godpowers 0.15.0

package/AGENTS.md

@@ -0,0 +1,37 @@
+# Godpowers - Agent Brief
+
+This is the Godpowers repository: an AI-powered development system that takes
+projects from raw idea to hardened production.
+
+## Reading Order
+
+1. `README.md` - what this is, how to use it
+2. `SKILL.md` - the core skill (the brain)
+3. `skills/god-mode.md` - the autonomous orchestrator
+4. `skills/god-init.md` - project initialization
+5. `skills/god-prd.md` through `skills/god-harden.md` - individual tier skills
+
+## Architecture
+
+- `SKILL.md` is the main entry point, loaded by AI coding tools
+- `skills/` contains individual command skills (one per file)
+- `references/` contains per-tier reference material (antipatterns, examples)
+- `bin/` contains the CLI installer and `god` command
+- `scripts/` contains validation and testing scripts
+- `templates/` contains artifact templates
+
+## Conventions
+
+- Every skill file has YAML frontmatter with `name`, `description`, and trigger phrases
+- Every tier skill documents its gate check (upstream dependency)
+- Every tier skill documents its have-nots (failure modes)
+- Artifacts are written to `.godpowers/<tier>/` paths
+- State is tracked in `.godpowers/PROGRESS.md`
+- Disk state is authoritative; conversation memory is not
+
+## Quality Rules
+
+- No em dashes or en dashes (use commas, colons, semicolons, or parentheses)
+- No emojis (use `+`, `-`, `x`, `!` for status indicators)
+- Every sentence in generated artifacts must be labeled: DECISION, HYPOTHESIS, or OPEN QUESTION
+- Every claim must fail the substitution test (swap in a competitor, sentence must break)

package/CHANGELOG.md

@@ -0,0 +1,639 @@
+# Changelog
+
+All notable changes to Godpowers will be documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [0.15.0] - 2026-05-11
+
+Distribution release. Godpowers and its three first-party packs are
+now publishable to npm with supply-chain provenance. OTel exporter
+turns events.jsonl into something a real observability backend can
+consume. Cost tracker distinguishes live (provider-reported) from
+estimated (heuristic) per-call token counts.
+
+### Added (cost-tracker live integration)
+- `cost.recorded` events now carry `source: 'live' | 'estimated'`.
+  Live = AI tool surfaced real per-call token counts from the
+  provider API; estimated = orchestrator's heuristic byte-based
+  count.
+- `lib/cost-tracker.recordModelCall(handle, attrs)`: canonical entry
+  point for AI tools with real usage data. Tags `source: 'live'`
+  automatically.
+- `lib/cost-tracker.isStrictLive(projectRoot, runIds?)`: returns
+  `{strict, live_calls, estimated_calls, total_calls}` to drive CI
+  gates.
+- `aggregate()` splits `totals` into `live_calls / live_usd /
+  live_tokens` and `estimated_*`.
+- `formatReport()` breaks the Spent line into Live and Estimated
+  sublines.
+- `/god-cost --strict`: exit non-zero if any in-scope record is
+  estimated. Wire into CI once live reporting is reliable.
+
+### Added (OTel exporter)
+- `lib/otel-exporter.js`: converts a run's events.jsonl into
+  OTLP/JSON ResourceSpans and (optionally) POSTs to an OTLP HTTP
+  collector. No external deps; uses Node's built-in http/https.
+- Event -> span mapping: `workflow.run` + `workflow.complete` form
+  the root span; `agent.start` + `agent.end` (matched by span_id)
+  become child spans parented to the root; other named events
+  (`cost.recorded`, `gate.fail`, `error`, `decision.route`,
+  `tool.call`) attach as span events; `error` or `gate.fail`
+  flips parent span status to ERROR.
+- Honors `OTEL_EXPORTER_OTLP_TRACES_ENDPOINT` (takes precedence),
+  `OTEL_EXPORTER_OTLP_ENDPOINT` (with `/v1/traces` auto-appended),
+  and `OTEL_EXPORTER_OTLP_HEADERS` (comma-separated `k=v` for auth
+  like `x-honeycomb-team=...`).
+- New skill `/god-export-otel`: opt-in; nothing exports until
+  invoked. Supports `--run-id`, `--all`, `--endpoint`, `--stdout`,
+  `--service-name`.
