claude-harness

Use when executing implementation plans with independent tasks in the current session

You MUST use this before any creative work - creating features, building components, adding functionality, or modifying behavior. Explores user intent, requirements and design before implementation.

Use when completing tasks, implementing major features, or before merging to verify work meets requirements

Use when starting any conversation - establishes how to find and use skills, requiring skill invocation before ANY response including clarifying questions

Use when creating new skills, editing existing skills, or verifying skills work before deployment

Use when you have a spec or requirements for a multi-step task, before touching code

Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes

Use when implementation is complete and tests pass — guides completion of development work by executing merge, push, and cleanup, or presenting interactive options based on finish-mode variable

Use when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via native tools or git worktree fallback

Generate a well-formed SKILL.md from a structured workflow context. Use this skill whenever the self-evolution reviewer (auto via Stop hook OR manual via /evolve-review) decides to CREATE or UPDATE a skill in ~/.claude/skills/. Returns complete SKILL.md content following Anatomy, Progressive Disclosure, naming conventions, and frontmatter schema; does NOT run evals or open viewers.

Use when viewing or modifying superpowers-pro plugin variables. Invoke without args to list all variables, or with name and value to change a variable.

调研驱动的 PRD 生成 — 目标对齐→竞品调研→问题提取→二次澄清→方案扩展→MVP 收敛→PRD 撰写→验证审查。Use when users ask to create a PRD, write product requirements, document a feature, 需求文档, 产品需求, or need help structuring product specifications.

Use when systematically scanning a specified domain for all potential issues — not diagnosing a single known bug. Triggers include "check X for all problems", "audit module Y", "find issues in Z", or starting the /fix workflow.

Use when evaluating code for refactoring — assessing structure, maintainability, performance, or testability. Triggers include "refactor X", "restructure Y", "clean up Z", or starting the /refactor workflow.

Generate comprehensive Product Requirements Documents (PRDs) for product managers. Use this skill when users ask to "create a PRD", "write product requirements", "document a feature", or need help structuring product specifications.

Use when designing system architecture for a 0→1 project after PRD is validated — produces implementable architecture with tech selection comparisons, code structure design, algorithms, state machines, data schemas, and ADRs. Triggers on: design architecture, system architecture, architect from scratch, architecture design, 架构设计, 系统架构

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.

Add a new source to the wiki and integrate it into existing pages. Use when the user says "ingest this", "add this to the wiki", "process this source", or drops a file into raw/. Handles reading the source, discussing key takeaways, creating a wiki page, updating the index, updating affected pages, and logging the operation.

Health-check the wiki for consistency and completeness. Use when the user says "lint the wiki", "health check", "what's missing", "check the wiki", or periodically after multiple ingests. Scans for contradictions, stale claims, orphan pages, missing cross-references, and gaps.

Answer questions from wiki content. Use when the user asks about a topic that might be covered in the wiki, says "look up X in the wiki", "what does the wiki say about Y", or asks about AI coding tools, features, configuration, or any topic the wiki covers. Routes to the right wiki pages via the index, reads them, and synthesizes an answer.

Build Claude Code plugins and plugin marketplaces from scratch. Use this skill whenever the user wants to create a plugin, build a marketplace, publish skills/agents/hooks as a distributable plugin, migrate standalone .claude/ configs to a plugin, or set up a plugin distribution system. Also use when the user mentions plugin development, plugin packaging, marketplace creation, plugin marketplace setup, sharing plugins with a team, or submitting plugins to the official Anthropic marketplace. Covers the full lifecycle: plugin structure, manifest, skills, agents, hooks, MCP/LSP servers, marketplace.json, hosting, versioning, and distribution. **Reference implementation**: see `references/codex-plugin-cc.md` for an architecture analysis of the openai/codex-plugin-cc production plugin (https://github.com/openai/codex-plugin-cc), covering companion scripts, embedded subagent skills, SessionStart/SessionEnd/Stop hooks, and command frontmatter patterns.

