Skip to content

CHANGELOG.md

Latest commit

 

History

History
423 lines (337 loc) · 27.3 KB

CHANGELOG.md

File metadata and controls

423 lines (337 loc) · 27.3 KB

Changelog

All notable changes to agent-bom are documented here.

Format follows Keep a Changelog. Versions follow Semantic Versioning.

[0.75.12] – 2026-03-29

Added

  • First-class skills scanning — agent-bom skills scan and agent-bom skills verify are now top-level CLI surfaces for instruction-file trust, provenance, and findings
  • Live MCP tool capability scoring — added capability-based tool/server risk assessment from tools/list introspection, surfaced through MCP and scan outputs
  • Release demo refresh — updated hero demo and release surfaces for v0.75.12

Changed

  • MCP tool surface — docs and public surfaces now reflect the current 36-tool MCP server accurately
  • Quickstart alignment — CLI/docs/demo flows now point to the grouped first-class commands (agents, skills, image, iac)
  • Advisory labeling — unscored vulnerabilities are presented as advisories/pending severity instead of ambiguous unknowns
  • Resolver continuity — npm version resolution now prefers cached/bundled continuity under rate limiting instead of repeated blocking retries
  • Supply-chain enrichment — --enrich resolves package source metadata before OpenSSF Scorecard lookup and reports explicit enrichment coverage state

Fixed

  • Blast radius serialization — package name/version/stable ID now propagate correctly in filesystem and JSON outputs
  • Filesystem posture credibility — fs/project scans no longer get penalized for missing MCP-only config context; posture and framework mapping now reflect the actual scan mode
  • Framework tagging — intrinsic vulnerability findings now carry framework tags in both agents and fs modes
  • Remediation output — remediation JSON now includes populated priority and action fields
  • UI dependency hygiene — aligned UI eslint peer range with Next.js-supported versions to remove install drift
  • Scorecard source resolution — npm/PyPI direct dependencies now fall back to source metadata resolution paths so repo URLs, homepages, and Scorecard repos populate end to end

Security

  • No new release-blocking regressions introduced across the 0.75.12 stabilization lane; focused regression suites, UI tests, build checks, and release consistency checks remained green

[0.75.0] – 2026-03-23

Added

  • Dashboard UX — posture grade (A-F) hero, top 5 attack path cards, security graph page with interactive React Flow, insight layer toggle (risk/credentials/default), 14-framework compliance heatmap
  • Remediation page — priority table sorted by blast radius impact, Jira ticket creation per finding, compliance impact summary, severity/framework filters, JSON export
  • Compliance narratives — GET /v1/compliance/narrative generates auditor-ready text per framework with control-level detail and remediation-compliance bridge
  • --posture flag — 5-line workstation posture summary for solo developers
  • --fixable-only flag — show only vulnerabilities with available fixes
  • agent-bom doctor — preflight diagnostic (Python, DB, network, Docker, MCP configs, API keys)
  • Cross-agent behavioral detection — CrossAgentCorrelator detects lateral movement (3+ agents same tool in 5min), anomaly baseline per agent
  • SSE proxy transport — agent-bom proxy --url for remote SSE/HTTP MCP servers
  • SBOM multi-hop graph — dependency depth tracking (A→B→C) + CycloneDX vulnerabilities[] ingest
  • API rate-limit headers — X-RateLimit-Limit/Remaining/Reset on all responses, X-API-Version: v1
  • Jira API endpoint — POST /v1/findings/jira with ephemeral credentials, SSRF-validated
  • False positive feedback — POST/GET/DELETE /v1/findings/false-positive with tenant-scoped persistence
  • Break-glass endpoint — POST /v1/shield/break-glass with admin RBAC + audit logging
  • Prometheus /metrics — fleet_total and fleet_quarantined gauges
  • 75 UI component tests (Vitest + @testing-library/react)
  • 8 intent-based OpenClaw skills — discover, scan, scan-infra, enforce, comply, monitor, analyze, troubleshoot
  • CONTRIBUTING.md — contributor onboarding guide
  • Enterprise Deployment guide — MDM push, fleet API, zero-credential architecture

