipeline integrity, compliance attestation, and more.",
   "type": "module",
   "main": "src/index.js",
   "bin": {
@@ -21,7 +21,8 @@
     "./sast": "./src/sast/index.js",
     "./sca": "./src/sca/index.js",
     "./secrets": "./src/secrets/index.js",
-    "./report": "./src/report/index.js"
+    "./report": "./src/report/index.js",
+    "./posture/fix-style-mirror.js": "./src/posture/fix-style-mirror.js"
   },
   "files": [
     "src/",
@@ -55,19 +56,20 @@
     "lodash-es": "^4.18.1"
   },
   "scripts": {
-    "build": "ncc build bin/agentic-security.js -o dist --minify -e web-tree-sitter -e tree-sitter-wasms && mv dist/index.js dist/agentic-security.mjs && rm -f dist/package.json && chmod +x dist/agentic-security.mjs && node -e \"const fs=require('fs');const p='dist/agentic-security.mjs';const c=fs.readFileSync(p,'utf8');if(!c.startsWith('#!'))fs.writeFileSync(p,'#!/usr/bin/env node\\n'+c);\" && node -e \"const fs=require('fs');const c=require('crypto');const h=c.createHash('sha256').update(fs.readFileSync('dist/agentic-security.mjs')).digest('hex');fs.writeFileSync('dist/agentic-security.mjs.sha256',h+'  agentic-security.mjs\\n');\"",
+    "build": "ncc build bin/agentic-security.js -o dist --minify -e web-tree-sitter -e tree-sitter-wasms && rm -rf dist/compliance-frameworks && mkdir -p dist/compliance-frameworks && cp src/posture/compliance-frameworks/*.json dist/compliance-frameworks/ && mv dist/index.js dist/agentic-security.mjs && rm -f dist/package.json && chmod +x dist/agentic-security.mjs && node -e \"const fs=require('fs');const p='dist/agentic-security.mjs';const c=fs.readFileSync(p,'utf8');if(!c.startsWith('#!'))fs.writeFileSync(p,'#!/usr/bin/env node\\n'+c);\" && node -e \"const fs=require('fs');const c=require('crypto');const h=c.createHash('sha256').update(fs.readFileSync('dist/agentic-security.mjs')).digest('hex');fs.writeFileSync('dist/agentic-security.mjs.sha256',h+'  agentic-security.mjs\\n');\"",
     "prepare": "node ../scripts/pre-push-gate.mjs --install-hook",
     "prepublishOnly": "npm run build && node ../scripts/sync-scanner-changelog.mjs && node ../scripts/release-check.mjs",
-    "test": "npm run test:smoke && npm run test:glob && npm run test:sast && npm run test:posture && npm run test:dataflow && npm run test:mcp && npm run test:report && npm run test:bench-modules && npm run test:lifecycle && npm run test:eval && npm run test:discovery && AGENTIC_SECURITY_CPP_DATAFLOW=1 node --test test/cpp-dataflow.test.js",
+    "test": "npm run test:smoke && npm run test:glob && npm run test:sast && npm run test:posture && npm run test:dataflow && npm run test:mcp && npm run test:report && npm run test:bench-modules && npm run test:lifecycle && npm run test:eval && npm run test:discovery && AGENTIC_SECURITY_CPP_DATAFLOW=1 node --test test/cpp-dataflow.test.js && npm run test:python",
+    "test:python": "node ../scripts/run-python-tests.mjs",
     "test:smoke": "node --test test/smoke.test.js",
     "test:glob": "node --test test/glob-compat.test.js",
-    "test:sast": "node --test test/crypto-specialist.test.js test/llm.test.js test/llm-cost-advisor.test.js test/llm-owasp.test.js test/logic.test.js test/authz.test.js test/model-load.test.js test/prompt-template.test.js test/business-logic.test.js test/python-sinks.test.js test/phase1-detectors.test.js test/phase2-detectors.test.js test/phase3-v3.test.js test/phase7-extensions.test.js test/phase8-extensions.test.js test/new-cwe-detectors.test.js test/file-upload.test.js test/llmsecops-detectors.test.js test/db-taint.test.js test/dart-swift.test.js test/redos-nfa.test.js test/weak-randomness.test.js test/csharp-pipeline.test.js test/post-quantum-crypto.test.js test/web3-advanced.test.js test/cloud-iam-k8s.test.js test/crypto-protocol.test.js