speckit-product-forge

Product Forge — SpecKit Extension & Claude Plugin

Full product lifecycle: Problem Discovery → Research → Product Spec → Revalidation → SpecKit → Pre-Impl Review → Implement → Code Review → Verify → Test → Release Readiness → API Docs · Security · Analytics · Retrospective

Installs as a SpecKit extension or as a Claude Code / Claude plugin from this same repo — see Installation and docs/claude-plugin.md.

Product Forge is a SpecKit extension that adds a complete product discovery, specification, and quality pipeline before and after any SpecKit implementation work. Instead of jumping straight to spec.md, you first research competitors, UX patterns, and your codebase — craft an approved product spec — review design/architecture/risks — let SpecKit implement it with progressive verification — run multi-agent code review — then automatically generate and run Playwright tests with a bug-fix loop until the feature is ready to ship.

New in v1.7.0 ("Ivysaur"):

• --dry-run preview (normative) — any phase (or forge end-to-end) runs fully but mutates nothing real: writes are redirected under .forge-dry-run/<phase>/, status is untouched, no external side-effects, and each phase emits a DRY-RUN-REPORT.md of what would change.
• Cross-model code review — code-review --cross-model exports the consolidated gate surface + diff and has a different model review it out-of-band, ingesting findings back into the single F-NNN namespace.
• Parallel implementation — implement --parallel runs path-disjoint task groups concurrently (proven independent by the portfolio conflict matrix), reconciled serially under the state-lock with a single gate.
• Constitution↔code drift layer — sync-verify Layer 10 / verify-full Layer 11 re-assert the project architecture constitution against the code as a standing check, not just at planning time.
• Lessons → reusable skills — retrospective promotes a lesson that recurs across features into a reusable skill (Hermes host), cross-project and idempotent.
• status --cost — rolls up per-phase token / tool-call telemetry per feature and across the portfolio (dollar cost when a rate is supplied).
• Phase-map single source — docs/schema/phase-map.yml is the canonical phase set + per-mode matrix; the forge.md tables render it and the linter enforces agreement.

Full change list in CHANGELOG.md.

New in v1.6.0:

• Express mode — first-class feature_mode: express: a 4-phase combined pass (product-spec minimal → plan inline → implement → verify) for the smallest changes, escalatable to lite/standard.
• Design System Harvest (Phase 2H) — discover the project's existing design system (component library, tokens, Storybook, Tailwind/CSS config) and emit a read-only manifest so mockups, component decomposition, and verification stay grounded in real code (/speckit.product-forge.design-system-harvest).
• Spec Merge (Phase 10, living spec) — merge a shipped feature's delta specs into the canonical specs/ source of truth and archive the change with audit history (/speckit.product-forge.spec-merge).
• Sync-verify now 10 layers — adds Layer 8 (FE↔BE contract drift), Layer 9 (doc↔code), and Layer 10 (constitution↔code) on top of the seven artifact-pair layers.
• Structured journeys — journeys/ (journeys.yml + JRN-*.md) as the machine-readable E2E source of truth for spec, tests, and tracking.
• Automated WCAG-AA accessibility floor — Phase 8A generates one @axe-core/playwright check per journey and Phase 8B runs it.

Full change list in CHANGELOG.md.

New in v1.5.0:

• Portfolio view across all features — conflict matrix, dependency graph, suggested merge order (/speckit.product-forge.portfolio).
• Brown-field backfill — reverse-engineer a feature folder from existing code with a gaps report.
• Lite mode — 5-phase lifecycle for small features and bug fixes (feature_mode: lite).
• Phase 9.5 monitoring-setup, Phase 5.5 migration-plan, post-bridge i18n-harvest, Phase 9B experiment-design, and cross-cutting feature-flag-cleanup — all opt-in, artifact-producing.
• Schema v3 for .forge-status.yml (additive over v2, no action required).
• State-lock protocol prevents concurrent-writer corruption.
• Per-phase digests keep context budgets small across long lifecycles.
• Learning loop via .product-forge/lessons.md — retrospectives teach research.
• Skip-reason policy and sync-verify drift budget with opt-in cosmetic auto-resolve.
• Release-readiness actually creates monitoring dashboard and flag registry instead of only checking them.

Full change list in CHANGELOG.md.