+
+### Added (first-party packs publishable as npm packages)
+- Each pack ships its own `package.json` with the scoped name,
+  `publishConfig.access=public`, `files` allowlist, and
+  `peerDependencies.godpowers`:
+  - `@godpowers/security-pack` (SOC 2, HIPAA, PCI-DSS auditors)
+  - `@godpowers/launch-pack` (Show HN, Product Hunt, Indie Hackers,
+    OSS release strategists)
+  - `@godpowers/data-pack` (ETL, ML features, dashboards)
+- Each pack starts at `0.1.0`, decoupled from the godpowers root
+  version.
+- Manifest engines fix: `>=1.0.0 <2.0.0` (impossible while godpowers
+  is on 0.x) -> `>=0.14.0 <2.0.0`.
+
+### Added (distribution)
+- `.github/workflows/publish.yml`: tag-triggered publish. On a `v*`
+  tag push, runs the full test suite as a gate, then `npm publish
+  --provenance --access public` using `${{ secrets.NPM_TOKEN }}`.
+- `.github/workflows/publish-pack.yml`: workflow_dispatch entry
+  point. Pick `security-pack`, `launch-pack`, or `data-pack` from
+  the GitHub Actions UI to publish that pack to npm after bumping
+  its `package.json` version.
+- CHANGELOG remains human-curated; `npm version minor` + push tags
+  is the release procedure.
+
+### Tests
+- 9 new tests for `source: 'live' | 'estimated'` cost-tracker
+  behavior, `recordModelCall()`, `isStrictLive()`, and aggregate
+  splits (35 total in `test-cost-saver.js`).
+- 15 new tests for the OTel exporter, including a real in-process
+  HTTP collector verifying POST, content-type, path, body, and
+  auth-header propagation.
+- 43 new tests for extension-pack publish readiness across all
+  three first-party packs: package.json well-formed, name+version
+  match manifest, `npm pack --dry-run` succeeds, tarball includes
+  required files, tarball excludes node_modules / .git / tests.
+- Full suite now 36 test suites, 1864+ behavioral tests.
+
+## [0.14.0] - 2026-05-11
+
+The big release. Workflow runtime is now executable, locks and
+checkpoints are wired into the orchestrator, a full token cost saver
+ships, on/off budget shortcut for casual users, GitHub Actions CI,
+and npm publish prep.
+
+### Added (lock + checkpoint wiring)
+- `lib/state-lock.js`: acquire / release / peek / isStale / reclaim /
+  withLock. Reentrant; scope-based conflict; stale-lock reclaim;
+  withLock guarantees release on thrown error.
+- `lib/checkpoint.syncFromState`: one-call orchestrator helper that
+  rebuilds CHECKPOINT.md from state.json + events.jsonl tail.
+- 28 artifact-producing skills gained `## Locking` sections that
+  document the contract for end users.
+- `agents/god-orchestrator.md`: new Concurrency section that wires
+  lock acquisition + CHECKPOINT auto-update into every sub-step.
+
+### Added (token cost saver)
+- `lib/cost-tracker.js`: per-call cost recording + per-tier /
+  per-agent / per-model aggregation. Built-in pricing table for
+  major models (Claude / GPT / Gemini / O-series). recordCost /
+  recordCacheHit / recordCacheMiss / aggregate / formatReport /
+  priceTokens.
+- `lib/agent-cache.js`: opt-in agent-output cache keyed by
+  sha256(agent + version + sorted inputs + state-hash). TTL-bounded.
+  Sharded storage. Cache hits emit `cache.hit` with savings_usd.
+- `lib/context-budget.js`: per-agent budget planner. Reads required /
+  optional context from agent frontmatter; loads under cap; emits
+  `budget.exceeded` when required alone overflows.
+- `schema/intent.v1.yaml.json`: new `budgets` block with
+  `default-max-tokens`, `model-profile` (cheap / standard /
+  expensive), `cache`, `cache-ttl-hours`, per-agent overrides.
+- `lib/events.js`: 4 new event names (`cost.recorded`, `cache.hit`,
+  `cache.miss`, `budget.exceeded`).
+- 3 new skills: `/god-cost` (spend + savings report), `/god-budget`
+  (view + set budgets), `/god-cache-clear` (invalidate cache).
+- `lib/budget.js`: applyOn / applyOff / set / read / summary so
+  `/god-budget --on` and `--off` work as one-shot toggles.
+- `agents/god-orchestrator.md`: new "Cost-conscious agent dispatch"
+  section wiring cache check, context-budget loadout, model selection
+  by profile, and cost recording into every spawn.
+
+### Added (v0.14 workflow runtime)
+- `lib/workflow-runner.js`: listWorkflows + loadByName + plan +
+  writePlan + readPlan. Reads workflows/*.yaml via
+  lib/workflow-parser; computes dependency-ordered waves; serializes
+  plans to .godpowers/runs/<id>/plan.yaml.
