claude-autopilot

Use when the user invokes /slim-task or requests structured multi-phase task execution (structured task, SOP execution, parallel sub-agent dispatch) with blind-audit quality review.

Use when autopilot orchestrator enters Phase 1 to understand requirements, drive multi-round clarification, and produce the structured requirement packet checkpoint.

Use when the autopilot orchestrator enters Phase 5 after Phase 4 test case design has been approved and is about to choose an implementation path (parallel, serial, or TDD) and dispatch implementation work.

Use when the autopilot orchestrator reaches Phase 7 after Phase 6 tri-path reports are available and must finalize the delivery by collecting summaries, extracting knowledge, previewing Allure, and evaluating archive readiness.

Translates Figma designs into production-ready application code with 1:1 visual fidelity. Use when implementing UI code from Figma files, when user mentions "implement design", "generate code", "implement component", "figma to code", "figma 转代码", "figma 还原", "实现设计稿", provides Figma URLs, or asks to build components matching Figma specs. For Figma canvas writes via `use_figma`, use `figma-use`. For full-screen design generation in Figma, use `figma-generate-design`.

Use when the autopilot orchestrator main thread is about to dispatch a phase-specific sub-agent via the Task tool and needs the JSON envelope contract, explicit path injection, and parameterized prompt templates. Not for direct user invocation; skip when the current context is not the autopilot orchestrator.

Use when the autopilot orchestrator main thread reaches Phase 2 or Phase 3 and must dispatch the mechanical OpenSpec change creation and Fast-Forward artifact generation as background tasks against the prepared requirements bundle.

Use when the autopilot orchestrator enters Phase 4 to design test cases. Applies after Phase 3 alignment with tasks.md ready; gates Phase 5 dispatch.

Use when the autopilot orchestrator reaches Phase 6 after Phase 5.5 Red Team has passed and must generate test reports with tri-path parallel execution covering testing, code review, and quality scans before archival.

Use when the user requests an end-to-end spec-driven delivery pipeline (requirements → OpenSpec → implementation → testing → archive); orchestrates Phase 0-7.

Use when the user invokes /daily-report or asks in natural language to fill out, write, generate or submit an internal daily report (填写日报/写日报/生成日报/提交日报), aggregating recent git commits and Lark chat history into the company time-tracking system.

Use when the parallel-harness orchestrator enters the dispatch phase after planning and must spawn parallel sub-agents per batch schedule, construct task contracts as agent prompts, and handle failure retry/downgrade. Not for direct user invocation. 当 parallel-harness 编排器在规划完成后进入派发阶段、需要按批次调度并行启动子代理、构造任务契约 prompt 并处理失败重试/降级时使用；不面向用户直接调用。

Use when the parallel-harness orchestrator enters the planning phase and must turn user intent into a structured task DAG with file-ownership isolation, conflict detection, and a batch schedule. Not for direct user invocation. 当 parallel-harness 编排器进入规划阶段、需要将用户意图转化为带文件所有权隔离、冲突检测与批次调度的结构化任务 DAG 时使用；不面向用户直接调用。

Use when the user requests parallel multi-task engineering work that benefits from task-graph planning, file-ownership isolation, and parallel sub-agent execution with multi-gate verification (test/lint/type/security/policy); orchestrates plan -> dispatch -> verify -> synthesize phases. 当用户提出需要并行执行的多任务工程需求（依赖任务图规划、文件所有权隔离、并行子代理与多维度门禁验证）时使用；编排规划 → 派发 → 验证 → 综合四个阶段。

Use when the parallel-harness orchestrator enters the verification phase after dispatch and must run multi-gate quality checks (test, lint, type, security, policy/ownership, coverage), classify gate results as blocking/non-blocking, and emit a PASS/BLOCK conclusion. Not for direct user invocation. 当 parallel-harness 编排器在派发完成后进入验证阶段、需要运行多维度门禁检查（测试 / Lint / 类型 / 安全 / 策略与所有权 / 覆盖率）并将结果分类为阻断或非阻断、综合产出 PASS/BLOCK 结论时使用；不面向用户直接调用。

Use when the autopilot orchestrator or pre-commit hook needs deterministic static detection of documentation drift between SKILL.md, runtime scripts, CLAUDE.md and their README / .dist-include / root plugin table targets.

Use when the autopilot orchestrator has finished Phase 7 archive and needs to aggregate Episode records into the L1/L2/L3 learning tiers, cluster failure/success patterns, and emit L3 promotion candidates. ONLY for autopilot orchestrator; not for direct user invocation.

Use when the autopilot orchestrator restarts after a crash or interruption and must scan existing checkpoints, detect partial state, and determine the safe resume point. Not for direct user invocation; skip outside the orchestrator main thread.

Use when the user runs /autopilot-docs-fix scan|apply or asks to convert documentation drift candidates into git-applicable patches or manual suggestion templates under human confirmation.

Use when an autopilot phase boundary is reached and gate verification + checkpoint + event emission are required before advancing to the next phase.

Use when the autopilot orchestrator dispatches an adversarial Critic sub-agent just before a phase gate and needs that sub-agent to score the phase artifacts against structured rubrics and emit risk-report-phase{N}.json into the change context for downstream gate prechecks. ONLY for autopilot orchestrator; not for direct user invocation.

Use when a user initializes autopilot in a project that lacks .claude/autopilot.config.yaml, requests /autopilot-setup, asks to bootstrap autopilot configuration, or needs autopilot to detect project tech stack, services, agents, and test suites for the first time.

Use when the user wants a read-only static audit of test rot — for example asks to '检测测试腐烂' or 'audit tests', suspects deleted runtime scripts are still referenced in tests, hook files were renamed, weak assertions or duplicate case names may have crept in, or a candidate list at .cache/spec-autopilot/test-rot-candidates.json needs to be regenerated for manual review before any test edits.

Use when the user wants to consume the test-rot candidate list at .cache/spec-autopilot/test-rot-candidates.json and generate repair artifacts — deterministic sed-style patches for R1 orphan references and .suggestion.md files (including weak-assertion templates) for the rest. Typical triggers: '/autopilot-test-fix scan', '修复测试腐烂'. Does not directly modify tests/*.sh; application is a separate manual step via apply-fix-patch.sh.

Use when the user wants to quantify test effectiveness — running lightweight shell-layer mutation testing and static health scoring to confirm the test suite would actually catch regressions. Typical triggers include weekly CI sweeps, on-demand audits asking 'how healthy are our tests', or '/autopilot-test-health score|mutate|all' invocations. Not part of any phase gate.

Use when the autopilot orchestrator main thread enters Phase 0 to bootstrap a session (env checks, config load, recovery decision, lockfile, anchor commit). Not for direct user invocation.

Use when the autopilot orchestrator has completed Phase 5 implementation and must run a Red Team adversarial critic pass before Phase 6 reporting begins, validating the implementation against attack scenarios. ONLY for autopilot orchestrator; not for direct user invocation.

Use when the user wants to discover, install, list, hot-swap, or get recommendations for phase-specific autopilot sub-agents sourced from community catalogs (OMC, Anthropic official, VoltAgent, etc.), or when the autopilot setup flow needs to resolve a missing agent for a given phase.

Use when the user wants to view or change the per-phase AI model routing strategy for autopilot — for example choosing a cost-optimized, balanced, or quality-max preset, inspecting the current routing table, or customising an individual phase's model binding.