📦 Release Notes

Full version history for Agenthood.
Generated automatically — do not edit manually.

---

🐛 Bug Fixes

• Milestones: leave CHANGELOG.md and release notes to semantic-release

---

✨ Features

• Skills: merge Copilot review, readme, api-doc, and unit-test patterns into existing members (#449)
• Skills: port bug-fix-teammate, cleanup-specialist, pull-request-assistant from Copilot library (#447)
• Skills: port concept-explainer, issue-manager, onboarding-plan from Copilot library (#445)
• Skills: port implementation-planner, debugging-tutor, accessibility-auditor from Copilot library (#446)
• Skills: port remember skill from deepagents (#448)

---

🐛 Bug Fixes

• resolve auditor follow-up warnings for scripts and delegation (#442)

---

🐛 Bug Fixes

• Agents: align docs with memory.write scoping, unify constructors, delegate exit to CLI
• Agents: align member tool grants with permission docs and close delegation boundaries
• Agents: apply review findings on trust boundaries and redaction
• Agents: close user_query injection, raise decision id entropy, align DeveloperAgent style
• Agents: delimit retrieved and project context as untrusted data
• Agents: escape retrieved context, redact traces by default, fail gate on ambiguous verdicts
• Agents: fail redaction closed, trim advertised tools, and delegate run errors to the CLI
• Agents: forward residualMemory through agent option bags
• Agents: restore user_query directive and close remaining trust-boundary escapes
• Agents: restrict delegation to read-only roles
• Agents: share redaction helpers, label prompt boundaries, and flatten the review gate
• Agents: share user_query wrapping, opt-in delegation, and redact sentry payloads
• Ci: harden decision gate and test-changed script against review findings
• Members: fail closed on empty tools and gate delegation for restricted members
• Reasoning: bound ReActLoop iterations with a maxSteps guard
• Scripts: escape academy site title
• Scripts: guard flag-like test paths and signal-death exits

✨ Features

• Observability: keep background-failure console visibility without a DSN

---

🐛 Bug Fixes

• Agents: restore Oracle model attribution on failures after runWithExecutor refactor (#436)

---

🐛 Bug Fixes

• Agents: emit trace envelope from OracleAgent.run
• Commands: match persisted pattern prefixes in status --learner
• Observability: correct health tracer probe and add config-dependent checks
• Observability: redact decision and provenance payloads and align envelope hashing

✨ Features

• Cli: add replay evaluation mode to eval command
• Cli: surface anomaly alerts in status --alerts
• Config: scaffold observability block in init config and make trace path configurable
• Core: accumulate tool-level LLM usage into trace token counts
• Evals: add EmbeddingIndex with ANN similarity search and upsert persistence
• Evals: add versioned re-index migration for legacy zero-vector patterns
• Evals: query embedding index before hash fallback in EpisodeLearner
• Observability: allow source override through ExecutionContext
• Observability: wire AnomalyDetector into trace flush with alert persistence
• Runtime: inject EpisodeLearner with EmbeddingIndex into agent construction

---

✨ Features

• Observability: add optional sentry error reporting
• Observability: expose episode learner learning status

---

✨ Features

• Observability: add health check command and API

---

✨ Features

• Observability: stamp traces with eval baseline quality

---

✨ Features

• Observability: add anomaly detection for cost and quality
• Observability: add trace retention and export policy

---

✨ Features

• Observability: add redaction filter for trace payloads

---

✨ Features

• Evals: add replay evaluator for behavior drift

---

✨ Features

• Cli: add eval command with baseline regression gating

---

✨ Features

• Evals: add baseline comparison for eval reports

---

✨ Features

• Evals: add eval runner with llm-as-judge scoring

---

🐛 Bug Fixes

• Observability: address review findings on trace pipeline

✨ Features

• Core: add OpenCode Go model pricing to cost estimator
• Metrics: per-member cost and quality summaries

---

✨ Features

• Cli: add trace command — npx agenthood trace

---

✨ Features

• Observability: add workflow and session correlation IDs

---

✨ Features

• Observability: persist traces to a queryable store

---

✨ Features

• Evals: define eval suite format with Ajv validation

---

✨ Features

• Core: implement TokenCounter and CostEstimator #297

---

🐛 Bug Fixes

• Deps: bump nanoid to 3.3.18 to clear auditor gate

✨ Features

• Observability: emit trace envelope and in-memory ring-buffer tracer #295

---

🐛 Bug Fixes

• Ci: harden doorman gate, extract audit check, and dedupe workflow boilerplate
• Ci: run setup-env composite after checkout in pr.yml