+- `/god-mode --workflow=<name>` and `/god-mode --plan` flags added.
+- All 13 workflows/*.yaml had their "NOT YET AUTHORITATIVE" header
+  replaced with "Authoritative (v0.14+)".
+
+### Added (release engineering)
+- `.github/workflows/ci.yml`: matrix on Node 18/20/22; runs npm test
+  on every PR + main push. Separate `package` job runs npm pack
+  --dry-run and verifies bin entry + CHANGELOG has a current entry.
+- `package.json` files array: routing/, workflows/, extensions/,
+  INSPIRATION.md added (previously missing -> would have shipped a
+  broken package). prepublishOnly runs the full test suite before
+  any npm publish.
+- npm pack --dry-run: 364KB tarball / 1.1MB unpacked / 439 files.
+
+### Tests
+- 34 suites (was 30 at v0.13.0), 1863 passing (was 1764). +99.
+- New: test-state-lock.js (21), test-cost-saver.js (26),
+  test-budget-onoff.js (13), test-workflow-runner.js (12).
+
+### Changed
+- bin/install.js VERSION: 0.13.0 -> 0.14.0
+- package.json version: 0.13.0 -> 0.14.0
+
+## [0.13.0] - 2026-05-10
+
+Context-rot protection (the major new feature), extension runtime, and
+observability readers. Ships earlier-than-roadmapped observability
+because it cost very little once events.jsonl already existed.
+
+### Added (context-rot protection)
+- `lib/checkpoint.js`: read/write/append API for `.godpowers/CHECKPOINT.md`,
+  the disk-authoritative "where you are" pin. Frontmatter holds id,
+  project, mode, mode-d-suite, lifecycle, current-tier, current-substep,
+  last-action, last-actor, last-update, facts-hash. Markdown body holds
+  Where-you-are + Last actions (up to 20) + Held facts (up to 10) +
+  Next suggested command + "if you are a new session" guide.
+- `skills/god-locate.md`: orient a new chat session or new AI tool against
+  disk reality. Reads CHECKPOINT + state.json + events.jsonl tail +
+  reflog + intent.yaml. Single-screen output. Replaces guesswork when
+  switching tools or returning to a project.
+- `skills/god-context-scan.md`: detect drift between the AI's stated mental
+  model and disk. The defensive cousin of /god-status. Use in long
+  sessions before commits. Surfaces hallucinated facts and stale memory.
+- `lib/events.js`: hash chain on events.jsonl. Each event includes
+  `prev` = sha256 of the previous line (or `genesis` for the first).
+  New `verifyChain(file)` detects any truncation / tampering /
+  reordering. Cheap; cryptographically tamper-evident audit log.
+- `hooks/session-start.sh`: now prefers CHECKPOINT.md over PROGRESS.md
+  when present. Prints the "Where you are" + Next-suggested sections
+  for instant orientation.
+
+### Added (v0.13 extension runtime)
+- `schema/extension-manifest.v1.json`: schema for extension manifests
+  (apiVersion, kind, metadata.name + version, engines.godpowers SemVer
+  range, provides {agents,skills,workflows,have-nots}, activation rules).
+- `lib/extensions.js`: install / list / info / remove + SemVer range
+  matcher (exact, ^X.Y.Z, ~X.Y.Z, >=, <, compound). Capability handshake
+  rejects install when engines.godpowers doesn't include the running
+  godpowers version.
+- 5 new skills: `/god-extension-add`, `/god-extension-list`,
+  `/god-extension-info`, `/god-extension-remove`, `/god-test-extension`.
+
+### Added (v0.15 observability, shipped early)
+- `lib/event-reader.js`: timeline / metrics / trace / summarize over
+  events.jsonl. Pairs agent.start with agent.end to compute durations.
+  Aggregates per-tier counts, durations, pauses, errors. Reads one run
+  or all runs in the project.
+- 3 new skills: `/god-logs`, `/god-metrics`, `/god-trace`.
+
+### Tests
+- 27 -> 30 test suites; 1629 -> roughly 1670 passing.
+- New: scripts/test-checkpoint.js (16), scripts/test-extensions.js (19),
+  scripts/test-event-reader.js (9).
+- All existing tests still pass; the events hash chain didn't break
+  anything because emit() was always single-line append.
+
+### Changed
+- bin/install.js VERSION: 0.12.0 -> 0.13.0
+- package.json version: 0.12.0 -> 0.13.0
+
+## [0.12.0] - 2026-05-10
+
+Mode D (multi-repo suites), agent discipline, story-file workflow, Pi + T3
+support, and routing sweep. Test suite at 1415 passing across 22 suites.
