V1-V3 Completion Audit

Audit date: 2026-05-05

Scope: local-first architect-mcp behavior from V1 through V3. Deferred V3 adoption/productization work is excluded: hosted service, database/storage, GitHub-backed memory adapter, CLI UX, and typed client SDK.

Verdict

V1 through V3 local-first scope is complete and release-gated.

Current evidence:

• npm run check:v3 passes.
• MCP self-review reports gate: pass, findings: 0, and score: 100.
• Stack-pack validation passes.
• Foundation-pack validation passes.
• Tool documentation sync tests pass.
• Hosted mode does not expose local workspace scanning.
• Package dry-run includes required V3 docs, examples, scripts, packs, and built output.

V1 Scope

Complete:

• Ask the next important project-intake question before implementation.
• Run /grill-me intake to identify blockers, challenges, readiness, and next focused questions.
• Generate architecture contracts with repo structure, module boundaries, file rules, and agent instructions.
• Generate repo artifacts including AGENTS.md, architecture contract docs, Cursor rules, .architectignore, and build plan docs.
• Review supplied file trees and local workspaces for architecture and repo-hygiene violations.

V2 Scope

Complete:

• Resolve stack-specific architecture packs from requested or inferred stack hints.
• Enrich generated contracts with framework/database/auth/payment-specific boundaries.
• Export repo-level agent instructions for AGENTS.md, CLAUDE.md, and Cursor rules.
• Review import direction, client/server leaks, direct database access in UI files, scattered env reads, and supported stack-pack file rules.
• Execute supported stack-pack fileRules during review.
• Apply file-type-aware line thresholds.
• Return stable finding codes and confidence levels.
• Generate review baselines so mature repos can fail only on new findings.
• Use TypeScript AST parsing during local scans for import and environment-access summaries.
• Enforce repo hygiene for oversized type/schema aggregation files, root source clutter, generated-file noise, manifest integrity, and release scripts.

V3 Scope

Complete:

• Guided-yolo harness for vague novice implementation intent.
• Ambiguity, blast-radius, reversibility, destructive-command, dependency-upgrade, and one-way-door gates.
• Pre-edit contracts and implementation drift review.
• Assumption ledger entries and stateless memory proposals.
• Triggered stack guidance from local ingested llms.txt snapshots.
• Source-backed stack-pack promotion workflow with review, conflict detection, versioning, dry-run output, and manifest updates.
• Promoted packs for Hono, Zod, Vitest, Auth0, Stripe, Expo, and Vercel AI SDK.
• Pack scoring and stack-pack coverage matrix.
• MCP config security review and local config scanner.
• Built-in skills catalog and supplied-skill metadata review.
• Agent artifact quality scoring for AGENTS.md and llms.txt.
• Final response honesty review.
• Combined review_agent_session for intent, contract drift, memory relevance, final response, and verification honesty.
• Hosted/local tool policy audit.
• Dedicated V3 eval harness.
• Hosted API shape documented without implementing hosted state.
• Release readiness gate through npm run check:v3.

Findings Fixed During Audit

• src/domain/stackPacks.ts was over the source-file hygiene threshold. Fixed by moving schema and manifest validation into focused modules.
• src/domain/stackPackWorkflow.ts was over the source-file hygiene threshold. Fixed by moving candidate-rule generation and pack-promotion file helpers into focused modules.

Deferred From V3

Still intentionally deferred:

• Hosted service and dashboard.
• Database/storage for review history or memory.
• GitHub-backed memory repository and PR adapter.
• CLI UX.
• Typed client SDK.

These are adoption/productization layers. They are not blockers for V4 Standards Depth or V4 Agent Workflow, but they must stay out of V4 and wait for the backend-concrete V5 gate.