test/ml-supply-chain.test.js test/wrong-context-sanitizer.test.js test/sanitizer-context.test.js test/frontend-hygiene.test.js test/csv-injection.test.js test/stored-taint.test.js test/tree-sitter-sinks.test.js test/kotlin-structural.test.js test/ruby-php-structural.test.js test/java-csharp-structural.test.js test/guard-recognition.test.js test/js-python-framework-structural.test.js test/go-structural.test.js test/secret-concat.test.js test/xss-reflected-multilang.test.js test/code-injection-multilang.test.js test/xxe-multilang.test.js test/xpath-injection-multilang.test.js test/gapfill-batch13.test.js test/agent-untrusted-flow.test.js test/api-authz.test.js test/event-entrypoint.test.js test/iac-terraform.test.js test/cross-service.test.js test/rbac-consistency.test.js",
-    "test:posture": "node --test test/material-change.test.js test/drift.test.js test/scorecard.test.js test/accuracy-scorecard.test.js test/scorecard-gate.test.js test/release-check.test.js test/pre-push-gate.test.js test/dependency-currency.test.js test/package-contents-check.test.js test/gate-verdict-cache.test.js test/mttr.test.js test/license-policy.test.js test/aibom.test.js test/sbom.test.js test/api-inventory.test.js test/iam-policy.test.js test/container.test.js test/container-runtime.test.js test/image-packages.test.js test/kev.test.js test/dep-confusion.test.js test/sca-deprecated.test.js test/sca-batch.test.js test/composite-risk.test.js test/sca-coverage.test.js test/gradle-deps.test.js test/sca-route-reachable.test.js test/sca-policy.test.js test/sca-verdict.test.js test/install-script.test.js test/sca-linked-findings.test.js test/packs.test.js test/flow-narration.test.js test/regression-test-gen.test.js test/deterministic-fix.test.js test/falsification.test.js test/verification-separation.test.js test/logic-claims.test.js test/attestation.test.js test/comparison.test.js test/determinism-cross-machine.test.js test/fix-honesty-gate.test.js test/model-routing.test.js test/root-cause-sweep.test.js test/entrypoint-inventory.test.js test/relevance.test.js test/untrusted.test.js test/agent-hardening.test.js test/rule-synthesis.test.js test/policy-gate.test.js test/agents-memory.test.js test/cve-lookup.test.js test/cve-alert-daemon.test.js test/fix-verify-loop.test.js test/fix-verify-tests.test.js test/fix-acceptance.test.js test/exploitability-probability.test.js test/history-scan.test.js test/viral-features.test.js test/viral-v074.test.js test/state-dir.test.js test/ignore-pragma.test.js test/license-graph.test.js test/secret-live-check.test.js test/attack-taxonomy.test.js test/triage-memory.test.js test/pr-augment.test.js test/chat-batch2.test.js test/chat-batch3.test.js test/chat-batch4.test.js test/chat-batch5.test.js test/chat-batch6.test.js test/router.test.js test/legacy-alias-redirect.test.js test/cache-economics.test.js test/coverage-report.test.js test/corpus-status.test.js test/provenance.test.js test/secret-history.test.js test/execution-proof.test.js test/fix-metrics.test.js test/corpus-enroll.test.js test/prove-findings.test.js test/corpus-match.test.js test/corpus-provenance.test.js test/learning-quorum.test.js test/llm-cache-integrity.test.js test/proof-artifact.test.js test/providers.test.js test/fleet.test.js test/autopilot.test.js test/autopilot-cli.test.js test/verifier-independence.test.js test/integrity-legacy-key.test.js test/suppression-visibility.test.js test/cost-ceiling.test.js test/local-endpoint.test.js test/model-trust.test.js test/vuln-archaeology.test.js test/scan-checkpoint.test.js test/llm-redact.test.js",