✨ Features

• Ci: enforce PR descriptions link to an issue via doorman gate, closes #N #N

---

🐛 Bug Fixes

• Ci: drop flag-terminator in pr comment helper, reject reflog refnames
• Ci: indent heredoc comment bodies to keep workflow YAML valid
• Ci: make new check scripts executable, document heredoc safety
• Ci: remove empty expression literal from workflow comment
• Cli: allowlist --provider values, drop unused go import
• Cli: component-level lock refname rule, case-insensitive per check-ref-format
• Cli: execFileSync for all pr-sync git/gh calls
• Cli: neutral skills-dir module, strict config parse, init failure exit
• Cli: refname validation mirrors git check-ref-format
• Cli: sanitize pr-sync marker shas and private temp files
• Cli: strict --pr parsing, untrusted prompt marker, sanitized key echo, diff-failure gate
• Cli: validate lockfile keys and use execFileSync in rollback
• Init: eject removes only member subdirs, never foreign skills
• Llm: friendly ollama connect errors and permanent 400 classification
• Llm: groq fails fast on missing key, validate chain primary, setup key guidance
• Llm: strip sampling extras for opencode-go to avoid upstream 400
• Verify: run lane checks before lockfile write; harden CI scripts per review

✨ Features

• Init: dry-run flag and agenthood-aware eject of skills dirs
• Verify: real lane-overlap checks for --strict

---

🐛 Bug Fixes

• Ci: severity-filter audit gate, three-dot diffs, revert pass-through, pin vsce

---

🐛 Bug Fixes

• Ci: anchor herald verdict updates to bot marker
• Ci: drop eval from herald summary, add issues permission, escape names
• Ci: grant actions read, dedupe herald concurrency, add script tests
• Ci: herald checkout step, full markdown escaping, scoped trials, multi-PR
• Ci: restore herald summary via inline github-script input

---

🐛 Bug Fixes

• Ci: exempt nested lockfiles from size limits, survive SIGPIPE in diff caps
• Ci: reuse agent-analysis action and harden verdict parsing
• Hooks: deduplicate doorman banner, POSIX-safe checks, lock vs extension deps
• Hooks: support breaking changes and revert commits, make pre-push advisory

✨ Features

• Agents: record decision and provenance per member run
• Memory: add causal chains and provenance store for member decisions
• Memory: add precedent search and society graph snapshots

---

🐛 Bug Fixes

• Packaging: ship scripts/ so postinstall resolves (#392)

---

🐛 Bug Fixes

• Rag: cache a tree-sitter parser per language (#390)

---

🐛 Bug Fixes

• Ci: enforce audit, declare commitlint, add build gate (#386)

---

🐛 Bug Fixes

• Lint: resolve lint errors and add ci gates
• Llm: clamp retry-after to 300 seconds
• Llm: guard NaN retry-after header in rate limit errors

---

🐛 Bug Fixes

• Ci: byte-truncate review diff to stay within argument limits
• Ci: guard member dir glob in librarian check
• Ci: ignore lifecycle scripts in agent analysis install
• Ci: mark truncated diffs in review prompt
• Ci: require RANGE env in agent analysis script
• Ci: review first push of a branch against empty tree
• Ci: truncate review diff at line boundaries

---

🐛 Bug Fixes

• Ci: fail agent analysis on missing or malformed decision block
• Ci: isolate agent analysis temp files per run

---

🐛 Bug Fixes

• Ci: always build before agent analysis runs
• Docs: point member skill references to skills canonical home
• Docs: update member counts and skill links after canonicalization
• Marketplace: add new members to agenthood-all bundle and update counts
• Marketplace: align copy with 19 members
• Members: satisfy sentinel and librarian checks for the-builder
• Runtime: route skill file changes to oracle and sentinel triggers
• Runtime: watch skills dir for operator drift and test sentinel pattern

✨ Features

• Members: add builder member
• Members: bring the-builder to full society standards
• Runtime: add the-builder to member triggers