Changed

  • Homepage reworked — posture grade + blast radius chains at top, stats compressed to one row
  • Compliance page — now shows all 14 frameworks (was 6)
  • Security graph — uses pre-computed blast_radius scores (risk_score, is_kev, epss_score)
  • All dashboard pages — consistent Loader2 spinners, overflow-x-auto tables, confirmation dialogs on destructive actions, Snowflake-only banners in error state
  • Vulns page — pagination (50/page), search, FP feedback button, confidence scores
  • Jobs page — status filter tabs, search, pagination (25/page), JSON export
  • Fleet page — search, JSON export, confirmation on state transitions
  • Agents page — search by name
  • CLI output — severity text labels alongside colors (accessibility)
  • CycloneDX — formulation field identifies agent-bom as generator
  • GitHub Action — exclude-unfixable input for CI gating
  • Architecture diagram — compact horizontal layout (LR)
  • Count alignment — release surfaces were brought back to a single source of truth for that release's framework, rule, pattern, page, tool, and format totals

Fixed

  • Full-stack alignment — severity_source, confidence, nist_800_53_tags, fedramp_tags, automation_settings, vector_db_scan, gpu_infra now serialized in JSON output (were silently dropped)
  • Compliance router — /v1/compliance/narrative no longer shadowed by /{framework} wildcard
  • UI field names — risk_score ?? blast_score, summary ?? description, is_kev ?? cisa_kev with backward compat
  • Offline mode strict — no silent network fallback when --offline set
  • AST prompt detector — description, help, title fields no longer misclassified as system prompts
  • CodeQL SSRF — defense-in-depth validate_url() at transport layer
  • HSTS header — Strict-Transport-Security added to all API responses
  • OIDC SSRF — validate_url() on discovery URL
  • ECS/EKS test mocks — updated to paginator pattern
  • Protection engine — stop() persists cleared kill-switch state, semaphore cache bounded to 8 entries
  • Chain-hashed audit log — each entry includes previous entry's HMAC for tamper-evidence
  • Multi-tenancy isolation — tenant_id enforced at middleware level
  • Quarantine enforcement — quarantined agents excluded from fleet list by default
  • Log file permissions — 0o600 on audit DB, fleet DB, log files
  • Node.js 20 deprecation — FORCE_JAVASCRIPT_ACTIONS_TO_NODE24 on dependency-review
  • Local fonts — Inter + JetBrains Mono bundled (no Google Fonts network dependency)

Security

  • 10/10 OWASP web categories PROTECTED (verified by code audit)
  • SLSA L3 provenance on releases, Sigstore signing on PyPI
  • 91 SHA-pinned GitHub Actions, 0 npm/Python vulnerabilities
  • scrypt KDF for API keys, HMAC constant-time comparison, parameterized SQL everywhere

[0.74.1] – 2026-03-22

Security

  • Runtime: Fix threat level comparison — ThreatLevel enum uses ordinal comparison; previously CRITICAL > HIGH was False (lexicographic), so escalations were never detected
  • Runtime: Per-session shield engine isolation — zero trust, no cross-session threat contamination
  • Runtime: Rate limiter now blocks (HIGH/CRITICAL severity) instead of alert-only (MEDIUM)
  • Runtime: HTTP protection server auth via AGENT_BOM_PROTECTION_API_KEY with constant-time comparison
  • Runtime: Single zero-width space now detected (was requiring 3+ consecutive chars, bypass fixed)
  • Runtime: Unicode path traversal (\u002e\u002e) detection added
  • Runtime: 3 new credential patterns (HashiCorp Vault, AWS session tokens, PagerDuty) — 112 total
  • Runtime: Drift baseline persisted to disk, survives engine restarts in shield mode
  • Supply chain: Go module provenance verification via sum.golang.org checksum database
  • Supply chain: Provenance results persisted in Package model (integrity_verified, provenance_attested)