-    "test:dataflow": "node --test test/fn-reach.test.js test/deep-taint.test.js test/calibration.test.js test/holdout-eval.test.js test/cross-lang-meta.test.js test/cross-lang-queues.test.js test/phase5-xlang.test.js test/phase5-coverage.test.js test/phase6-taint.test.js test/llm-validator-consistency.test.js test/llm-validator-default-on.test.js test/llm-validator-preset.test.js test/parser-py-cst.test.js test/parser-cs-kt.test.js test/parser-go.test.js test/parser-php-rb.test.js test/interproc-k2.test.js test/proven-clean.test.js test/backward-default.test.js test/incremental-cache.test.js test/string-regex-lattice.test.js test/closure-capture.test.js test/points-to.test.js test/type-stubs.test.js test/soft-taint.test.js test/ifds.test.js test/symbolic-exec-proof.test.js test/ifds-summary-edges.test.js test/stub-aware-filter.test.js test/cross-repo.test.js test/proof-gate.test.js test/proof-safe.test.js test/collection-taint.test.js test/kcfa-context.test.js test/kcfa-callstring.test.js test/flow-parity.test.js test/callgraph-resolve.test.js test/import-reachability.test.js test/ir-stats.test.js test/parser-cpp.test.js test/parser-js-decorators.test.js test/cpp-integration.test.js test/engine-reconnect.test.js test/phase2-scoping.test.js test/engine-recall.test.js",
-    "test:mcp": "node --test test/mcp.test.js test/mcp-audit.test.js test/audit-cli.test.js test/mcp-scratchpad.test.js test/mcp-offload.test.js test/sca-upgrade.test.js",
-    "test:report": "node --test test/sarif-ingest.test.js test/junit.test.js test/ci.test.js test/poc-generator.test.js test/verifier.test.js test/verifier-target.test.js test/annotator-errors.test.js test/grader-calibration.test.js test/pr-delta-gate.test.js test/vex.test.js test/report-render.test.js",
+    "test:sast": "node --test test/crypto-specialist.test.js test/llm.test.js test/llm-cost-advisor.test.js test/llm-owasp.test.js test/logic.test.js test/authz.test.js test/model-load.test.js test/prompt-template.test.js test/business-logic.test.js test/python-sinks.test.js test/phase1-detectors.test.js test/phase2-detectors.test.js test/phase3-v3.test.js test/phase7-extensions.test.js test/phase8-extensions.test.js test/new-cwe-detectors.test.js test/file-upload.test.js test/llmsecops-detectors.test.js test/db-taint.test.js test/dart-swift.test.js test/redos-nfa.test.js test/weak-randomness.test.js test/csharp-pipeline.test.js test/post-quantum-crypto.test.js test/web3-advanced.test.js test/cloud-iam-k8s.test.js test/crypto-protocol.test.js test/ml-supply-chain.test.js test/mcp-audit-scan.test.js test/wrong-context-sanitizer.test.js test/sanitizer-context.test.js test/frontend-hygiene.test.js test/csv-injection.test.js test/stored-taint.test.js test/tree-sitter-sinks.test.js test/kotlin-structural.test.js test/ruby-php-structural.test.js test/java-csharp-structural.test.js test/guard-recognition.test.js test/js-python-framework-structural.test.js test/go-structural.test.js test/secret-concat.test.js test/secret-redaction.test.js test/xss-reflected-multilang.test.js test/code-injection-multilang.test.js test/xxe-multilang.test.js test/xpath-injection-multilang.test.js test/gapfill-batch13.test.js test/agent-untrusted-flow.test.js test/api-authz.test.js test/event-entrypoint.test.js test/iac-terraform.test.js test/cross-service.test.js test/rbac-consistency.test.js test/adversarial-robustness.test.js test/secret-entropy.test.js test/comment-strip.test.js",
