Skills
96,770 packaged capabilities and playbooks for agents. Install any of them with one command.
Agent workflows
A skill that converts the tacit knowledge of existing features into specifications and generates an as-is Context Pack (a documentation folder organized by topic). Triggered by keywords "기능 문서화", "기존 기능 분석", "as-is 정리", "docs/기능별설명 생성". If you need a Spec for a new feature, use peach-gen-spec.
Coordinate installed skills and native subagents for bounded multi-stage outcomes, parallel work, context isolation, or independent judgment. Use when the user requests delegation or the desired outcome materially benefits from several owned results. Exclude single-skill routing, host setup, scheduling, and persistent agent runtimes.
Implements Manus-style file-based planning for complex tasks. Creates task_plan.md, findings.md, and progress.md. Use when starting complex multi-step tasks, research projects, or any task requiring >5 tool calls. Now with automatic session recovery after /clear.
Transforms workflow to use Manus-style persistent markdown files for planning, progress tracking, and knowledge storage. Use when starting complex tasks, multi-step projects, research tasks, or when the user mentions planning, organizing work, tracking progress, or wants structured output.
Search community legal skills in watched registries, show matches with their descriptions, and offer the option to view the full SKILL.md before installing. Use when the user says "browse," "search skills," "find a skill for X," "what's available," or wants to add a new registry to the watch list.
Recommend community skills based on recent activity in other plugins. Checks whether the community has already built something relevant to a task and mentions it once, non-intrusively. Use when the user says "is there a community skill for this", "what else is available", or asks for skill recommendations; also runs as a passive part of other plugin workflows.
Plan or run a measured goal loop with anti-goal guardrails. Planning mode drafts metrics, discovery, and work structure without changing the workspace. When the user asks to run, delegate, automate, or resume the loop, execution mode may run bundled local helper scripts, write durable `.okra` records, spawn subagents, and optionally route writer work through an already-installed Codex or Claude CLI. Use for OKRs, metric-led improvement, bounded research, or progress that must preserve budget, quality, safety, trust, or another measured constraint. All actions remain inside the user's request and runtime permissions; the skill does not install tools, add credentials, grant authority, or contact external services on its own.
Use when Codex or code/plan review findings need Definition-of-Done scope triage, especially to separate must-fix issues from minimal hygiene, out-of-scope hardening, repeated noise, or self-induced refinements. Also use when codex-review-cycle invokes scope guard between validity checking and summary render. Do not use for single-shot lint or unrelated changes.
Orchestrates the RLM repo workflow end-to-end with phase gates, locked artifacts, addenda, traceability, and automatic bootstrap/upsert of AGENTS/PLANS scaffolding. Trigger phrases: "Implement requirement <run-id>", "Run RLM Phase <N>", "resume requirement", "lock Phase <N>", "verify locks".
Query decomposition and multi-source search orchestration. Breaks natural language questions into targeted searches per source, translates queries into source-specific syntax, ranks results by relevance, and handles ambiguity and fallback strategies.
Sets up an `## Agent skills` block in AGENTS.md/CLAUDE.md and `docs/agents/` so the engineering skills know this repo's ticket tracker (GitHub or local markdown), triage label vocabulary, domain doc layout, and review-prep conventions (commit style and breadcrumb schema). Run before first use of `to-tickets`, `to-spec`, `tdd`, `improve-codebase-architecture`, `wayfinder`, or `prepare-for-review` — or if those skills appear to be missing context.
Use this skill to run the Shipshape Shipwright role: in-harbour code inspection. Discovers existing behaviour and policy violations from production code, writes @captain-tagged scenario skeletons and @planks annotations for Captain review. Run for fitting out, onboarding an existing codebase, and between releases.
High-performance routing engine for AI AgentSkills. Classifies user intent, scans for specialized skills, and generates optimal dispatch decisions (HANDOFF, SEQUENCE, or NO_MATCH). Use this to manage complexity in large skill environments and ensure the best-equipped skill handles every task.
Package and distribute Claude Code skills for sharing via GitHub, Claude.ai uploads, or team deployment. Creates install scripts, documentation, and .skill packages. Use when user says "publish skill", "share skill", "package skill", "distribute skill", or "release skill".
Install community skills from watched registries. Reads the allowlist first, fetches, shows the raw SKILL.md (not just a summary), runs structural trust checks, runs skills-qa, and writes files only after explicit user approval. Use when the user says "install [skill]," picks install while browsing, or provides a direct skill URL.
Reference: detailed uninstall, disable, and re-enable workflows for community skills installed through the legal builder hub. Safe by default — refuses to touch first-party plugin skills, confirms before deleting files, and logs every action. Loaded by the /legal-builder-hub:uninstall and /legal-builder-hub:disable skills.