Fixed

  • Compliance: Wire NIST 800-53 Rev 5 tags in scanner pipeline (was defined but never called)
  • Compliance: Wire FedRAMP Moderate baseline tags in scanner pipeline
  • Policy: Add 9 missing framework tag fields to policy engine expressions (all 14 frameworks now accessible)
  • Scanner: Surface non-OSV ecosystem skips to users with per-ecosystem breakdown
  • Scanner: Track per-package CVE lookup errors and report at scan completion
  • Scanner: Add severity_source field for audit trail (cvss, osv_database, ghsa_heuristic)
  • Scanner: Bound semaphore cache to 8 entries (prevents memory leak in long-running API servers)
  • Scanner: Kill-switch state now persisted on stop() (was leaving stale blocked state on disk)
  • Cloud: ECS/EKS/SageMaker discovery uses boto3 paginators (was missing results beyond page 1)
  • MITRE ATLAS: Expanded from 8 to 30+ technique mappings across all tactic categories

Added

  • make dev — starts API server + Next.js dashboard in parallel
  • make install-all — installs all development extras (dev + ui + mcp-server)
  • dev-all extra in pyproject.toml for one-command setup
  • AGENT_BOM_CLOUD_DISCOVERY_TIMEOUT config (default 45s)
  • Portable .claude/launch.json for dev server configs
  • Mermaid blast radius diagram in README
  • Mermaid architecture diagram in README
  • Compliance frameworks table in README

Changed

  • README: ecosystem count 11 → 15, credential patterns 31 → 34, compliance frameworks 14 → 16
  • MITRE ATLAS coverage documented accurately (30+ observable techniques)

[0.74.0] – 2026-03-21

Dashboard & HTML Report

  • Sidebar navigation — collapsible left sidebar replaces top nav across Next.js dashboard (18 pages) and HTML report, with 5 grouped sections, command palette (Cmd+K), active link highlighting, mobile drawer (#993)
  • React Flow edge fix — topology graph nodes now connect properly with Handle components, eliminating all "null handle id" errors (#993)
  • Topology enhancements — nodes show server count, vuln count, credential indicators; animated edges for vulnerable connections; zoom controls; empty state (#993)
  • HTML report sidebar — fixed 220px sidebar with IntersectionObserver active section tracking, mobile responsive hamburger toggle, print-friendly layout (#993)

Security

  • Dependency CVE fixes — pillow ≥12.1.1 (CVE-2026-25990 PSD OOB write), protobuf ≥6.33.5 (CVE-2026-0994 ParseDict DoS), pyjwt ≥2.12.0 (CVE-2026-32597 crit header bypass), tornado ≥6.5.5 (CVE-2026-31958 multipart DoS, GHSA-78cv-mqj4-43f7 cookie injection)
  • CI/CD hardening — all GitHub Actions pinned to SHA digests, no @master refs, scoped permissions

Fixed

  • Dashboard layout — wider content area (1400px), CSS page transitions, thin scrollbar, card hover effects
  • HTML report unused variables — removed dead nav link variables (ruff F841)

[0.73.0] – 2026-03-19

Architecture

  • 5-product CLI — agent-bom, agent-shield, agent-cloud, agent-iac, agent-claw — one package, five entry points, zero duplication (#967)
    • agent-bom — BOM generation + vulnerability scanning
    • agent-shield — runtime protection (proxy, protect, guard, watch, audit)
    • agent-cloud — cloud infrastructure (aws, azure, gcp, snowflake, databricks, huggingface, ollama, posture)
    • agent-iac — IaC security (scan, policy, validate)
    • agent-claw — fleet governance (fleet, serve, api, schedule, report, connectors)
  • Shared entry point factory (_entry.py) — consistent error handling + update check across all 5 products
  • Parameterized help categories — each product shows relevant command groups in --help
  • Full backward compat — agent-bom runtime proxy, agent-bom cloud aws, etc. still work