+    "test:posture": "node --test test/privacy-framework.test.js test/material-change.test.js test/drift.test.js test/scorecard.test.js test/accuracy-scorecard.test.js test/scorecard-gate.test.js test/predeploy-gate.test.js test/release-check.test.js test/pre-push-gate.test.js test/dependency-currency.test.js test/package-contents-check.test.js test/gate-verdict-cache.test.js test/independent-population.test.js test/evidence-bundle.test.js test/discovery-memory.test.js test/mttr.test.js test/license-policy.test.js test/aibom.test.js test/sbom.test.js test/api-inventory.test.js test/iam-policy.test.js test/container.test.js test/container-runtime.test.js test/image-packages.test.js test/kev.test.js test/dep-confusion.test.js test/sca-deprecated.test.js test/sca-batch.test.js test/composite-risk.test.js test/epss-enrich.test.js test/sca-coverage.test.js test/gradle-deps.test.js test/sca-route-reachable.test.js test/sca-policy.test.js test/sca-verdict.test.js test/install-script.test.js test/sca-linked-findings.test.js test/packs.test.js test/flow-narration.test.js test/regression-test-gen.test.js test/deterministic-fix.test.js test/falsification.test.js test/verification-separation.test.js test/logic-claims.test.js test/attestation.test.js test/verify-attestation-cli.test.js test/comparison.test.js test/determinism-cross-machine.test.js test/fix-honesty-gate.test.js test/model-routing.test.js test/root-cause-sweep.test.js test/entrypoint-inventory.test.js test/relevance.test.js test/untrusted.test.js test/agent-hardening.test.js test/rule-synthesis.test.js test/policy-gate.test.js test/agents-memory.test.js test/cve-lookup.test.js test/cve-alert-daemon.test.js test/fix-verify-loop.test.js test/fix-verify-tests.test.js test/fix-acceptance.test.js test/exploitability-probability.test.js test/history-scan.test.js test/viral-features.test.js test/viral-v074.test.js test/state-dir.test.js test/ignore-pragma.test.js test/license-graph.test.js test/secret-live-check.test.js test/attack-taxonomy.test.js test/triage-memory.test.js test/pr-augment.test.js test/chat-batch2.test.js test/chat-batch3.test.js test/chat-batch4.test.js test/chat-batch5.test.js test/chat-batch6.test.js test/router.test.js test/legacy-alias-redirect.test.js test/cache-economics.test.js test/coverage-report.test.js test/corpus-status.test.js test/provenance.test.js test/secret-history.test.js test/execution-proof.test.js test/fix-metrics.test.js test/corpus-enroll.test.js test/prove-findings.test.js test/corpus-match.test.js test/corpus-provenance.test.js test/learning-quorum.test.js test/llm-cache-integrity.test.js test/proof-artifact.test.js test/providers.test.js test/fleet.test.js test/autopilot.test.js test/autopilot-cli.test.js test/verifier-independence.test.js test/integrity-legacy-key.test.js test/suppression-visibility.test.js test/cost-ceiling.test.js test/local-endpoint.test.js test/model-trust.test.js test/vuln-archaeology.test.js test/scan-checkpoint.test.js test/llm-redact.test.js test/world-class-batch2.test.js test/world-class-modules.test.js test/integration-end-to-end.test.js test/streak-cli-wiring.test.js test/stack-playbook.test.js test/machine-output-flag.test.js test/only-flag-channels.test.js test/confidence.test.js test/security-trend.test.js test/calibration-drift.test.js test/validator-metrics.test.js",
+    "test:dataflow": "node --test test/fn-reach.test.js test/deep-taint.test.js test/calibration.test.js test/holdout-eval.test.js test/cross-lang-meta.test.js test/cross-lang-queues.test.js test/phase5-xlang.test.js test/phase5-coverage.test.js test/phase6-taint.test.js test/llm-validator-consistency.test.js test/llm-validator-default-on.test.js test/llm-validator-preset.test.js test/parser-py-cst.test.js test/parser-cs-kt.test.js test/parser-go.test.js test/parser-php-rb.test.js test/interproc-k2.test.js test/proven-clean.test.js test/backward-default.test.js test/incremental-cache.test.js test/string-regex-lattice.test.js test/closure-capture.test.js test/points-to.test.js test/type-stubs.test.js test/soft-taint.test.js test/ifds.test.js test/symbolic-exec-proof.test.js test/ifds-summary-edges.test.js test/stub-aware-filter.test.js test/cross-repo.test.js test/proof-gate.test.js test/layer-recall.test.js test/java-taint-flow.test.js test/kt-taint-flow.test.js test/sanitizer-typed-flow.test.js