📦 Release Notes

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

---

🐛 Bug Fixes

• Opencode: support bare plugin specifier without killing host (#991), refs #990

---

✨ Features

• Goals: add goal command backed by GoalChain (#987)
• Runtime: arbitrate member runs through ConcurrencyQueue by origin (#986)
• Trace: persist one-step trajectory on run success and failure (#985)

---

🐛 Bug Fixes

• Ci: add sync-skills validation to prevent stale generated artifacts (#961)
• Tests: achieve full typecheck coverage for tests/ directory (#963)
• Tests: address remaining typecheck errors in test files (#963)
• Tests: address remaining typecheck errors with type assertions (#963)
• Tests: align 7 test files with current source API types (#963)
• Tests: fix shared helpers and cascade mocks for typecheck coverage (#963)
• Tests: resolve opencode-plugin ChildProcess type mismatches (#963)
• Tests: restore spy mocks and vitest import, clear typecheck gate (#976), refs #963
• Tests: revert opencode-plugin type casts to as any (#963)

---

🐛 Bug Fixes

• Ci: the publish job must clean release notes before tagging

---

🐛 Bug Fixes

• Ci: the bare-ref gate rejects cleanLine's own output (#971), refs #969, #663, #967
• Herald: clean generated release notes before gating and publishing (#969), refs #663, #968
• Herald: the release gate missed the linkified closing keyword (#968), refs #663, #967

✨ Features

• Members: add Jev-inspired confidence patterns to mediator and steward (#935)

---

✨ Features

• Routing: validate the confidence-gated cascade as a record (#958)
• Skills: budget the context window at skill activation (#966), refs #663

---

🐛 Bug Fixes

• Dependabot: stop the minor-patch group from swallowing 0.x majors, and land the four safe bumps (#948), refs #945, #946, #900, #947
• Herald: dedupe trailing ref pile in the changelog too (#956), refs #949, #948, #946, #900, #947, #945, #1, #953, #954
• Herald: release notes stop leaking issue refs, changelog stops claiming false closures (#951), refs #949, #948, #946, #900, #947, #945, #1, #950
• Herald: stop linkify artifacts corrupting release notes (#954), refs #951, #953, #950

---

🐛 Bug Fixes

• Test: raise the vitest timeout to fix the class, not the symptom (#934), refs #927, #914
• Vscode-extension: hold @types/vscode at the engines.vscode floor (#927), refs #914

---

🐛 Bug Fixes

• Deps-dev: bump vitest to 5 with the coverage provider and hoist nested mocks (#925), refs #920

---

🐛 Bug Fixes

• Deps: pin tree-sitter-go to 0.23.4 so npm ci resolves without ERESOLVE (#911), refs #910
• Deps: restore optional peers dropped from the lockfile by #916 (#924), refs #910

✨ Features

• v3.68 Convert & Trust — onboarding, fail-closed skill trust, cost observability (#913), refs #574, #619, #620, #578, #152, #151, #514, #515, #604, #606, #624, #625, #651, #465

---

✨ Features

• Health: probe providers with a real request (#905)
• Llm: configurable per-request timeout for provider calls (#903)
• Memory: add ttl option to short-term memory, prune residual signals on record (#906)
• Safety: enforce allowed-tools from SKILL.md frontmatter (#904)

---

🐛 Bug Fixes

• Ci: reviewer empty-tree range (#815) (#888)
• Docs: correct non-existent member references and tool vocabulary (#897), refs #790

✨ Features

• Cli: --json flag with pino structured logging (#647) (#892), refs #646
• Cli: add shared CLI error formatter with consistent exit codes (#891), refs #646
• Create: scaffold new skills from a template (#879), refs #602
• Diff: show drift between installed members and lockfile (#876), refs #664
• Eject: sweep lockfiles alongside artifacts (#874), refs #650
• Init: add --ci for non-interactive setup with --runtime and --members (#882), refs #673, #761
• Init: add --target flag for cross-client compatibility (#662) (#895)
• Init: create .agents/skills/ as primary cross-client location (#654) (#894)
• Init: prompt before overwriting an existing setup (#873), refs #643
• List: show token counts and context budget (#875), refs #657
• Remove: clean uninstall with lock pruning (#877), refs #590
• Run: add --sandbox flag with strict local profile (#880), refs #665
• Sandbox: add Docker container isolation detection (#884) (#896)
• Skills: discover skills in bare skills/ directory (#642) (#893)
• Upgrade: self-upgrade agenthood with config backup (#881), refs #667

