claude-md-plugin

Use when the user wants to autonomously develop a feature end-to-end without manual steps. Runs requirements → CLAUDE.md → code generation as a pipeline. Autonomous execution from start to finish given only requirements, without step-by-step commands. Trigger keywords: auto develop, end-to-end, autonomous implementation

Converges an existing project to the current plugin schema. Uses fix-schema CLI's converge_schema to deterministically handle section rename/remove/add. No source version detection needed — target-state-driven migration.

Adds or updates the Conventions section (unified 6 subsections) in project/module CLAUDE.md. For existing projects, extracts conventions from source code; for new projects, collects them interactively. Use the --update option to modify an existing Conventions section (absorbs the former /convention-update).

Inter-session spec workflow step executor. Reads .claude/workflows/{dir-safe}/state.json and executes the next pending step. Use when a spec workflow was interrupted, or to run pipeline steps in separate sessions/CI jobs. Trigger keywords: resume spec, resume workflow, resume spec

Use this agent when performing 3-layer root cause analysis for a reported bug. Traces through CLAUDE.md (Layer 1), DEVELOPERS.md (Layer 2), and source code (Layer 3). Composes superpowers:systematic-debugging for evidence-based root cause tracing. For Layer 3 unambiguous cases: writes failing test → fixes code → verifies. Returns a structured result block for the SKILL to act on. <example> <user_request> Session file: .claude/tmp/bugfix-session-src-auth.md Target path: src/auth </user_request> <assistant_response> 1. Session read — E: "AuthError on expired token", A: "User returned" 2. Layer 1: REQ-2 found — "expired tokens rejected" — E matches S ✓ 3. Layer 2: CONST-2 found — "max 7 days" — constraint exists ✓ 4. Layer 3: source_changed=false, changed_requirements=[] — code diverged from spec 5. Judgment: unambiguous (E == S AND A != S) 6. [RED] wrote test_authenticate_rejects_expired_token — fails as expected 7. [GREEN] fixed authenticate() at auth.ts:42 — added expiry check 8. Tests: 1 passed, 0 failed ---bugfix-result--- status: fixed root_cause_layer: 3 judgment: unambiguous fix_type: code_fix fix_description: Added token expiry check in authenticate() at auth.ts:42 test_result: passed ---end-bugfix-result--- </assistant_response> </example> <example> <user_request> Session file: .claude/tmp/bugfix-session-src-payment.md Target path: src/payment </user_request> <assistant_response> 1. Session read — E: "refund within 30 days", A: "refund rejected after 14 days" 2. Layer 1: no Requirement found for refund window — S = null 3. Judgment: ambiguous (S == null) ---bugfix-result--- status: escalated root_cause_layer: unknown judgment: ambiguous fix_type: none fix_description: CLAUDE.md has no Requirement for refund window. Cannot determine if code or spec is wrong. test_result: skipped escalation: expected: "refund within 30 days" actual: "refund rejected after 14 days" spec: "none" reason: "S == null — missing requirement in CLAUDE.md" choices: [A, B, C] ---end-bugfix-result--- </assistant_response> </example>

Use this agent when analyzing source code to generate CLAUDE.md drafts for a single directory. Orchestrates CLI tools (resolve-boundary, analyze-code, format-analysis) and generates documents directly. Input is a pre-extracted session file with tree info and children context. <example> <context> The decompile skill calls decompiler agent with a session file for each directory in leaf-first order. </context> <user_request> Session file: ${TMP_DIR}decompile-session-src-auth.md Target: src/auth Save results to ${TMP_DIR} and return only the path </user_request> <assistant_response> ---decompiler-result--- status: success target_dir: src/auth validation: passed developers_md: generated ---end-decompiler-result--- </assistant_response> </example>

Use this agent when a large spec needs to be split into individual spec units. Analyzes natural language requirements and produces a module decomposition plan: target paths, requirement distribution, tree structure, and dependency order. Does NOT write CLAUDE.md — that is impl agent's responsibility. Returns result as a file to protect SKILL context window. <example> <context> The spec skill calls decompose agent before dispatching impl agents. </context> <user_request> Session file: ${TMP_DIR}decompose-session.md Save results to ${TMP_DIR} and return only the path </user_request> <assistant_response> 1. Scope Classification: multi (3 independent purpose groups identified) 2. Module Identification: src/auth, src/payment, src/notification 3. Requirement Distribution: 12 requirements assigned, 0 unassigned 4. Tree Validation: INV-1 passed (flat siblings, no circular deps) 5. Result written: ${TMP_DIR}decompose-result.json ---decompose-result--- result_file: ${TMP_DIR}decompose-result.json scope: multi module_count: 3 ambiguous_count: 0 ---end-decompose-result--- </assistant_response> </example>