+
+### Added (Mode D - multi-repo suites)
+- `agents/god-coordinator.md`: Tier-0 peer agent for cross-repo coordination
+- `lib/suite-state.js`: Mode D suite registration + version table
+- `lib/meta-linter.js`: cross-repo lint for byte-identical files
+- `skills/god-suite-init.md`: register siblings + shared standards
+- `skills/god-suite-status.md`: side-by-side repo status
+- `skills/god-suite-sync.md`: byte-identical file propagation
+- `skills/god-suite-patch.md`: coordinated multi-repo change
+- `skills/god-suite-release.md`: release coordination across siblings
+- `references/shared/multi-repo-suite-layout.md`
+
+### Added (agent discipline - phase 17)
+- `lib/agent-validator.js`: validates every agents/*.md against the agent
+  contract (frontmatter, required sections, output schema)
+- `skills/god-agent-audit.md`: `/god-agent-audit` runs the validator
+
+### Added (story-file workflow - phase 18)
+- `lib/story-validator.js`: parses + validates STORY.md files
+- `agents/god-storyteller.md`: STORY.md writer
+- `skills/god-story.md`: write a new story
+- `skills/god-stories.md`: list stories by status
+- `skills/god-story-build.md`: implement a story
+- `skills/god-story-verify.md`: run acceptance criteria as headless tests
+- `skills/god-story-close.md`: close after build + verify
+
+### Added (runtime support)
+- Pi (earendil-works/pi) support in installer (`--pi` flag, ~/.pi/skills/)
+- T3 Code (pingdotgg/t3code) transparent support via underlying agent
+- Cross-tool Agent Skills standard at .agents/skills/
+
+### Added (brownfield depth)
+- `agents/god-archaeologist.md`: deep code archaeology
+- `agents/god-reconstructor.md`: reverse-engineer planning artifacts
+- `agents/god-reconciler.md`: cross-artifact reconciliation
+- `agents/god-debt-assessor.md`: technical-debt scorer
+- `skills/god-archaeology.md`, `god-reconstruct.md`, `god-reconcile.md`,
+  `god-tech-debt.md`
+
+### Changed (routing sweep + integration)
+- Phase 13: routing sweep + beyond-arc skill linkage participation
+- Phase 14: documentation surface for runtime / linkage / design
+- Phase 15: runtime heuristic improvements (parseFlow verb coverage)
+- Audit-driven fixes: closed 4 misconnections + disconnections between
+  beyond-arc workflows and linkage / reverse-sync
+
+### Documentation
+- `INSPIRATION.md`: single canonical acknowledgement of prior-art
+- Doc deck refreshed to current state (82 skills, 38 agents, v0.11+)
+
+### Tests
+- 22 test suites, 1400+ passing (was 18 suites, 1235 at 0.11.0)
+
+## [0.11.0] - 2026-05-10
+
+Major release. Production-ready validation, full design pipeline, and
+runtime verification. 18 commits since 0.4.0; full test suite at 1235
+passing across 18 suites.
+
+### Added (validation foundation)
+- `lib/have-nots-validator.js`: 11 mechanical have-nots checks (em/en
+  dash, emoji, unlabeled paragraphs, phantom references, future dates,
+  generic claims, PRD/ARCH structure violations)
+- `lib/artifact-linter.js`: per-artifact orchestrator with detectType,
+  lintFile, lintAll, formatReport, aggregate
+- `lib/artifact-diff.js`: regression detection between artifact versions
+- `skills/god-lint.md`: `/god-lint` mechanical validation
+- `references/HAVE-NOTS.md` integrated into linter
+
+### Added (exemplars and antipatterns parity)
+- `examples/saas-mrr-tracker/` complete UI project (PRD/ARCH/ROADMAP/STACK/DESIGN)
+- `examples/cli-tool/` backend-only project (PRD/ARCH/ROADMAP/STACK)
+- `references/planning/ROADMAP-ANTIPATTERNS.md`
+- `references/planning/STACK-ANTIPATTERNS.md`
+- `references/building/BUILD-ANTIPATTERNS.md`
+- `references/shipping/{DEPLOY,OBSERVE,HARDEN,LAUNCH}-ANTIPATTERNS.md`
+- `references/design/{DESIGN-ANATOMY,DESIGN-ANTIPATTERNS}.md`
+
+### Added (design foundation - integrations)
+- `lib/design-detector.js`: UI presence detection across 24+ frameworks
+- `lib/design-spec.js`: Google Labs design.md parser + linter (frontmatter
+  schema, section order, token resolution, WCAG contrast)
+- `lib/impeccable-bridge.js`: detect-and-delegate to Impeccable's 23 commands
+- `lib/awesome-design.js`: 71-site catalog from VoltAgent's awesome-design-md