test/sanitizer-gate-unit.test.js test/proof-safe.test.js test/collection-taint.test.js test/kcfa-context.test.js test/kcfa-callstring.test.js test/flow-parity.test.js test/callgraph-resolve.test.js test/import-reachability.test.js test/ir-stats.test.js test/parser-cpp.test.js test/parser-js-decorators.test.js test/parser-js-if-else-cfg.test.js test/cpp-integration.test.js test/engine-reconnect.test.js test/phase2-scoping.test.js test/engine-recall.test.js test/catalog-expanded.test.js test/builtin-summaries.test.js test/interproc-mutation-crash.test.js test/mutated-param-taint.test.js test/higher-order-inline-callback.test.js test/interproc-findings-discard.test.js test/runscan-deep-option.test.js test/summary-cache-convergence.test.js test/parser-rb-calls.test.js test/balanced-call.test.js test/destructuring-taint.test.js test/computed-member-taint.test.js test/string-interpolation-taint.test.js",
+    "test:mcp": "node --test test/mcp.test.js test/mcp-audit.test.js test/audit-cli.test.js test/mcp-scratchpad.test.js test/mcp-offload.test.js test/sca-upgrade.test.js test/lsp-server.test.js",
+    "test:report": "node --test test/junit.test.js test/ci.test.js test/poc-generator.test.js test/verifier.test.js test/verifier-target.test.js test/annotator-errors.test.js test/grader-calibration.test.js test/pr-delta-gate.test.js test/vex.test.js test/report-render.test.js",
     "test:bench-modules": "node --test test/phase4-harness.test.js test/pipeline.test.js test/proof-corpus-lib.test.js test/proof-corpus-runner.test.js",
-    "test:lifecycle": "node --test test/dead-code.test.js test/no-dead-modules.test.js test/stop-hook.test.js test/plugin-self-check.test.js test/skills-registry.test.js test/bodyguard.test.js test/cache-invalidator-guard.test.js test/dispatch-pre-tool.test.js test/dispatch-user-prompt.test.js test/session-start-model-capture.test.js test/sandbox.test.js test/sandbox-escape.test.js && node ../scripts/lint-command-descriptions.mjs",
+    "test:lifecycle": "node --test test/dead-code.test.js test/no-dead-modules.test.js test/no-orphan-scripts.test.js test/no-stray-state.test.js test/tree-integrity.test.js test/stop-hook.test.js test/plugin-self-check.test.js test/skills-registry.test.js test/bodyguard.test.js test/cache-invalidator-guard.test.js test/dispatch-pre-tool.test.js test/dispatch-user-prompt.test.js test/session-start-model-capture.test.js test/sandbox.test.js test/sandbox-escape.test.js test/ci-templates.test.js test/agent-mcp-tools.test.js test/triage-command.test.js test/no-orphan-tests.test.js test/supply-command.test.js test/posture-command.test.js test/labs-command.test.js test/compliance-command.test.js test/check-doc-drift.test.js test/dist-chunks-tracked.test.js && node ../scripts/lint-command-descriptions.mjs",
     "test:eval": "node --test test/independent-eval.test.js test/realworld-recall.test.js",
     "test:discovery": "node --test test/discovery-partition.test.js test/discovery-lenses.test.js test/discovery-hunter.test.js test/discovery-confirm.test.js test/discovery-disprove.test.js test/discovery-judge.test.js test/discovery-run.test.js test/discovery-wiring.test.js test/discovery-llm-invoke.test.js",
     "eval:independent": "node ../bench/independent-eval/runner.mjs",
@@ -95,9 +97,15 @@
     "bench:engine-reconnect": "node ../bench/engine-reconnect/measure.mjs",
     "bench:self-scan": "node ../bench/self-scan/measure.mjs",
     "bench:self-scan:check": "node ../bench/self-scan/check.mjs",
+    "bench:mutation:check": "node ../bench/mutation/runner.mjs",
+    "bench:layer-recall": "node ../bench/layer-recall/runner.mjs",
+    "bench:layer-recall:check": "node ../bench/layer-recall/runner.mjs --check",
+    "bench:layer-recall:update-baseline": "node ../bench/layer-recall/runner.mjs --update-baseline",