Added

  • CycloneDX 1.6 ML BOM extensions — native modelCard, data components, machine-learning-model type with quantitativeAnalysis for training metrics (#967)
  • Agent-shield deep defense mode (--shield) — correlated multi-detector threat scoring, ThreatLevel escalation (NORMAL→ELEVATED→HIGH→CRITICAL), automatic kill-switch on CRITICAL (#967)
  • Graph-native AIBOM export — to_graphml() for yEd/Gephi/NetworkX, to_cypher() for Neo4j with AIBOM node labels (AIAgent, MCPServer, Package, Vulnerability) (#967)
  • CLI graph formats — agent-bom graph --format graphml|cypher alongside existing json/dot/mermaid (#968)
  • CLI shield flags — agent-shield protect --shield --correlation-window 30 (#968)
  • API: graph export — GET /v1/scan/{job_id}/graph-export?format=graphml|cypher|dot|mermaid|json (#970)
  • API: shield control — POST /v1/shield/start, GET /v1/shield/status, POST /v1/shield/unblock (#970)
  • MCP graph tool — graph_export MCP tool for graphml/cypher/dot/mermaid from scan results (#972)
  • agent-cloud new commands — snowflake, databricks, huggingface, ollama, posture (#967)
  • agent-claw fleet — fleet sync|list|stats commands + connectors command (#967)

Fixed

  • Fleet min_trust Python 3.14 — explicit float() coercion + defensive (trust_score or 0.0) (#969)
  • API test noise — mock _run_scan_sync with pytest monkeypatch fixture scoped to test file only (#970)
  • flatted 3.4.1→3.4.2 — GHSA-rf6f-7fwh-wjgh prototype pollution in /ui (#970)
  • CVE-2026-33231 suppression — nltk wordnet_app remote shutdown (not reachable — never imported) (#971)
  • Stale test name — test_total_agent_types_is_18 → is_31 (#967)

[0.72.0] – 2026-03-19

Added

  • 30 MCP clients — added Pieces, mcp-cli, Trae, Aide, Void, Replit Agent, Aider, Sourcegraph Cody, Tabnine, Copilot CLI, Junie, JetBrains AI
  • AI BOM tools — 32 MCP tools for scanning, compliance, runtime, cloud in the 0.72.0 release
  • Version accuracy — all surfaces updated with correct version and client counts
  • Compliance noise reduction — actionable findings only in default output

[0.71.4] – 2026-03-18

Fixed

  • SARIF output: paths were absolute (/Users/.../Agent-Bom) causing GitHub Security tab to show "No summary of scanned files". Now relative paths pointing to manifest files (#945)
  • Scanner accuracy: git commit SHAs in fixed_version from OSV data no longer generate false positive "assuming affected" warnings — silently skipped as non-version data (#946)
  • Docker pip CVEs: upgraded pip in all 6 Docker images — fixes CVE-2025-8869 (5.9M) and CVE-2026-1703 (2.0L) (#947)
  • Release pipeline: self-scan gate now blocks ALL publish jobs (PyPI, Docker, GHCR, Sigstore, SLSA). No release ships with HIGH+ CVEs (#944)
  • Self-scan noise: post-merge scan now only uploads HIGH+ findings to GitHub Security tab (was uploading all severities, flooding with 78 low-severity base-image CVEs) (#945)