+- `lib/skillui-bridge.js`: SkillUI fallback for sites not in catalog
+- `agents/god-designer.md`: lifecycle owner of DESIGN.md + PRODUCT.md
+- `agents/god-design-reviewer.md`: two-stage gate (spec + quality)
+- `skills/god-design.md`: 26 subcommands bridging impeccable + catalog + skillui
+- `routing/god-design.yaml`
+- `templates/DESIGN.md`
+
+### Added (linkage + propagation)
+- `lib/linkage.js`: bidirectional artifact-to-code map with 7 stable ID types
+  (P-MUST/SHOULD/COULD-NN, ADR-NNN, C-{slug}, M-{slug}, S-{slug}, token paths, D-{slug})
+- `lib/code-scanner.js`: 6 discovery mechanisms (annotations, filenames,
+  imports, style-system, test descriptions, manual)
+- `lib/drift-detector.js`: design token drift, stack version drift, ARCH
+  container drift
+- `lib/impact.js`: forward propagation (artifact change -> affected code)
+- `lib/cross-artifact-impact.js`: 6 rule classes for artifact-to-artifact impact
+- `lib/review-required.js`: REVIEW-REQUIRED.md + REJECTED.md managers
+- `lib/reverse-sync.js`: code -> artifact fenced footer appender
+  (PRD/ARCH/ROADMAP/STACK/DESIGN)
+- `skills/god-design-impact.md`: what-if analysis
+- `skills/god-review-changes.md`: walk REVIEW-REQUIRED.md
+- `skills/god-scan.md`: manual reverse-sync
+- `skills/god-link.md`: manual link entry
+
+### Added (runtime verification - headless)
+- `lib/browser-bridge.js`: headless-only browser launch (cascade:
+  agent-browser preferred, Playwright fallback)
+- `lib/agent-browser-driver.js`: vercel-labs/agent-browser CLI wrapper
+- `lib/runtime-audit.js`: design verification on rendered DOM (computed
+  styles vs DESIGN.md tokens, real-DOM WCAG contrast)
+- `lib/runtime-test.js`: PRD acceptance criteria as user-flow assertions
+- `agents/god-browser-tester.md`
+- `skills/god-test-runtime.md`
+
+### Added (light-impeccable - 7 design domain references)
+- `references/design/TYPOGRAPHY.md` (~140 lines)
+- `references/design/COLOR.md` (~145 lines)
+- `references/design/SPATIAL.md` (~110 lines)
+- `references/design/MOTION.md` (~120 lines)
+- `references/design/INTERACTION.md` (~150 lines)
+- `references/design/RESPONSIVE.md` (~125 lines)
+- `references/design/UX-WRITING.md` (~130 lines)
+
+### Added (ai-tool context)
+- `lib/context-writer.js`: AGENTS.md / CLAUDE.md / GEMINI.md / .cursor/ /
+  .windsurf/ / others fenced section manager (11 AI tools detected)
+- `agents/god-context-writer.md`
+- `skills/god-context.md`
+
+### Added (front-door)
+- `skills/god.md`: free-text intent matcher
+
+### Changed
+- `god-orchestrator.md`: extended Quarterback responsibilities;
+  detection-driven Tier 1 routing; mid-arc DESIGN/PRODUCT change
+  detection; extended critical-finding gate (drift, lint errors,
+  design-review BLOCK, validator errors); explicit YOLO behavior table
+- `god-updater.md`: now calls reverse-sync.run on /god-sync
+- `lib/state.js`: schema additions (tier-1.design, tier-1.product,
+  linkage slot, yolo-decisions array)
+
+### Documentation
+- `docs/change-propagation.md`: forward + reverse + cross-artifact propagation
+- `docs/linkage.md`: stable IDs, 6 discovery mechanisms, drift detection
+- `.planning/2026-05-10-production-ready-and-design.md`: comprehensive plan
+- `.planning/dogfood-001-results.md`: end-to-end validation results
+
+### Tests
+- 18 test suites, 1235 passing, 0 failing (was 4 suites, ~360 tests at 0.4.0)
+- All new tests behavioral, not just structural
+
+### External integrations (5; all detect-and-delegate, none vendored)
+- Google Labs design.md (format spec)
+- Impeccable (design intelligence; 7 domain refs + 23 commands + 27 anti-patterns)
+- VoltAgent awesome-design-md (71-site curated catalog)
+- SkillUI (static analysis fallback for arbitrary URLs)
+- vercel-labs/agent-browser + Playwright (runtime verification backends)
+
+## [0.8.1] - 2026-05-09
+
+### Changed
+- Automatic mode detection across `/god-init` and `/god-mode`: greenfield /
+  brownfield / bluefield decided invisibly from disk signals (package.json,
+  source dirs, org-context markers). Users no longer need to know the
+  jargon; the system asks plain-English clarifying questions only when
+  signals are ambiguous.
