Spec Audit En

spec-audit-en

Treat spec documents as the Single Source of Truth. Detect implementation gaps and create GitHub Issues.

Prerequisites

• Claude Code environment
• gh CLI (for Issue creation)

Arguments

• No arguments: Auto-discover docs and run full check
• Path (e.g., /spec-audit-en docs/api-spec.md): Check specific document(s)

Phase 1: Discovery

If no path argument is given, auto-discover documents from the project root using Glob:

• docs/, spec/ directories
• README.md, ARCHITECTURE.md, CLAUDE.md
• OpenAPI/Swagger definitions (*.yaml, *.json)
• Other .md files

Show the discovered document list to the user and confirm the check targets.

Detection granularity is per function/endpoint. Report specific items like "DELETE /users/:id is not implemented", not entire sections.

Detection covers all of:

1. Spec vs Implementation: Features defined in specs but not implemented
2. TODO/FIXME/HACK: Incomplete markers in code
3. Skipped tests: skip, pending, xit, xdescribe, etc.
4. API schema gaps: OpenAPI/Swagger definitions vs actual handlers

Phase 2: Document Analysis

1. Read target documents determined in Phase 1
2. Extract concrete claims and specifications:
   • Endpoint definitions, function signatures, CLI arguments
   • Data models, table definitions
   • Feature lists, behavior descriptions
   • Config items, environment variables
3. Create tasks with TaskCreate to track check targets

Phase 3: Implementation Cross-check

Refer to references/check-criteria.md for detection criteria.

3-1: Spec vs Implementation

1. Extract claims from documents (function names, paths, config values, behaviors) one by one
2. Search for corresponding implementation using Explore agent or Grep/Glob
3. Determine: matches / mismatches / not implemented
4. Record document-side line numbers and implementation-side file:line for mismatches

3-2: TODO/FIXME/HACK

1. Grep for TODO, FIXME, HACK, XXX, TEMP
2. Record each marker's content, file, and line number
3. Link to spec if related

3-3: Skipped Tests

1. Grep for:
   • JS/TS: it.skip, describe.skip, xit, xdescribe, test.skip
   • Go: t.Skip
   • Python: @pytest.mark.skip, @unittest.skip, self.skipTest
   • Ruby: skip, pending
   • Others: adjust based on project language
2. Record skip reason comments, file, and line number

3-4: API Schema Gaps

1. Read OpenAPI/Swagger definitions if present
2. Search for handlers/routes corresponding to defined endpoints
3. Record unimplemented endpoints

Phase 4: Report Generation and Output

1. Aggregate results by severity (see references/check-criteria.md)
2. Generate report using templates/report.md format
3. Ask user via AskUserQuestion for output destination:
   • Create GitHub Issues (Recommended)
   • Save as local MD file: Generate spec-audit-report.md at project root
   • Console output: Display report in conversation

Phase 5: Issue Creation

When "Create GitHub Issues" is selected:

1. Show a summary list of all findings (title + severity)
2. Ask via AskUserQuestion for Issue creation mode:
   • Single Issue (Recommended): Combine all findings into one Issue with a checklist
   • Individual Issues: Create one Issue per finding
3. Duplicate check: gh issue list --state open --search "<summary>" for existing Issues. If duplicate found: ask to skip or create new
4. Issue body follows templates/issue.md
5. gh issue create with labels: spec-audit + severity label (severity:critical, severity:high, severity:medium, severity:low)
6. Report created Issue URL(s) to user

Rules

• Never report gaps based on speculation. Verify against actual implementation code
• Show both document-side and implementation-side locations for every gap
• Do not modify code. Deliverables are Issues and reports only
• Always get user approval before creating Issues
• Track progress with TaskCreate/TaskUpdate
• When many gaps are found, prioritize severity 🔴🟠; 🟡🟢 may be omitted