---

🐛 Bug Fixes

• Pr: address reviewer and warden findings on PR #371

✨ Features

• Llm: add OpenRouter provider

---

🐛 Bug Fixes

• Ci: point member structure checks at canonical skills/ source
• Members: make skills/ the single source of truth for member SKILL.md
• Members: make tool tier construction order-independent
• Project: scope supersedes regex to its section
• Security: replace execSync postinstall eval and drop esbuild allowScripts

---

🐛 Bug Fixes

• Cli: add missing run command to COMMANDS map
• Cli: fix flag parsing and status member count
• Status: remove readMetrics duplication, restore MetricsCollector with centralized usage

✨ Features

• Commands: add adapter for MetricsCollector to isolate infrastructure dependency

---

✨ Features

• add The Mailman — 17th Society member for delivery and cross-posting
• Init: strip junk files from init (#360)

---

🐛 Bug Fixes

• Security: sanitize error logging and pin dep versions

---

✨ Features

• fix vector store crash, seed during init, add semantic pattern matcher #312 #354 #312

---

🐛 Bug Fixes

• add root commitlint.config.ts for repo CI
• Init: resolve 11 failing health checks by correcting source paths and workflow
• pin commitlint versions, tighten CI perms, split check, extract stripConfig
• Struct: sort workflow entries alphabetically in STRUCTURE.md

---

🐛 Bug Fixes

• Groq: add error mapping, shared stream utils, and OpenAIProvider validation
• Groq: resolve all Auditor, Warden, and Reviewer findings in GroqProvider
• Providers: restore custom tool call handling in shared parseToolCall

---

🐛 Bug Fixes

• Ci: track .agenthood/config.json for CI agent analysis

✨ Features

• Skills: add 16 platform integration skills (CLI-focused)

---

🐛 Bug Fixes

• Ci: update Sentinel to warn on clear-named skills, remove bridge directories
• Skills: add bridge docs/members/ dirs for clear-named skills

✨ Features

• Skills: add clear-named skill mirrors and shared reference checklists

---

🐛 Bug Fixes

• Core: address Warden findings — extract memberLore, refactor run.ts, deduplicate constants, add injection guards
• Core: address Warden findings round 3 — indentation, dead code, boolean naming, nesting, unused dep
• Core: address Warden/Auditor/Reviewer findings round 2

✨ Features

• Core: rename src/skills to src/tools, add skill discovery infrastructure

---

🐛 Bug Fixes

• Groq: update default model from decommissioned llama-3.1-70b to llama-3.3-70b-versatile
• Groq: update test expectation to match default model llama-3.3-70b-versatile

---

🐛 Bug Fixes

• Ollama: read baseUrl from config before falling back to env

---

🐛 Bug Fixes

• align 3 remaining stale doc references
• update all project-wide references after moving doc dirs to docs/

---

🐛 Bug Fixes

• Ci: add explicit return 0 in decision function and guard call
• Ci: add explicit return 0 in else branch for consistency
• Ci: add explicit return 0 to check_agenthood_decision
• Ci: address auditor findings — test-runner flag injection, npm ci, npm audit, temp paths
• Ci: address auditor high and medium findings
• Ci: address review findings — setup-node in action, smart tester, extract scripts, pin tools
• Ci: address reviewer and warden findings
• Ci: address warden stale-comment quoting, temp_dir casing, revert vscode npm ci
• Ci: align blocking detection in reviewer.yml with line-start grep pattern
• Ci: anchor blocking regex, ignore-scripts, extract helpers
• Ci: anchor blocking=true grep to require closing -->
• Ci: change AGENTHOOD_DECISION format to avoid false-positive true|false
• Ci: export OPENCODE_API_KEY, add setup-node cache, smart test selection
• Ci: fail workflows on blocking findings, restore api-key, fix registry paths, add execute permission
• Ci: install commitlint packages locally for config resolution
• Ci: match [blocking] in summary table cells not line start for reviewer
• Ci: narrow stale-comment matching, remove awk guard
• Ci: prevent set -e from killing script on CLI failure
• Ci: resolve commitlint tsx resolution and librarian false-positive
• Ci: resolve TASK unbound variable from subshell scoping and commitlint tsx resolution
• Ci: restrict blocking check to table rows only
• Ci: revert stale_previous_comment --arg in jq filter, add pipefail guard
• Ci: tighten blocking detection to line-start grep, bump header-max-length to 150
• Ci: use AGENTHOOD_DECISION for stale comment matching