Claude Code 官方文档参考技能。When working with Claude Code features, configuration, hooks, skills, MCP, permissions, settings.json, CLI commands, slash commands, sub-agents, Agent SDK, or any other Claude Code internals — either when users ask OR when the Agent needs to know how something works. Fetch from https://code.claude.com/docs/ as the authoritative source instead of relying on training data. Make sure to use this skill whenever the Agent encounters a Claude Code feature it doesn't fully understand, needs to configure settings.json, create hooks/scripts, use MCP servers, work with skills, or needs accurate information about Claude Code commands, permissions, or APIs.

当在本仓库中审查功能改动、缺陷修复、Pull Request 或 git diff，且需要按业务功能实现、代码质量、架构合理性和项目 rules 四个部分输出固定中文审查结果时使用。

以内容优先、证据支撑的方式，从现有代码库生成详细的架构解析、工作流解析与参考文档体系。

Cursor AI 编辑器官方文档参考技能。When working with Cursor features, hooks, skills, MCP, CLI commands, rules, permissions, configuration, or any other Cursor internals — either when users ask OR when the Agent needs to know how something works. Fetch from https://cursor.com/docs as the authoritative source instead of relying on training data.

Generate a well-formed SKILL.md from a structured workflow context. Use this skill whenever the self-evolution reviewer (auto via Stop hook OR manual via /evolve-review) decides to CREATE or UPDATE a skill in ~/.claude/skills/. Returns complete SKILL.md content following Anatomy, Progressive Disclosure, naming conventions, and frontmatter schema; does NOT run evals or open viewers.

OpenAI Codex 官方文档参考技能。When working with OpenAI Codex features, configuration, hooks, skills, MCP, AGENTS.md, CLI, plugins, subagents, sandboxing, or any other Codex internals — either when users ask OR when the Agent needs to know how something works. Fetch from https://developers.openai.com/codex as the authoritative source instead of relying on training data. Make sure to use this skill whenever the Agent encounters a Codex feature it doesn't fully understand, needs to configure Codex settings, create hooks/scripts, use MCP servers, work with skills, or needs accurate information about Codex commands, permissions, or APIs.

OpenCode AI CLI 官方文档参考技能。When working with OpenCode features, configuration, plugins, skills, MCP, permissions, agents, custom tools, TUI, CLI commands, SDK, or any other OpenCode internals — either when users ask OR when the Agent needs to know how something works. Fetch from https://opencode.ai/docs/ as the authoritative source instead of relying on training data. Make sure to use this skill whenever the Agent encounters an OpenCode feature it doesn't fully understand, needs to configure opencode.json, create plugins/tools, use MCP servers, work with skills/agents, or needs accurate information about OpenCode commands, permissions, or APIs.

Review and fix UI/UX issues in web pages. Use whenever the user reports layout bugs, chart failures, or asks to improve page design.

内部技能——从 wiki 页面提取交叉引用，构建图谱数据。由 serve 技能自动调用。

内部技能——从 raw + wiki 内容构建搜索索引。由 serve 技能自动调用。

当需要理解模块对外暴露的接口、协议、参数格式、或调用方式时使用。读取结构地图，逐模块提取 API 接口文档。

当需要理解模块的数据结构、实体关系、字段定义、或数据生命周期时使用。读取结构地图，逐模块提取数据模型文档。

当需要理解模块的业务流程、状态转换、时序关系、或异常处理方式时使用。读取结构地图，逐模块提取流程文档。

当需要对代码库进行多维度深度分析时使用。路由到具体维度的提取技能，或一键全量提取。scan 的自然后续步骤。

当需要将 raw 提取产出转化为结构化 wiki 页面、建立交叉引用时使用。路由到 ingest 和 cross-ref 技能。extract 的自然后续步骤。

当需要在已转化的 wiki 页面之间建立交叉引用、发现跨维度关联、或生成综合分析时使用。ingest 的自然后续步骤。

当需要理解模块内部结构、职责边界、上下游交互方式、或关键设计决策时使用。读取结构地图，逐模块产出深度拓扑分析文档。

当需要理解模块的核心概念、领域术语、概念关系、或命名规范时使用。读取结构地图，逐模块提取概念术语文档。

当需要将 raw 提取文档转化为结构化 wiki 页面时使用。逐维度读取 raw 产出，按模板生成 entity/concept/synthesis 页面。