[0.71.3] – 2026-03-18

Fixed

  • Scanner accuracy: empty string versions no longer silently bypass OSV queries — now logged as unresolvable (#933)
  • HTTP reliability: all 19 urllib.request.urlopen callers migrated to httpx with exponential backoff, Retry-After handling, and configurable timeouts (#878, #932)
  • Context graph OOM: BFS queue bounded at 10,000 entries + O(1) cycle check via frozenset to prevent memory explosion on dense deployments (#877, #930)
  • Docker Hub token leak: release.yml now masks API token with ::add-mask::, validates before use (#929)
  • CI self-scan gap: added agent-bom scan --os-packages inside built Docker image to catch base-image CVEs (glibc, sqlite3, dpkg) that Docker Hub found but CI missed (#931)
  • Alpine CI image: pinned python:3.12-alpine by SHA256 digest (#929)
  • Glama Dockerfile: added missing HEALTHCHECK (#929)
  • Container rescan: weekly Trivy scan escalated from informational to exit-code 1 with warning annotation (#931)
  • action.yml: shortened description to 91 chars (was 141, GitHub Marketplace truncates at 125) (#933)

Added

  • CMMC 2.0 in REST API: wired 17 CMMC practices into /v1/compliance endpoint — 14 frameworks now accessible via API (#933)
  • Sync HTTP client: create_sync_client(), sync_request_with_retry(), fetch_bytes(), fetch_json() in http_client.py (#932)

[Unreleased]

Added

  • Next.js Insights page (/insights) — SupplyChainTreemap, BlastRadiusRadial, PipelineFlow, EpssVsCvssChart, VulnTrendChart wired to real scan data; treemap cells are clickable and drill down to /vulns
  • Gateway enforcement chart — audit tab shows stacked bar of blocked/alerted/allowed actions per tool
  • Governance findings chart — stacked bar of finding severity by governance category
  • Activity query chart — bar chart of agent query pattern frequency
  • Fleet lifecycle chart — bar chart of agents by lifecycle state
  • Jobs status donut — pie chart summarising job queue by status
  • ui/components/empty-state.tsx — reusable EmptyState + ErrorBanner components used across pages
  • Retry buttons — Activity and Governance error states now include a Retry button
  • SECURITY.md expanded — response SLA, known limitations, API security model, disclosure timeline
  • PR template — added Related Issues, TypeScript check, breaking changes, checklist sections
  • pre-commit hooks — added check-yaml, check-json, check-toml, end-of-file-fixer, trailing-whitespace, detect-private-key, check-merge-conflict, mixed-line-ending

Fixed

  • JSON report import — file upload now validates size (10 MB), schema, prototype-pollution keys, and finite numeric values before use (ui/lib/validators.ts)
  • generated_at TypeScript error — ScanResult does not have generated_at; use scan_timestamp instead
  • JetBrains claim — removed from active integrations; filed as issue #412 for proper implementation

Security

  • JSON file upload — ui/lib/validators.ts guards against DoS via oversized files, prototype pollution, and schema-invalid payloads (no new npm dependencies)

[0.71.2] – 2026-03-16

Fixed

  • Symlink cycle in parser walk — _walk() now tracks visited real paths via os.path.realpath() to prevent the same directory being scanned up to 5× through symlinks (#876)
  • Transitive dep fetch failures now logged — silent except (ValueError, KeyError): pass in npm/PyPI fetch functions replaced with _logger.warning(); gather exceptions also routed through logger in addition to console (#879)

Docs

  • ARCHITECTURE.md — added agent-bom run node to CLI Commands subgraph wired to Runtime Proxy
  • SKILL.md — bumped version, Docker tag, Sigstore reference, and test count to current values

