Skip to content

V3 Status

V3 is concrete enough to stay local-first and continue without hosting or CLI work.

Covered

  • MCP config security review through review_mcp_config_security.
  • Dedicated deterministic V3 eval reporting through run_v3_eval_harness.
  • Stronger generated AGENTS.md and llms.txt scoring through score_agent_artifacts.
  • Source-backed stack-pack promotion through promote_stack_pack_to_files, including dry-run outputs, readiness metadata, manifest diffs, overwrite blocking, and explicit version bumps.
  • Promoted source-backed stack packs for Hono, Zod, Vitest, Auth0, Stripe, Expo, and Vercel AI SDK.
  • Executable detector coverage for auth boundaries, payment boundaries, test policy, validation boundaries, route thinness, screen/component size, and AI tool safety.
  • Client integration recipes through list_client_integration_recipes and JSON examples under examples/.
  • Local MCP config file scanning through scan_mcp_config_files.
  • Final response review through review_agent_final_response.
  • Release readiness through npm run check:v3, covering typecheck, tests, build, audit, package dry-run inclusion, and mcp_readiness_report.
  • Hosted API request/response boundaries documented in docs/hosted-api-shape.md without adding hosting state.
  • Expanded owned V3 fixtures for Expo, Auth0, Stripe, Supabase, and AI SDK boundaries.
  • Focused V3 tests for security, artifact scoring, evals, recipes, promotion, and self-contract review.
  • Schema hygiene: src/tools/schemas.ts is now a thin barrel over focused modules in src/tools/schemas/.

Remaining Future Work

  • GitHub-backed memory adapter with repo, branch, and PR workflow.
  • Hosted/team memory, analytics, review inbox, and persistence.
  • CLI packaging and local command UX.
  • Optional typed client SDK helpers around the recipe JSON.

Decision

Treat V3 as implemented for the local-first MCP core. The next work should be hardening and expanding stack packs, not starting hosting or CLI unless the product direction changes.

The completion audit lives in docs/v1-v3-completion-audit.md. V4 scope lives in docs/v4-scope.md; the V4 scope audit lives in docs/v4-scope-audit.md. Advanced compatibility criteria live in docs/v5-scope.md, followed by local governance maturity in docs/v6-scope.md, local strategic intelligence in docs/v7-scope.md, and local governance automation in docs/v8-scope.md.

Released under the MIT License.