+
+## [0.8.0] - 2026-05-09
+
+### Added (brownfield + bluefield support)
+- `agents/god-archaeologist.md`: deep code archaeology (history, decisions,
+  conventions, risks, tribal knowledge) beyond `/god-map-codebase`
+- `agents/god-reconstructor.md`: reverse-engineers PRD / ARCH / ROADMAP /
+  STACK from existing code so brownfield projects can use full
+  reconciliation
+- `agents/god-debt-assessor.md`: 8-category debt assessment with P0-P3
+  prioritization (code, design, dependency, security, test, doc,
+  operational, knowledge)
+- `agents/god-org-context-loader.md`: bluefield support for org-level
+  shared standards, conventions, infrastructure, libraries
+- Skills: `/god-archaeology`, `/god-reconstruct`, `/god-tech-debt`,
+  `/god-org-context`
+
+## [0.7.3] - 2026-05-09
+
+### Added
+- Greenfield artifact coverage: `/god-mode` reliably creates the 10 core
+  artifacts across Tier 0-3 sub-steps. The 4 capture artifacts (BACKLOG,
+  SEEDS, TODOS, THREADS) remain lazy by design (created only when used).
+- Mandatory final sync: `/god-mode` always runs `/god-sync` at end of arc
+  regardless of flags (`--yolo`, `--conservative`, `--with-hygiene`).
+
+## [0.7.2] - 2026-05-09
+
+### Added (comprehensive multi-artifact reconciliation)
+- Extended reconciliation from roadmap-only to all 13 impacted artifacts:
+  PRD, ARCH, ROADMAP, STACK, REPO, DEPLOY, OBSERVE, HARDEN, LAUNCH,
+  BACKLOG, SEEDS, TODOS, THREADS
+- New skills: `/god-reconcile` (before feature work), `/god-sync` (after)
+- Cross-artifact divergence detection (e.g. PRD/roadmap moving-target,
+  STACK drift vs lock files)
+
+## [0.7.1] - 2026-05-09
+
+### Added (roadmap reconciliation loop)
+- `agents/god-roadmap-reconciler.md`: classifies user intent vs ROADMAP.md
+  with 6 verdicts (already-done, in-progress, enhancement,
+  prerequisite-needed, new, ambiguous)
+- `agents/god-roadmap-updater.md`: keeps ROADMAP.md as a living artifact
+  after feature work; detects PRD divergence
+- Skills: `/god-roadmap-check`, `/god-roadmap-update`
+
+## [0.7.0] - 2026-05-09
+
+### Added (recipes as programmatic input)
+- `schema/recipe.v1.json`: structured recipe definition with triggers
+  (intent-keywords + state-conditions), decision-tree, sequences, default
+- 33 structured recipes at `routing/recipes/<recipe>.yaml` covering
+  starting, feature-addition, production, maintaining, recovering,
+  collaborating, knowledge categories
+- Agents consult recipes for fuzzy intent decisions (no longer just human docs)
+
+## [0.6.0] - 2026-05-09
+
+### Added (unified decision engine + routing + standards gates)
+- `schema/routing.v1.json`: JSON Schema for routing definitions
+  - Prerequisites (required + recommended) with auto-complete commands
+  - Execution (spawns, secondary-spawns, reads, writes)
+  - Standards (substitution-test, three-label-test, have-nots,
+    gate-on-failure)
+  - Success-path + failure-path
+- Standards gates between command stages (artifact discipline runs
+  independent of the producing agent)
+
+## [0.5.0] - 2026-05-09
+
+### Added (reference content depth)
+- 12 substantive per-tier reference documents at `references/`:
+  - `planning/PRD-ANATOMY.md`, `PRD-ANTIPATTERNS.md`
+  - `planning/ARCH-ANATOMY.md`, `ARCH-ANTIPATTERNS.md`
+  - `planning/ROADMAP-ANATOMY.md`, `STACK-ANATOMY.md`
+  - `building/BUILD-VERTICAL-SLICES.md`, `BUILD-WAVES.md`
+  - `shipping/DEPLOY-PATTERNS.md`, `OBSERVE-SLO-EXAMPLES.md`,
+    `HARDEN-OWASP-WORKSHEETS.md`
+  - `orchestration/MODE-DETECTION.md`, `SCALE-DETECTION.md`
+  - `shared/GLOSSARY.md`, `ORCHESTRATORS.md`
+- Each with worked examples, antipattern catalogs, concrete guidance
+
+## [0.4.0] - 2026-05-09
+
+### Added
+- **god-mode lifecycle awareness**:
+  - god-orchestrator now has explicit Post-Launch Transition phase
+  - After `/god-mode` (full-arc), project enters STEADY STATE
+  - Steady-state hand-off message lists all 11 ongoing workflows
+  - New flag `/god-mode --with-hygiene` runs audit + deps + docs verification
+  - `/god-mode --yolo --with-hygiene` enables autonomous hygiene (still pauses on Critical CVEs)
+- **2 new lifecycle slash commands**:
+  - `/god-hygiene`: composite health check (audit + deps + docs)
+  - `/god-lifecycle`: shows project phase and contextually appropriate workflows