[0.71.1] – 2026-03-15

Fixed

  • GHSA false positives (authlib, pyjwt) — GitHub Advisory API now returns patched_versions=null; vulnerable_version_range is the authoritative field. Added _installed_version_is_affected() helper parsing compound constraints (<=, <, >=, >). Fixed packages already past the vulnerable window no longer flagged (#895)
  • Empty-version packages passed to scanners — packages with version="" were unconditionally matched against all advisories. Added "" to the scannable exclusion filter in scanners/__init__.py
  • NVIDIA scanner unconditional matches — CSAF vulns with a fixed_version now call compare_versions() before appending; unversioned packages skipped
  • Proxy credential detection in error responses — check_credentials was only applied to "result" fields; JSON-RPC "error" objects can carry exception messages with API keys. Now applies to both (#896)
  • Rate limit advisory-only — --rate-limit logged alerts but never blocked. Enforcement block path added matching replay-detection pattern (#896)
  • Audit log rotation silent failure — except OSError: pass replaced with logger.warning() to surface disk-full/permission errors (#896)

[0.70.4] – 2026-03-11

Added

  • Cloud provisioning scripts — read-only, least-privilege provisioning for all 11 providers (AWS IAM + EKS RBAC, Azure Managed Identity, GCP Workload Identity, Snowflake key-pair JWT, Databricks PAT/OAuth, HuggingFace fine-grained token, W&B Viewer service account, Nebius IAM, CoreWeave namespace RBAC, NVIDIA NGC Viewer key) in scripts/provision/
  • Nebius pagination — cursor-based pagination with nextPageToken/next_page_token wired into all 3 discovery functions
  • Post-merge self-scan — GitHub Actions workflow scans agent-bom with agent-bom on every merge; blocks release on critical CVE (#648)
  • Two-tier severity gate — --warn-on for CI warning gates, --fail-on-severity for hard failures (#625)
  • Trivy/Grype/Syft ingestion — import external scanner JSON output with blast radius enrichment (#624)
  • Delta scanning — --delta flag reports only new findings since last scan; exit code based on new-only (#630)
  • Local embedded vulnerability database — SQLite schema, OSV/EPSS/KEV sync, fast lookup (#631)
  • Bun, NuGet (.NET), pip-compile parsers (#660)
  • Gradle and conda parsers for AI/ML ecosystems (#659)
  • go.sum integrity verification + GOPROXY version resolution (#658)
  • Maven Central and crates.io version resolution for unpinned deps (#661)
  • GHSA and NVD local DB sync sources (#653)
  • Multi-source asset deduplication — cross-cloud dedup with stable IDs (#654)
  • Deterministic UUID v5 stable IDs for assets and findings (#655)
  • Auto-refresh stale vuln DB flag — --auto-refresh-db, NIM/NeMo/NemoClaw NVIDIA tracking
  • Production-quality Go/Maven/RPM parser improvements (#656)

Fixed

  • Credential security — URL validation (jira.py, slack.py), timing-safe metrics token compare (proxy.py), API key sanitization from exception messages (vector_db.py) (#662)
  • Databricks enum bug — EndpointStateReady.value fix; str(enum) returned full name not value, causing all serving endpoints to be skipped (#664)
  • CoreWeave/NVIDIA provisioning — kubeconfig + namespace RBAC, NGC Viewer key, DCGM exposure detection (#664)
  • Normalization gaps — CLI check, scan_agents, rescan, postgres cache (#615)
  • GHSA PEP 503 normalization — advisory matching + resolver debug logging (#619)
  • Multi-arch container rescan — arm64 support, SARIF to Security tab (#650)
  • OTel hardening — schema validation, file size cap, framework expansion (#642)
  • Stale DB warning on outdated local cache (#642)
  • Local vuln DB security — chmod 0600, HTTPS-only sync, path validation, integrity check (#634)
  • HTML report — delta/warn-gate banners, vendor_severity display (#632)
  • Silent exception handlers — logging added to all bare except blocks (#620)
  • Documentation accuracy — detector count (6→7), architecture client and tool counts (#657)
  • MCP tool count — replaced hardcoded counts with dynamic assertions (#626)