     "bench:self-scan:update-baseline": "node ../bench/self-scan/check.mjs --update-baseline",
     "bench:engine-recall": "node ../bench/engine-recall/measure.mjs",
     "scorecard": "node ../scripts/scorecard.mjs",
+    "check-doc-drift": "node ../scripts/check-doc-drift.mjs",
+    "harness-score": "node ../scripts/harness-score.cjs",
     "corpus:enroll": "node ../scripts/enroll-proven-finding.mjs",
     "corpus:provenance": "node ../scripts/corpus-provenance-check.mjs",
     "fleet": "node ../scripts/fleet.mjs",
@@ -108,7 +116,11 @@
     "release:check": "node ../scripts/release-check.mjs",
     "release:check:fast": "node ../scripts/release-check.mjs --fast",
     "gate:prepush": "node ../scripts/pre-push-gate.mjs",
-    "gate:prepush:install": "node ../scripts/pre-push-gate.mjs --install-hook"
+    "gate:prepush:install": "node ../scripts/pre-push-gate.mjs --install-hook",
+    "bench:independent": "node ../bench/independent/runner.mjs",
+    "bench:independent:fetch": "node ../bench/independent/fetch.mjs",
+    "bench:independent:mine": "node ../bench/independent/mine.mjs",
+    "bench:independent:materialise": "node ../bench/independent/materialise-cli.mjs"
   },
   "author": "Ross Young <ross@clearcapabilities.com>",
   "license": "PolyForm-Internal-Use-1.0.0"

package/src/badge.js

@@ -28,6 +28,7 @@
 import * as fs from 'node:fs';
 import * as path from 'node:path';
 
+import { statePath } from './posture/state-dir.js';
 const COLORS = {
   critical:    '#e05d44',  // red
   high:        '#fe7d37',  // orange
@@ -42,7 +43,7 @@ const SEVERITIES = ['critical', 'high', 'medium', 'low', 'info'];
 
 function _readLastScan(scanRoot) {
   if (!scanRoot) return null;
-  const fp = path.join(scanRoot, '.agentic-security', 'last-scan.json');
+  const fp = statePath(scanRoot, 'last-scan.json');
   if (!fs.existsSync(fp)) return null;
   try { return JSON.parse(fs.readFileSync(fp, 'utf8')); }
   catch { return null; }

package/src/dataflow/CLAUDE.md

@@ -6,8 +6,12 @@ Layer-2 taint engine. Walks the Layer-1 IR (`../ir/`) with field-sensitive forwa
 
 - **Intra-procedural field-sensitive taint** with access-path lattice (`access-paths.js`). `user.email` is distinguishable from `user.password`.
 - **Value-context-sensitive interprocedural taint (FR-SEM-2).** `SummaryCache` (`summaries.js`) holds a distinct summary per distinct entry-taint-state. A pre-pass computes the empty-entry base for every function; call sites then lazily compute the summary under their actual tainted-arg context (at both assign-from-call and plain-call sites) so a helper that is clean with clean args but tainted with user input is detected per call site. Bounded by a per-function context cap (`AGENTIC_SECURITY_KCFA_MAX_CONTEXTS`, default 16; 0 = monovariant). Over the cap → reuse the empty-entry summary.
-- **Catalog-driven source/sink matching.** Add entries in `catalog.js`. Each entry: `kind` ∈ {source, sink, sanitizer}, plus language + framework + match shape. 200+ entries spanning Express/Flask/FastAPI/Django/Rails/PHP/Go-net-http/Gin/Echo.
-  - **Sanitizer entries are NOT consulted by this engine.** `matchSinkOrSanitizer()` returns hits of both kinds, but every consumer in `dataflow/*.js` keeps only `e.kind === 'sink'` — there is no `'sanitizer'` branch anywhere in this directory. Taint dies here only on clean re-assignment of a variable (`removePathAndDescendants`), independent of whether the RHS is a catalog sanitizer. Consequence: `engine.js`'s `applySanitizerGate` call is **inert** (`sanitizersOnPath` is always `{}`) because there is nothing to populate it from. Making it live needs (1) a `kind === 'sanitizer'` branch at the call site in `engine.js`, and (2) the sanitizer callee threaded onto the finding.