✨ Features

• Ci: mark previous agent PR comments as outdated instead of deleting
• Ci: use structured decision block from LLM for blocking detection

---

🐛 Bug Fixes

• Evals: import ExecutionContext from correct module
• Evals: import ExecutionContext from correct module in tests

✨ Features

• Evals: implement EpisodeLearner — update LongTermMemory and ResidualMemory from eval scores

---

🐛 Bug Fixes

• correct vitest JSON field names (num prefix), remove duplicate detect block in run.ts

✨ Features

• Core: implement metrics collector, status --watch/--json/--drift, quality gates drift detection and config

---

🐛 Bug Fixes

• address all review findings across 7 phase branches
• remove hardcoded member count from steward readme (maintenance trap)
• steward pre-load count should be 15 (16 members - 1 for steward itself)
• sync skills/ directory with members/ for ci compliance
• use 16 not 15 for steward pre-load count (matches registry)

✨ Features

• Phase: phase 0 - decision log, postmortem, auto-discover #280 #114
• Phase: phase 1 - protocol interfaces and workflow engine #116
• Phase: phase 2 - workflow checkpoint and goal chain #118
• Phase: phase 3 - oracle, strategist, and operator agents #277 #278
• Phase: phase 4 - verify, rollback, and status commands #276 #281
• Phase: phase 5 - diff impact analyzer and quality gates #282
• Phase: phase 6 - review-pr workflow end-to-end

---

🐛 Bug Fixes

• Docs: address Reviewer findings on init check count and LanceDBStore API example
• Docs: correct check count to 21 and fix insert->add API example
• Docs: fix mentioned shipped version
• Init,check: align init ceremony with health check expectations
• Skills: add output format section to the-reviewer SKILL.md for consistent rendering
• sync skills/the-reviewer/SKILL.md with members/ changes
• The-reviewer: address review findings on output format and README
• The-reviewer: flatten heading hierarchy and add intra-section spacing example
• The-reviewer: use [SEVERITY] placeholder and move meta-instruction outside template

✨ Features

• load .env file automatically via dotenv

---

🐛 Bug Fixes

• address PR review feedback and update docs
• Ci: add GITHUB_TOKEN to Run Reviewer step
• Ci: address reviewer feedback on gh pr view error handling
• Ci: convert skills/ symlinks to regular files
• Ci: ensure all gh commands have GITHUB_TOKEN auth
• Ci: fix YAML indentation in sentinel, auditor, warden workflows
• Ci: install gitleaks binary before pre-check step
• Ci: remove noisy gitleaks pre-check step
• Ci: update sentinel to check file content instead of symlinks
• Cli: wire detect flag through CLI parser
• implement all review findings from architect and reviewer
• Security: address all Auditor findings from PR #285

📝 Documentation

• Governance: create member RACI map and release policy

✨ Features

• Ci: make API usage smart and economic
• Llm: add OpenCode Go provider
• Llm: add OpenCode Zen provider
• Llm: fix OpenCode provider for DeepSeek tool format compatibility
• Orchestration: implement MemberOrchestrator detection
• Rag: implement AgenticRAG with RetrievalDecisionSkill
• Rag: implement HierarchicalChunkStrategy with parent-child chunking

🔹 BREAKING CHANGES

• Governance: announcements, deprecation policy, compliance

---

🐛 Bug Fixes

• Memory: align ProjectMemoryImpl return types with ProjectMemory interface

---

🐛 Bug Fixes

• Deps: pin tree-sitter-go and tree-sitter-python to v0.23.x to resolve peer dependency conflict

✨ Features

• Memory: implement PersonalisationStore for per-project agent adaptation, closes hi#weight #112
• Memory: implement ShortTerm, LongTerm, Episodic, and Project memory tiers
• Rag: implement baseline RAG pipeline — ChunkStrategy, Indexer, Retriever
• Rag: implement SocietyIndexer for members, ADRs, and conventions
• Rag: implement TreeSitterParser for AST-based code structure extraction