---

🐛 Bug Fixes

• Llm: wrap tools in OpenAI function shape for chat-completions providers (#860)

---

🐛 Bug Fixes

• Ci: poll npm registry instead of fixed sleep in release verify (#857), refs #856

---

🐛 Bug Fixes

• Ci: gate ritual matrix entries in a step instead of job if (#855), refs #854

✨ Features

• Eval: cross-provider comparison with repeatable --provider flags (#849)

---

🐛 Bug Fixes

• Ci: cross-check blocking verdicts against listed findings (#851)

---

🐛 Bug Fixes

• Llm: runtime stream guard and structured provider logging (#847), refs #323, #322

---

🐛 Bug Fixes

• Skills: harden agenthood install against SSRF and path traversal (#843), refs #833, #842

---

✨ Features

• Skills: include packaged tool skills in runtime catalog (#833), refs #832

---

🐛 Bug Fixes

• Verify: narrow placeholder scan to marker form and skip code spans (#830), refs #816, #753

---

🐛 Bug Fixes

• Scripts: drop removed marked options for v15/v18 compat (#824)

---

🐛 Bug Fixes

• Release: route herald compute through a file, keep stdout json-only
• Release: untrack the herald notes artifact, keep the changelog-derived view

---

🐛 Bug Fixes

• Cost: add missing afterEach import and update docs
• Cost: remove unused vars, fix require(), simplify tests
• Release: keep semantic-release peer and install before pending check
• Vitest: exclude cost command from coverage gate

✨ Features

• Cli: add cost command with per-provider/per-member breakdown (#676)
• Cli: add doctor command for comprehensive diagnostics
• Cli: add rotate-key command for API key rotation with hot-reload (#678) (#784)
• Rituals: implement ritual command and scheduled rituals workflow (#794)
• Skills: add integrity verification to SkillDiscovery on load (#658) (#785)
• Verify: add --conflicts flag for overlapping skill-description detection

---

🐛 Bug Fixes

• Deps: bump js-yaml 4.3.1 → 4.3.2 in vscode-extension (#781)

✨ Features

• Evals: add description optimization command (#584) (#779)

---

✨ Features

• Evals: add blind A/B comparison with significance testing (#778), refs #558
• Members: validate skill output against declared format pattern (#774), refs #594

---

✨ Features

• Evals: add skill regression testing with iteration tracking and convergence detection (#775), refs #562

---

🐛 Bug Fixes

• Packaging: export dist/checkpoint/RunCheckpoint.js for host CheckpointStore use (#772), refs #771

---

🐛 Bug Fixes

• Runtime: resume restores loop state and answers the pending ask_human call (#768), refs #767

---

✨ Features

• Runtime: accept an injectable CheckpointStore for host-managed persistence (#766), refs #764

---

✨ Features

• Evals: add assertion-based output grading to the eval harness (#763), refs #559

---

🐛 Bug Fixes

• Ci: tolerate whitespace before the decision marker's --> (#757), refs #755, #756

---

🐛 Bug Fixes

• Llm: import chain providers concurrently to stop cold-start timeouts (#754), refs #465

---

🐛 Bug Fixes

• Verify: unify member-path resolution and add a lockfile CI integrity gate (#752), refs #740

---

🐛 Bug Fixes

• React-loop: emit tool.approval instead of a duplicate tool.called (#751), refs #748

---

🐛 Bug Fixes

• Skills: strip UTF-8 BOM so frontmatter parses on Windows-authored files (#750), refs #563

---

🐛 Bug Fixes

• Exports: re-expose ./dist/llm and ./dist/core for programmatic consumers (#749), refs #637, #739