+- **Catalog-driven source/sink matching.** Add entries in `catalog.js`. Each entry: `kind` ∈ {source, sink, sanitizer}, plus language + framework + match shape. 641 entries (136 source, 124 sink, 381 sanitizer — re-derive via `CATALOG.length` / a per-`kind` count rather than trusting a hardcoded number here again) spanning Express/Flask/FastAPI/Django/Rails/PHP/Go-net-http/Gin/Echo.
+  - **Sanitizer entries are RECORDED, never trusted to kill taint.** `_sanitizersForExpr()` collects the sanitizer callees applied to the value reaching each sink argument — inline in the argument, or inherited from the variable it reads (`_sanitizersByVar`, updated on every assignment and cleared when the RHS is clean). They are stamped on the finding as `_sanitizersOnPath` and consumed by `sanitizer-gate.js`, which sets `sanitized` + `sanitizerProof` **only when the sanitizer's `appliesTo` family covers the finding's threat class** (`*` covers everything). The proof gate then demotes. Taint itself still dies only on clean re-assignment (`removePathAndDescendants`).
+    - **Why the sanitizer does not kill the taint.** A blanket "any sanitizer clears the flow" rule scores well on benchmarks and silently drops a real SQL injection whenever the code applied an HTML escaper — `htmlspecialchars` neutralises XSS and does nothing for SQLi. Labelling demotes confidence; killing taint would delete the finding. Recall-preserving, same precedent as `falsification.js` / `proof-gate.js`: never removed, never severity-touched.
+    - **The sanitization signal is may-, not must-.** `_sanitizersByVar` is keyed per variable within a function and is not branch-sensitive, so a value sanitized on only one path still carries the label. That is deliberate and safe *because* the label only demotes — the same imprecision under a taint-killing design would be a false negative.
+    - **The gate runs twice**, for the same reason `_applyIgnorePragmas` does: pass 1 in `engine.js` runs ~2300 lines before deep-mode IR findings are appended, so pass 2 runs immediately after `finalFindings.push(...irFindings)` and is scoped to `irFindings` only (re-running `annotateProofGate` over already-gated findings would demote them twice).
+    - Guarded by `bench/mutation/runner.mjs` (`npm run bench:mutation:check`), which scores verdict-flip correctness over metamorphic rewrites (verdict must hold) and adversarial near-misses (verdict must flip). Removing the family check makes it exit non-zero.
   - **`match.receiver`** (optional, regex source string) constrains a call to call sites whose receiver chain has a matching segment. Callee matching is by *bare name*, so `callee: 'write'` otherwise matches `process.stdout.write` and `fh.write` as readily as `document.write`. Works for both sinks (via `matchSinkOrSanitizer`) and sources (via `matchSource`'s CALL-source path). `match.receiverBase` is an optional, independent second pattern (both `_receiverAllowed` checks must pass) — use it when `receiver` alone would still match an unrelated same-named local, e.g. `py-flask-args-get` requires *both* a property segment (`args`/`form`/`headers`/…) *and* a `request`/`req` segment, so `args = parse(); args.get(x)` doesn't fire just because some catalog entry happens to be named `get`.
   - **`language`** is enforced by `_languageAllowed` (Phase 2) via a table (`_LANG_EXT` in `catalog.js`) covering all nine catalog languages — `js`, `py`, `cs`, `kt`, `go`, `php`, `rb`, `java` by literal regex kept byte-identical to `ir/index.js`'s own dispatch regexes (pinned by `test/phase2-scoping.test.js`, which compares `.source` directly, not just "does this regex literal appear somewhere in the file"), and `cpp` by delegating to `ir/parser-cpp.js`'s `cppExtRe()` so it can't drift from the parser. A language with no table entry stays permissive (matches any file) — additive by design, so it can't regress a language before its entry exists.
 - **Path feasibility.** Constant-folds `if` conditions to prune unreachable branches.