Use this agent when implementing production code to pass approved tests (GREEN phase). Receives approved tests and must make them pass with minimal implementation. NEVER modifies test assertions. Called by dev SKILL after test review loop approval. <example> <context> The dev skill calls green-coder with approved tests. </context> <user_request> Session file: ${TMP_DIR}green-session-src-auth.md Target directory: src/auth Detected language: typescript Save results to ${TMP_DIR} and return only the path </user_request> <assistant_response> 1. Session read — target: src/auth, language: typescript 2. Mapping loaded — 10 tests across 2 files 3. [GREEN attempt 1] Implementation generated 4. [GREEN attempt 1] Tests: 8 passed, 2 failed 5. [GREEN attempt 2] Fixed 2 failures 6. [GREEN attempt 2] Tests: 10 passed, 0 failed ---green-result--- result_file: ${TMP_DIR}green-result-src-auth.json status: success implemented_files: [src/auth/index.ts, src/auth/types.ts] tests_passed: 10 tests_failed: 0 ---end-green-result--- </assistant_response> </example>

Use this agent when critically reviewing a spec execution plan (plan.md) before CLAUDE.md generation. Applies Socratic method to verify Requirements completeness, Constraints precision, and Rationale traceability. Called by spec SKILL in the Socratic Loop, after impl agent produces plan.md and before mode=execute generates CLAUDE.md + DEVELOPERS.md. Returns verdict: approved | rejected with specific Critical Questions. <example> <context> spec SKILL calls impl-reviewer after plan.md is produced. </context> <user_request> Session file: .claude/tmp/spec-reviewer-session-src-auth-v1.md Save results to .claude/tmp/ and return only the path </user_request> <assistant_response> 1. Session read — plan_file: .claude/tmp/spec-plan-src-auth.md, round: 1 2. Plan loaded — 4 Requirements, 3 Constraints 3. Critique: - REQ-3: "handle appropriately" → unmeasurable expression - CONST-2: error type not specified - No Constraint corresponding to REQ-4 4. Verdict: rejected (3 Critical Questions) 5. Result written: .claude/tmp/spec-reviewer-result-src-auth-v1.md ---spec-reviewer-result--- result_file: .claude/tmp/spec-reviewer-result-src-auth-v1.md verdict: rejected round: 1 ---end-spec-reviewer-result--- </assistant_response> </example>

This skill should be used when the user reports a bug, unexpected behavior, or asks to "fix this bug", "debug this", "something is broken", "not working as expected", or uses "/bugfix". Traces root cause through 3 layers: CLAUDE.md (requirements) → DEVELOPERS.md (constraints) → source code. Fixes at the highest affected layer following Fix-Highest-Layer-First principle. Never patches code while leaving CLAUDE.md inconsistent. Trigger keywords: fix bug, debug, unexpected behavior, broken, not working

This skill should be used when the user asks to "decompile code to CLAUDE.md", "extract CLAUDE.md from code", "document existing codebase", "reverse engineer spec", "extract documentation from source", or uses "/decompile" or "/decom". Analyzes existing source code and creates CLAUDE.md + DEVELOPERS.md documentation for each directory. Uses parse-tree CLI for directory discovery, then runs decompiler agent per directory in leaf-first order. Trigger keywords: decompile, extract docs from code, document existing code

This skill should be used when the user asks to "develop from CLAUDE.md", "generate code from CLAUDE.md", "implement CLAUDE.md", "create source files", or uses "/dev". Processes changed CLAUDE.md files in the target path (or all with --all flag). Performs 4-agent TDD pipeline: test-writer → test-reviewer loop → green-coder → refactorer. Trigger keywords: code generation, develop, code from CLAUDE.md

This skill should be used when the user asks to "define requirements", "write spec", "create CLAUDE.md from requirements", "define behavior before coding", or uses "/spec". Analyzes natural language requirements and generates CLAUDE.md without implementing code. Follows ATDD principle: specification first, then code generation via /dev. Trigger keywords: define requirements, write spec, spec first

This skill should be used when the user asks to "validate CLAUDE.md", "check documentation-code consistency", "verify specification matches implementation", "check for drift", "lint documentation", "resolve drift", "fix documentation mismatch", "sync docs with code", or uses "/validate". Runs deterministic CLI validation (schema, convention structure, boundary, DEVELOPERS.md existence) and semantic validator agent for comprehensive drift detection, with interactive auto-fix. Trigger keywords: validate CLAUDE.md, document verification, drift check, resolve drift

1 hook across 1 event