Changed

  • cli/scan.py refactored — 3,079L monolith → modular scan/ package (#651)
  • Unified Finding model Phase 1 — core dataclasses, BlastRadius migration shim, dual-write (#628)
  • PEP 503 name normalization — configurable batch size, unresolved package warnings (#614)
  • server.py routes extracted — scan, discovery, connectors, governance, enterprise, schedules, observability, assets (#612, #613)

0.60.1 – 2026-03-08

Fixed

  • P0: Ecosystem case normalization — packages with ecosystem "PyPI" (uppercase) were silently returning 0 vulnerabilities. Fixed with .lower() normalization at every ECOSYSTEM_MAP lookup in query_osv_batch and scan_packages.
  • P0: OSV detail enrichment — OSV /v1/querybatch returns only {id, modified} per vuln (no summary, CVSS, aliases). Added parallel detail fetching via /v1/vulns/{id} (semaphore=10), called at every return path including cache-hit early returns.
  • P0: Cache blocking I/O — SQLite cache writes now run off the event loop via asyncio.to_thread + batched put_many() transaction, preventing event loop stalls on large scans.
  • Zero test failures — test_accuracy_baseline.py fully fixed: lowercase ecosystem, asyncio.run(), correct Severity enum values.

Added

  • Update notifications — background daemon thread checks PyPI on startup, 24-hour file cache at ~/.cache/agent-bom/, non-blocking notice shown on clean exit.
  • Improved --version output — shows Python version, platform, Syft/Grype install status.
  • Better first-run UX — zero-config scan shows actionable quick-start commands.

Changed

  • CONTRIBUTING.md rewritten — Quick start (5 min), good-first-issue guide, architecture table, DCO, security report path.

0.60.0 – 2026-03-07

Added

  • OpenSSF Best Practices passing badge (100% criteria met)
  • ClusterFuzzLite integration for continuous fuzz testing
  • ARCHITECTURE.md for contributor orientation
  • SAST CWE map expanded from 10 → 52 entries
  • 11 GitHub area labels for issue triage
  • Good-first-issue and help-wanted issues for new contributors

Changed

  • Split OpenClaw monolith SKILL.md into 4 focused skills (scan, compliance, registry, runtime)
  • Reduced SKILL.md surface area for better OpenClaw trust score

Fixed

  • ResponseInspector detector for cloaking and payload detection (wired into proxy)

0.59.3 – 2026-03-07

Added

  • Redesigned SVG diagrams: architecture, blast radius, topology, scan pipeline, compliance heatmap — dark and light variants
  • Animated demo SVG
  • Simplified README
  • Dashboard charts and compliance matrix (Next.js)
  • Grafana template
  • MkDocs documentation site
  • GitHub issue and PR templates
  • Roadmap accuracy fixes

Fixed

  • Credential pattern detection improvements
  • Policy conditions corrected to 17

0.59.2 – 2026-03-07

Added

  • Value-based credential scanning
  • Smithery/ClawHub/MCP Registry retry logic
  • Fleet scan and CIS benchmark input bounds
  • NVD cache 90-day TTL
  • Docker health check hardening
  • ClawHub SKILL.md trust transparency

0.59.1 – 2026-03-06

Added

  • MCP tool reachability and capability scanning via mcp_introspect.py
  • VEX (OpenVEX) support: load, generate, apply, export
  • SBOM ingest: CycloneDX 1.x and SPDX 2.x/3.0 JSON
  • OpenSSF Scorecard → risk boost integration
  • OTel trace ingestion and CVE cross-reference

0.59.0 – 2026-03-06

Added

  • Runtime proxy with 7 detectors: ToolDrift, ArgumentAnalyzer, CredentialLeak, RateLimit, SequenceAnalyzer, ResponseInspector, VectorDBInjectionDetector
  • Snowflake Native App integration
  • Fleet scan across multiple agent inventories
  • CIS benchmark checks (AWS)
  • Policy-as-code: 17 conditions (16 declarative + expression engine)
  • Prometheus metrics export from proxy
  • JSONL audit log from proxy