@@ -49,5 +53,7 @@ Layer-2 taint engine. Walks the Layer-1 IR (`../ir/`) with field-sensitive forwa
 
 - **Path attribution.** If you're adding a sink to the catalog, set `argIndex` carefully. `'all'` means "any tainted arg fires"; a numeric index pinpoints THE arg whose taint matters. Wrong here → noisy findings with confused traces.
 - **Cache invalidation.** `SummaryCache` is in-memory per-scan. Cross-scan persistence lives in `incremental.js` (FR-incremental) but it's behind a separate flag. Don't conflate the two.
-- **Recursion.** The cache returns a bottom summary (`_recursive: true`) when it hits a function already on the stack. The engine relies on fixed-point iteration to refine — but `runTaintEngine` does only ONE pass today. Recursive cycles will under-approximate.
-- **`AGENTIC_SECURITY_BLIND_BENCH=1` disables the deep engine** along with everything else bench-shape. If you're trying to bench taint quality, run with both `AGENTIC_SECURITY_DEEP=1` and `AGENTIC_SECURITY_BLIND_BENCH=0` (the default).
+- **Recursion.** The cache returns a bottom summary (`_recursive: true`) when it hits a function already on the stack. The engine relies on fixed-point iteration to refine, and `runTaintEngine` does run a multi-pass loop (`MAX_FP_ITERS = 3`, `engine.js`) — this gotcha previously claimed a single pass, which contradicted the "Recursion via fixed point" bullet above and was stale. Cycles that have not converged within 3 iterations still under-approximate.
+- **`AGENTIC_SECURITY_BLIND_BENCH=1` does NOT disable the deep engine.** It gates the bench-shape detectors; `_deepRequested` in `engine.js` keys on `AGENTIC_SECURITY_DEEP === '1'` alone, so deep mode runs regardless. (This bullet previously claimed the opposite — verified false while building `bench/layer-recall/`, where setting `BLIND_BENCH=1` left taint recall unchanged at 20/210.) To turn the taint engine off, set `AGENTIC_SECURITY_DEEP=0`.
+- **The corpus barely exercises this engine.** `bench/cve-replay` enables deep mode only for its 6 `deep`-tier entries, and with the deep engine off entirely, corpus detection falls only 210 → 204. Taint quality is measured by `npm run bench:layer-recall` (which forces deep on for all 210), not by the corpus gate — see `docs/METRICS.md`.
+- **Resolved: `stub-aware-filter.js` no longer mutates `severity`.** It used to (`f.severity = downgrade[f.severity]`, opt-in via `AGENTIC_SECURITY_TYPE_STUBS=1`), which violated the recall-preserving convention every other precision annotator in this directory follows. Flagged by documentation audit, then fixed on maintainer decision: it now demotes `confidence` (× 0.4, floor 0.01) + `confidenceTier` + `exploitabilityTier` by one step each, mirroring `proof-gate.js`'s own demotion shape exactly (`_confidenceBeforeStubFilter` records the pre-demotion value, a distinct field from `proof-gate.js`'s `_confidenceBeforeProofGate` so the two compose if both fire on the same finding). Severity is never touched.

package/src/dataflow/builtin-summaries.js

@@ -125,7 +125,7 @@ export function lookupBuiltinSummary(calleeName) {
   if (lastDot > 0) {
     const short = calleeName.slice(lastDot + 1);
     const fallback = BUILTIN_SUMMARIES.get(short);
-    if (fallback) return null;
+    if (fallback) return fallback;
   }
   return null;
 }

package/src/dataflow/cross-service-taint.js

@@ -44,12 +44,13 @@ import * as fs from 'node:fs';
 import * as path from 'node:path';
 import * as yaml from '../util/yaml.js';
 
+import { statePath } from '../posture/state-dir.js';
 const SERVICES_FILE_NAMES = ['services.yml', 'services.yaml'];
 
 export function loadServiceGraph(scanRoot) {
   if (!scanRoot) return null;
   for (const name of SERVICES_FILE_NAMES) {
-    const fp = path.join(scanRoot, '.agentic-security', name);
+    const fp = statePath(scanRoot, name);
     if (!fs.existsSync(fp)) continue;
     try {
       const raw = fs.readFileSync(fp, 'utf8');