---

✨ Features

• Skill: validate SKILL.md against the agentskills.io spec (#746), refs #634

---

🐛 Bug Fixes

• Skill: converge lockfile regeneration and make --update-lock actually work (#745), refs #635

---

🐛 Bug Fixes

• Core: graceful SIGINT shutdown for long-running agent sessions (#744), refs #633

---

🐛 Bug Fixes

• Security: delimit untrusted task and tool output against prompt injection (#743), refs #632

---

🐛 Bug Fixes

• Security: broaden SafetyGuard catastrophic command patterns (#742), refs #637

---

🐛 Bug Fixes

• Llm: redact secrets from outbound provider requests (#741), refs #631

---

🐛 Bug Fixes

• send x-opencode-session header for all OpenCode requests (#738)

---

🐛 Bug Fixes

• add x-opencode-session header for OpenCode Go tier (#736)

---

✨ Features

• Registry: add SkillRegistryClient, search, and upgrade commands (#729)

---

✨ Features

• Provisioning: add RemoteSkillSource for URL/git fetching with caching (#728)
• Trajectory: add TrajectoryStore and trace visualize/diff subcommands (#727)

---

✨ Features

• Interactive: add --interactive flag for human-in-the-loop confirmation (#726)

---

✨ Features

• Logs: add --tail and --follow flags to log command (#725)

---

✨ Features

• Cli: add agenthood completion command for bash/zsh/fish (#724)

---

✨ Features

• Debug: add --debug flag for LLM request/response logging (#723)

---

✨ Features

• Checkpoint: add run checkpoint persistence and --resume flag (#722)

---

🐛 Bug Fixes

• Install: use execFileSync to prevent shell injection

✨ Features

• Cli: add publish command and --dry-run for install

---

🐛 Bug Fixes

• Mcp: extract handlers, pin dep, update docs

✨ Features

• Mcp: add MCP server to expose skills as tools

---

✨ Features

• Cli: add install command for skills from URLs and git repos
• Skills: add quality tiers (official, community, experimental) with visual indicators

---

✨ Features

• Config: switch default model from deepseek-v4-flash to mimo-v2.5 (#686)

---

✨ Features

• Ci: dedupe analysis comments and post inline review findings (#684)

---

✨ Features

• Plugin: add official opencode plugin entry with member-run tool (#629)

---

✨ Features

• Logging: add structured log API and agenthood log command on trace store (#507)

---

🐛 Bug Fixes

• Security: surface absent-lockfile and missing-skill integrity states (#506)

---

✨ Features

• Hitl: ask_human park primitive with test stub provider (#502)

---

✨ Features

• Hitl: add ask_human park primitive for park-and-resume hosts (#501), refs #496

---

🐛 Bug Fixes

• Llm: harden provider failover priority, per-entry keys and cheap retry

---

🐛 Bug Fixes

• Agenthood: correct agent count 19 -> 20 in package description

---

🐛 Bug Fixes

• Llm: remove unused name parameter in stream callback
• Llm: use announce() in complete() for consistent logging

✨ Features

• Runtime: add RunEventBus tests and enrich reasoning telemetry, refs #474

---

🐛 Bug Fixes

• Audit: fail-closed tool gating, testable filter, and doc reconciliation
• Audit: harden audit filter against array and bad vulnerabilities shape
• Ci: exempt npm ecosystem tools from dependency audit
• Ci: fail closed on empty-node advisories in audit gate
• Ci: restore mixed-node advisory reporting in audit gate
• Members: address auditor and reviewer findings on member tool gating
• Members: prevent LLM from echoing SKILL.md content in responses, refs #473
• Review: address reviewer warnings on 6906423
• Review: sync institutional-knowledge steps 2/4 with oracle and harden audit filter

---

🐛 Bug Fixes

• Ci: remove stray Githubfx gitlink that broke CI checkout (#483)

✨ Features

• Members: reframe steward load routing and add the-mediator (#482), refs #474

---

🐛 Bug Fixes

• Core: clean up MemberAgent delegation smell and sanitize the-mailman (#469)