---

🐛 Bug Fixes

• address reviewer findings and update Phase 0 docs
• ignore entire .agenthood/ directory except config.example.json
• Llm: extract and granularize api key validation

✨ Features

• Core: move schema validator to core and harden error messages
• Memory: implement LanceDB vector store with IVectorStore interface
• Memory: implement memory governance with IMemoryStore and InMemoryStore
• Memory: implement ResidualMemory — decay-weighted trace signals
• Rag: implement KnowledgeGraphStore for relationship-aware retrieval
• Reasoning: add infinite loop detection to reactloop

---

🐛 Bug Fixes

• Workflows: correct yaml indentation in member attribution comments (#260)

---

🐛 Bug Fixes

• Academy: compute relative link from non-index pages at correct depth
• Failover: add embed model downgrade, 3-attempt backoff, JSDoc, align with spec
• Failover: trip permanent errors immediately, add model downgrade to stream()

✨ Features

• Cli: add provider selection logging, runtime guide, and failover integration tests
• Cli: wire provider failover config into CLI and LLMRouter
• Provider: implement model downgrade and circuit breaker config

---

🐛 Bug Fixes

• Academy: compute relative links from file dir instead of docs root

✨ Features

• Academy: replace MkDocs with Node.js build and deploy to GitHub Pages

---

🐛 Bug Fixes

• Ci: add npm ci step to gh-pages workflow before building, closes #pages
• Config: update stale commitlint.config.cjs references to .ts
• Llm: make provider SDK imports lazy, lower engines.node to 22.14.0

✨ Features

• Academy: replace MkDocs with Node.js build and deploy to GitHub Pages

---

🐛 Bug Fixes

• Agents: add missing contextCompressor property declaration
• Ci: add build step before pr-sync in The Manuscript workflow
• Ci: fail gracefully on push events and split workflow triggers
• Ci: use node dist/cli.js instead of npx to avoid permission denied
• Pr-sync: use PR head SHA and preserve existing PR body
• Providers: handle missing GROQ_API_KEY in GroqProvider constructor
• remove dead ContextCompressor import that breaks build
• remove duplicated docs
• Tests: correct Artifact interface usage and add JSON.parse error handling
• Test: update commitlint test to import .ts config directly
• Vscode-extension: move test config to src/ for 100% typescript compilation

🔹 chore

• Release: mark v2.0.0 breaking changes

✨ Features

• add GroqProvider tests, schema validation, and runtime documentation
• Agent: implement ArchitectAgent, ReviewerAgent, QAAgent runtime classes
• Commands: add pr-sync command, PrSyncSkill, and The Manuscript workflow, closes #based
• Core: add concurrency queue and safety guard
• Core: add RiskManager, SkillRegistry discovery, dynamic routing, and README rewrite #162 #102 #102 #103 #162
• Core: implement ContextCompressor with token-aware memory summarization
• Core: security hardening — Ajv, API key validation, symlink checks
• Llm: add Anthropic prompt caching with cache control breakpoint
• Llm: implement ProviderFailover for resilience (#161)
• Llm: provider failover with circuit breaker and per-member preferences
• Members: wire all 14 society members to agenthood run
• Reasoning: implement ContextCompressor for token management (#104)
• Release: generate user-friendly release notes via @semantic-release/exec
• Runtime: release v2.0.0 — TypeScript runtime with autonomous agent execution
• ship M4 foundation - TypeScript runtime with providers, agents, skills, and CLI
• Skills: export and register SubagentTaskSkill with delegate_task name #8 #9
• Skills: implement SubagentTaskSkill for agent delegation (#199)
• Skills: replace stub skills with real LLM and filesystem implementations
• Workflow: replace commit listing with LLM code review by The Reviewer

🔹 BREAKING CHANGES

• Release: The Manuscript PR body sync is replaced by The Reviewer commit review. The Python runtime and runtime/ directory are removed. The Society now runs exclusively on the TypeScript runtime with Groq as the default provider.