+- **v0.5 scaffolding**:
+  - `schema/intent.v1.yaml.json`: JSON Schema for godpowers.yaml (intent)
+  - `schema/state.v1.json`: JSON Schema for state.json (facts)
+  - `schema/events.v1.json`: OpenTelemetry-shape event vocabulary
+  - `lib/README.md`: planned runtime modules with target versions
+  - `docs/RFC/0002-workflow-yaml-v1.md`: workflow language design
+- **Distribution prep**:
+  - `.npmignore` excludes dev files from npm package
+  - `package.json` repository, homepage, bugs fields populated
+  - `scripts/release.sh`: tag + publish flow with verification
+- **First-party extension scaffold**:
+  - `extensions/security-pack/manifest.yaml`
+  - `extensions/security-pack/agents/god-soc2-auditor.md`
+  - `extensions/security-pack/skills/god-soc2-audit.md`
+  - `extensions/security-pack/README.md`
+  - Demonstrates extension shape for v0.8 implementation
+- **Integration test scaffold**:
+  - `tests/README.md`: three-layer test strategy
+  - `tests/integration/README.md`: planned end-to-end tests with record/replay
+  - Fixture project layout designed for v0.5 implementation
+
+- **8 new workflow slash commands** for real-world scenarios beyond greenfield:
+  - `/god-feature`: Add a feature to an existing project
+  - `/god-hotfix`: Urgent production bug fix (skips planning, expedited deploy)
+  - `/god-refactor`: Safe refactor with strict TDD (no behavior change)
+  - `/god-spike`: Time-boxed research with throwaway POC
+  - `/god-postmortem`: Post-incident investigation (root cause + class-of-bug)
+  - `/god-upgrade`: Framework/version migration (expand-contract pattern)
+  - `/god-docs`: Write/update docs verified against code
+  - `/god-update-deps`: CVE-aware incremental dependency updates
+- **5 new specialist agents**:
+  - `god-incident-investigator`: Postmortems with action items + runbook updates
+  - `god-spike-runner`: Time-boxed POC with honest findings
+  - `god-migration-strategist`: Incremental migrations with rollback per slice
+  - `god-docs-writer`: No-lying docs verified against code
+  - `god-deps-auditor`: CVE-aware dependency updates with bisect-able commits
+- **5 new templates** for workflow artifacts:
+  - `templates/POSTMORTEM.md`
+  - `templates/SPIKE.md`
+  - `templates/MIGRATION.md`
+  - `templates/DOCS-UPDATE-LOG.md`
+  - `templates/DEPS-AUDIT.md`
+- **HAVE-NOTS.md catalog extended** with new failure modes for each new
+  artifact type (PM-01..PM-08, SP-01..SP-05, MG-01..MG-07, DC-01..DC-05,
+  DP-01..DP-06)
+- **Architecture design documents**:
+  - `ARCHITECTURE.md`: 16-section canonical design for v1.0 (pure-skill model)
+  - `docs/RFC/0000-research-brief.md`: Synthesized research informing the design
+  - `docs/RFC/0001-state-model-v1.md`: First detailed RFC
+  - `docs/ROADMAP.md`: v0.4 -> v1.0 implementation plan
+
+### Changed
+- `package.json` bumped to 0.4.0 (was stuck on 0.3.0 despite v0.4 work)
+- `install.js` VERSION constant bumped to 0.4.0
+- `/god-next` routing extended to suggest workflows based on user intent
+  (feature add, hotfix, refactor, etc.) when project is in steady state
+- README "Other Workflows" section added to command table
+- SessionStart hook updated to suggest new workflows for ambient discovery
+- Smoke test PAIRS extended to verify 8 new skill-to-agent routings
+- Each new agent references its template for structural starting point
+
+### Architecture decisions (documented, not yet implemented)
+- v0.5+ will introduce three load-bearing artifacts: intent.yaml + state.json
+  + events.jsonl
+- v0.6 will add forward-only recovery (`/god-undo`, `/god-rollback`)
+- v0.7 will add OTel-shape observability
+- v0.8 will add the skill pack ecosystem
+- v1.0 will freeze schemas
+
+## [0.3.0] - 2026-05-09
+
+### Added
+- **Critical fixes**:
+  - Installer now copies `references/` directory (HAVE-NOTS catalog and
+    per-tier reference content) so agents can find it in production
+  - `--uninstall` flag now actually removes Godpowers from the target runtime
+  - Install verification message lists how many commands and agents were
+    installed and shows exact next steps
+  - Templates explicitly referenced from each tier agent (god-pm references
+    PRD.md, god-architect references ARCH.md, etc.)