---

✨ Features

• Skills: add conversational style section to all Society members (#463)

---

🐛 Bug Fixes

• Core: require distinct markers before propagation fires
• Security: complete runtime split, flatten delegation nesting, and escape user_query boundary

✨ Features

• Agents: add mind virus immunity warning to system prompts
• Core: add injection-time SKILL.md integrity check
• Core: add viral-persona and propagation anomaly signals
• Security: complete mind virus defense — corrupt lockfile handling, marker clamp, redaction default, and test coverage for 461 fixes
• Tools: forbid message propagation in delegated tasks

---

🐛 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), refs #435

---

🐛 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, refs #313
• 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, refs #319
• Observability: expose episode learner learning status, refs #303

---

✨ Features

• Observability: add health check command and API, refs #321

---

✨ Features

• Observability: stamp traces with eval baseline quality, refs #306

---

✨ Features

• Observability: add anomaly detection for cost and quality, refs #306
• Observability: add trace retention and export policy, refs #307

---

✨ Features

• Observability: add redaction filter for trace payloads, refs #305

---

✨ Features

• Evals: add replay evaluator for behavior drift, refs #314

---

✨ Features

• Cli: add eval command with baseline regression gating, refs #298

---

✨ Features

• Evals: add baseline comparison for eval reports, refs #311

---

✨ Features

• Evals: add eval runner with llm-as-judge scoring, refs #310

---

🐛 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, refs #300

---

✨ Features

• Cli: add trace command — npx agenthood trace, refs #302

---

✨ Features

• Observability: add workflow and session correlation IDs, refs #301

---

✨ Features

• Observability: persist traces to a queryable store, refs #299

---

✨ Features

• Evals: define eval suite format with Ajv validation, refs #293

---

✨ Features

• Core: implement TokenCounter and CostEstimator, refs #296, #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, refs #292, #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

---

🐛 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, refs #366
• Members: make skills/ the single source of truth for member SKILL.md, refs #366
• 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, refs #367
• 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, refs #354, #312

---

🐛 Bug Fixes

• add root commitlint.config.ts for repo CI
• Init: resolve 11 failing health checks by correcting source paths and workflow, refs #14
• 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, refs #324
• Providers: restore custom tool call handling in shared parseToolCall, refs #350

---

🐛 Bug Fixes

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

✨ Features

• Skills: add 16 platform integration skills (CLI-focused), refs #348

---

🐛 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, refs #346

---

🐛 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, refs #119

---

🐛 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, refs #279, #280, #114
• Phase: phase 1 - protocol interfaces and workflow engine, refs #116
• Phase: phase 2 - workflow checkpoint and goal chain, refs #117, #118
• Phase: phase 3 - oracle, strategist, and operator agents, refs #113, #277, #278
• Phase: phase 4 - verify, rollback, and status commands, refs #275, #276, #281
• Phase: phase 5 - diff impact analyzer and quality gates, refs #115, #282
• Phase: phase 6 - review-pr workflow end-to-end

---

🐛 Bug Fixes

• Docs: address Reviewer findings on init check count and LanceDBStore API example, refs #286
• Docs: correct check count to 21 and fix insert->add API example, refs #286
• Docs: fix mentioned shipped version
• Init,check: align init ceremony with health check expectations, refs #286
• Skills: add output format section to the-reviewer SKILL.md for consistent rendering, refs #286
• sync skills/the-reviewer/SKILL.md with members/ changes, refs #286
• The-reviewer: address review findings on output format and README, refs #286
• The-reviewer: flatten heading hierarchy and add intra-section spacing example, refs #286
• The-reviewer: use [SEVERITY] placeholder and move meta-instruction outside template, refs #286

✨ Features

• load .env file automatically via dotenv, refs #286

---

🐛 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, refs #285
• 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, refs #283

✨ 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, refs #285
• Orchestration: implement MemberOrchestrator detection, refs #201
• Rag: implement AgenticRAG with RetrievalDecisionSkill, refs #108
• Rag: implement HierarchicalChunkStrategy with parent-child chunking, refs #109