---

✨ Features

• Skills: add skills/ symlinks for all 14 members and Sentinel validation

---

🐛 Bug Fixes

• Academy: remove source CNAME to prevent gh-pages redirect loop, closes #pages

---

🐛 Bug Fixes

• Distribution: add owner email and align version with repo release v1.8.4
• Dot-folders: audit and repair githooks, devcontainer, gitignore, and stale dirs

✨ Features

• Distribution: add .claude-plugin marketplace.json for Claude Code plugin discovery
• Workflows: add Herald CI summary workflow that posts PR verdict comment

---

🐛 Bug Fixes

• Academy: revert GitHub Pages custom domain config (#194)

---

🐛 Bug Fixes

• Academy: move CNAME to docs root for GitHub Pages (#191)

---

🐛 Bug Fixes

• Academy: quote ADR nav title to fix YAML syntax (#187)

---

🐛 Bug Fixes

• Academy: resolve ADR rendering and broken cross-links (#186)

---

✨ Features

• Registry: submit Agenthood to SkillsMP and Skills.sh (#184)

---

🐛 Bug Fixes

• Skill: normalize SKILL.md structure for milestone M1 (#183)

---

🐛 Bug Fixes

• Docs: correct broken ADR-010 references and Academy CTA URLs (#180)

---

✨ Features

• Npm: improve package visibility with better keywords and badges

---

🐛 Bug Fixes

• Ci: remove registry-url and upgrade to Node 22 for npm OIDC
• Ci: switch to OIDC trusted publisher for npm publishing

---

🐛 Bug Fixes

• Ci: restore npm auth wiring for semantic-release (#164)
• Release: enable npm publishing in semantic-release (#146)

---

🐛 Bug Fixes

• Release: add semantic-release git plugin for changelog commits
• Release: enable npm publishing in semantic-release configuration

---

🐛 Bug Fixes

• Docs: restore missing changelog entries for v1.1.0-v1.2.1 (#83)

---

🐛 Bug Fixes

• Release: wire up npm publishing pipeline (#82)

---

✨ Features

• add integration test framework and improve TypeScript setup (#64)

---

✨ Features

• Vscode: implement workspace event bus for passive observation (#62)

---

🐛 Bug Fixes

• Security: remove embedded credential examples from docs (#61)

---

🐛 Bug Fixes

• Release: drop @semantic-release/git plugin (#55)

✨ Features

• Runtime: bootstrap Python package and 14-member registry (#51)
• Vscode: modernize with build, tests, and CI (#54)

---

🐛 Bug Fixes

• Release: remove prepublishonly script (#40)

---

🐛 Bug Fixes

• Ci: run npm ci before semantic-release to satisfy prepublishOnly (#38)
• Ci: use npm install instead of npm ci (no lockfile) (#39)

---

🐛 Bug Fixes

• Ci: pass NPM_TOKEN to semantic-release and install npm plugin (#37)
• Release: enable npm publish now that NPM_TOKEN is configured (#36)

---

🐛 Bug Fixes

• Agents: update stale member count from 13 to 14
• Check: validate all 14 members in health check (#27) #26
• Ci: add ADR presence check to librarian.yml (#20)
• Ci: add AGENTS.md to sentinel.yml trigger paths (#19)
• Ci: fix sentinel multi-word section checks
• Ci: use commitlint.config.cjs for esm compat
• Conventions: add vague-subject rule to commitlint config
• Docs: correct member count to fourteen
• Gitmessage: replace project-specific scope examples with generic placeholders
• Portals: create missing linear.md and jira.md connector docs (#28)
• Release: disable npm publish until NPM_TOKEN is configured (#33)

✨ Features

• Adr: create foundational ADRs for Agenthood's own architecture (#30)
• Agentic-workflows: clarify workflow files as manual-prompt templates (#31)
• Bootstrap: add .agenthood/config.example.json reference template
• Bootstrap: implement agenthood setup command and init CLI (#23)
• Doorman: add pre-push hook blocking direct push to main
• Hooks: add commit-msg hook
• Hooks: add pre-commit hook
• Members: add branch scope and PR scope validation to architect and doorman
• Members: add N+1 commit pattern and PR granularity to the-scribe
• Members: add the-envoy
• Members: add the-oracle
• Members: add the-sentinel
• Members: add the-steward
• Members: add the-warden
• Members: register the-oracle and the-envoy in indexes
• Members: register the-sentinel and the-warden in indexes
• Platform: add npm package, VS Code extension, portals rename, and INITIATION
• Setup: add setup.sh, makefile, devcontainer
• Society: add skill files, rituals, agentic workflows, CI, and intelligence