+- **Mode B (gap-fill) implementation**:
+  - god-orchestrator now scans existing artifacts on disk
+  - Detects which tiers have passing artifacts and skips them
+  - Uses codebase signals (package.json, CI configs, test dirs) to detect
+    partial progress
+- **Documentation**:
+  - CHANGELOG.md (this file)
+  - CONTRIBUTING.md
+  - SECURITY.md
+  - Per-tier reference subdirectories with placeholder content
+
+### Changed
+- `package.json` version bumped from 0.1.0 to 0.3.0 (was stuck on first commit)
+- Mode D (multi-repo) downgraded from "supported" to "future work" until real
+  implementation lands
+- Smoke test em/en-dash check rewritten to use Python instead of buggy bash
+  byte-class regex (false-positives on UTF-8 multi-byte chars starting with 0xE2)
+- `/god-init` skill now spawns god-orchestrator for mode/scale detection
+  instead of duplicating the logic
+- `/god-audit` skill now explicitly spawns god-auditor agent
+
+### Fixed
+- Smoke test no longer false-positives on block characters or other Unicode
+  starting with 0xE2
+
+## [0.2.1] - 2026-05-09
+
+### Added
+- Build phase orchestration explicit in god-orchestrator (4-agent chain per
+  slice: god-planner -> god-executor -> god-spec-reviewer -> god-quality-reviewer)
+- `--yolo` flag flows through to specialist agents with documented defaults
+- Critical-finding carve-out: god-harden-auditor never auto-resolves Critical
+  findings even with `--yolo`
+- 13 routing checks, 1 build-phase check, 6 YOLO handling checks, 1 carve-out
+  check added to smoke test
+
+### Fixed
+- /god-mode could stall on Build phase (orchestrator didn't reference reviewers)
+- /god-mode --yolo could pause at specialist agent pause conditions
+  (specialists didn't know about --yolo)
+
+## [0.2.0] - 2026-05-09
+
+### Added
+- 6 artifact templates in `templates/` (PRD, ARCH, ROADMAP, STACK, PROGRESS,
+  HARDEN-FINDINGS) with embedded have-nots checklists
+- `references/HAVE-NOTS.md` consolidated catalog with 115 named failure modes
+- 7 new runtimes in installer: Trae, Cline, Kilo, Antigravity, Qwen, CodeBuddy, Pi
+  (15 total)
+- 9 new slash commands:
+  - `/god-fast` - trivial inline edits
+  - `/god-quick` - TDD-discipline tasks below /god-build threshold
+  - `/god-explore` - Socratic ideation pre-init
+  - `/god-pause-work` - context handoff
+  - `/god-resume-work` - context restoration
+  - `/god-workstream` - parallel workspace management
+  - `/god-sprint` - sprint plan/status/retro
+  - `/god-party` - real multi-persona collaboration
+  - `/god-build-agent` - custom specialist agent generator
+- 2 new agents: `god-explorer`, `god-retrospective`
+- Mode A/B/C/D detection logic in god-orchestrator
+- Scale detection (trivial/small/medium/large/enterprise)
+- YOLO-DECISIONS.md emission for `--yolo` runs
+- `hooks/pre-tool-use.sh` safety hook
+
+## [0.1.0] - 2026-05-09
+
+### Added
+- Initial release
+- 17 slash commands (skills/) as thin orchestrators
+- 16 specialist agents (agents/) in fresh contexts
+- SessionStart hook
+- Installer for 8 AI coding tool runtimes
+- Smoke test and skill validation infrastructure
+- README, AGENTS.md, LICENSE