🔹 BREAKING CHANGES

• Governance: announcements, deprecation policy, compliance

---

🐛 Bug Fixes

• Memory: align ProjectMemoryImpl return types with ProjectMemory interface, refs #269

---

🐛 Bug Fixes

• Deps: pin tree-sitter-go and tree-sitter-python to v0.23.x to resolve peer dependency conflict, refs #269

✨ Features

• Memory: implement PersonalisationStore for per-project agent adaptation, refs #112
• Memory: implement ShortTerm, LongTerm, Episodic, and Project memory tiers, refs #262
• Rag: implement baseline RAG pipeline — ChunkStrategy, Indexer, Retriever, refs #263
• Rag: implement SocietyIndexer for members, ADRs, and conventions, refs #107
• Rag: implement TreeSitterParser for AST-based code structure extraction, refs #106

---

🐛 Bug Fixes

• address reviewer findings and update Phase 0 docs, refs #268
• ignore entire .agenthood/ directory except config.example.json, refs #110
• Llm: extract and granularize api key validation, refs #203

✨ Features

• Core: move schema validator to core and harden error messages, refs #205
• Memory: implement LanceDB vector store with IVectorStore interface, refs #261
• Memory: implement memory governance with IMemoryStore and InMemoryStore, refs #111
• Memory: implement ResidualMemory — decay-weighted trace signals, refs #110
• Rag: implement KnowledgeGraphStore for relationship-aware retrieval, refs #105
• Reasoning: add infinite loop detection to reactloop, refs #206

---

🐛 Bug Fixes

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

---

🐛 Bug Fixes

• Academy: compute relative link from non-index pages at correct depth, refs #243
• 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, refs #207
• Cli: wire provider failover config into CLI and LLMRouter
• Provider: implement model downgrade and circuit breaker config, refs #217

---

🐛 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
• Config: update stale commitlint.config.cjs references to .ts, refs #237
• 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
• Core: add concurrency queue and safety guard
• Core: add RiskManager, SkillRegistry discovery, dynamic routing, and README rewrite, refs #103, #162, #102
• Core: implement ContextCompressor with token-aware memory summarization, refs #104
• 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, refs #202
• ship M4 foundation - TypeScript runtime with providers, agents, skills, and CLI
• Skills: export and register SubagentTaskSkill with delegate_task name, refs #3, #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, refs #234

---

🐛 Bug Fixes

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

---

🐛 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, refs #224
• 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), refs #185

---

🐛 Bug Fixes

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

---

🐛 Bug Fixes

• Academy: resolve ADR rendering and broken cross-links (#186), refs #185

---

✨ Features

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

---

🐛 Bug Fixes

• Skill: normalize SKILL.md structure for milestone M1 (#183), refs #66

---

🐛 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), refs #63

---

✨ Features

• Vscode: implement workspace event bus for passive observation (#62), refs #56

---

🐛 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), refs #45
• Vscode: modernize with build, tests, and CI (#54), refs #52

---

🐛 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, refs #7
• Check: validate all 14 members in health check (#27), refs #25, #26
• Ci: add ADR presence check to librarian.yml (#20), refs #11
• Ci: add AGENTS.md to sentinel.yml trigger paths (#19), refs #10
• Ci: fix sentinel multi-word section checks
• Ci: use commitlint.config.cjs for esm compat
• Conventions: add vague-subject rule to commitlint config, refs #8
• Docs: correct member count to fourteen
• Gitmessage: replace project-specific scope examples with generic placeholders, refs #15
• Portals: create missing linear.md and jira.md connector docs (#28), refs #24
• Release: disable npm publish until NPM_TOKEN is configured (#33)

✨ Features

• Adr: create foundational ADRs for Agenthood's own architecture (#30), refs #12
• Agentic-workflows: clarify workflow files as manual-prompt templates (#31), refs #13
• Bootstrap: add .agenthood/config.example.json reference template, refs #9
• Bootstrap: implement agenthood setup command and init CLI (#23), refs #14
• 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