Changelog
Release history for governance-sdk. Follow on GitHub for updates.
v0.11.2latest2026-04-16
README sync automation — keeps npm README in lockstep with repo-root README
featscripts/sync-readme.mjs generates packages/governance/README.md from root README
featWired into prepublishOnly — every npm release ships an in-sync README
featCI guard fails the build if a manual README edit skips the sync
docsNo code changes. SDK behavior identical to 0.11.0.
v0.11.0stable2026-04-15
Scope honesty pass 2 — removes 5 unused modules, demotes 4 oversold ones. 1,328 tests, 0 failures.
breakRemoved governance-sdk/eval-trace, /eval-scorer, /eval-types and gov.eval — use a dedicated eval harness (inspect-ai, PyRIT, Phoenix, Langfuse) and route results to gov.audit.log()
breakRemoved governance-sdk/plugins/mcp-annotations — annotation-rule generator was a static template, not runtime governance
breakRemoved governance-sdk/supply-chain-sbom — use /supply-chain-cyclonedx (CycloneDX 1.5) instead
breakRemoved GovernMCPConfig.traceCollector — tool-call audit still fires via gov.audit
docsDemoted framing on metrics, otel-hooks, action-recorder, behavioral-scorer — they ship but are no longer headlined as built-in observability / dynamic-trust
docsReal OpenInference OTel exporter and TrustEngine promotion queued for future releases
test1,328 tests, 0 failures
perf49 → 44 export paths after dead-module removal
v0.10.02026-04-02
Scope honesty pass 1 — tightened public surface ahead of 1.0
docsSee packages/governance/CHANGELOG.md for the full list of cuts and demotions
v0.1.0 — 0.9.x2026-01-15 → 2026-03-31
Early iteration — core policy engine, framework adapters, audit chain, injection detection, scoring
featcreateGovernance() — core governance instance with rule evaluation
featFramework adapters: Mastra, Vercel AI SDK, LangChain, OpenAI Agents, Anthropic, Genkit, LlamaIndex, Mistral, Ollama + MCP & Bedrock
feataudit-integrity module — HMAC-SHA256 hash-chained audit with chain.verify() and chain.export()
featdetectInjection() — 54 regex patterns across 7 categories (F1 = 0.48 on 6,931-sample LIB benchmark)
featKill switch at priority 999 — kill(), killAll(), revive(), reviveAll() (per-process)
feat7-dimension governance scoring — composite 0–100 score, L0–L4 levels
featEU AI Act + ISO 42001 + NIST AI RMF self-assessment mappings
featPostgreSQL storage adapter via governance-sdk-platform (auto-migration)
featTypeScript-native, zero runtime dependencies, MIT license
docsPer-release notes at packages/governance/CHANGELOG.md on GitHub
Stay up to date
Watch releases on GitHub Releases or subscribe to the npm feed.