Evaluate a skill against the legal skill design framework — thirteen design parameters (including trust surface, freshness, schema validation, and conflict detection), three legal failure modes, and a three-tier verdict (ready / some concerns / major concerns). Use before deciding whether to trust a community skill for installation, before deploying a first-party skill to the team, or when the user asks "should I trust this?" or "is this skill well designed?" Runs automatically as part of /legal-builder-hub:skill-installer.
Compatibility entry point for deepening one non-trivial area of a roadmap design without reopening the whole interview or generating tasks. Use when the user says /slice, names a UX/API/data/worker/infra/security area, or provides a legacy slice that must be reconciled.
Draft specs, analyze requirements, plan implementation, implement from specs, review against specs (generator≠evaluator), extract specs from existing code, run surgical bugfix specs, multi-session handoff/progress, and size greenfield/brownfield work. Use when the user wants testable requirements, living specs, spec-driven development, session handoff, bugfix with regression fences, Design-First vs Requirements-First, or /draft-spec /plan-impl /implement-spec /review /extract-spec. Do not use for C4/solution architecture packets (use solution-architecture), DevOps/CI-CD strategy design, README authoring, Agent Skill optimization, AFK/Ralph automation products, or trivial one-line fixes.
Use when a user-gate task has requiresUserSpecification=true OR the agent's "do I know HOW?" self-check returns no. Locks down verification mechanics through a short AskUserQuestion sequence and writes the answers back into the task's metadata. Does NOT run the verification itself.
Create new spec-driven tickets and resume existing ones through the full lifecycle: research, specification, planning, and implementation. Use when: the user describes new work to be done; continuing a previously started ticket; a feature, bug fix, or task needs planning and implementation; starting or resuming planned work on a project with specs/. Triggers on: 'new ticket', 'new feature', 'fix bug', 'implement', 'create ticket', 'I want to build', 'let us work on', 'add feature', 'continue ticket', 'resume work', 'pick up where we left off', 'work on ticket', 'existing ticket', 'check ticket status', 'what is the state of ticket'.
Documentation-first development methodology. The goal is AI-ready documentation - when docs are clear enough, code generation becomes automatic. Triggers on "Build", "Create", "Implement", "Document", or "Spec out". Version 3.5 adds Phase 2.5 Adversarial Review and renames internal verification to Spec Gate (structural completeness). Clarity Gate is now a separate standalone tool for epistemic quality.
Full session shutdown — digest, retro, merge worktree branches, worktree cleanup, close session state, and commit handoff. This is the single entry point for ending a session. Triggers on: 'teardown', 'clean up', 'wrap up', 'session end', 'end session', 'close session', 'done', 'log off', 'sign off', or automatically when the decision budget is reached.
Use when the user has a coding prompt — a feature, bug fix, or refactor — and wants it implemented end to end by a self-checking software loop, not a single pass. It refines the prompt into an executable plan (running the plan-loop internally), then executes the plan task by task: an Engineer subagent implements each PR-sized task editing only source, a separate QA subagent authors the tests that prove the task's acceptance criteria (or grades quality only when tests already cover them), runs the task's tests plus the full accumulated regression suite, and grades the code against a strict conciseness/readability/style-match rubric; the Engineer and QA loop on each task until the tests pass and the quality gate holds, then it commits and opens a pull request for the task — stacked PRs (one per task, each based on the prior) by default, or one combined PR — and moves to the next task. Author and critic have disjoint write scopes (Engineer owns source, QA owns tests) so neither can game the gate. Not for producing a plan without building it (that is plan-loop), and not for tuning a metric on an existing artifact under a fixed correctness bound (that is optimize-loop).
Multi-AI Agent Orchestration System with configurable models and role-based workflows. Use when you need to coordinate multiple AI agents (Claude, Gemini, Codex) for complex tasks like planning, code generation, analysis, review, and execution. Supports agentic workflow patterns: parallel specialists, pipeline, and swarm orchestration. Compatible with Claude Code, Cursor, and OpenCode. Triggers: 'orchestrate agents', 'multi-agent workflow', 'plan and execute', 'code review pipeline', 'run synapse', 'agentic workflow'.
Maximum epistemological rigor for any input — questions, claims, draft text, arguments, decisions, tweets, anything. Always runs the full truth-seeking pipeline: strips prestige, malformed framing, false confidence, adversarial framing, and heuristic overreach. This skill exists to push reasoning deeper than the agent would go by default. Use for anything where you want honesty over agreeableness. Triggers on /truth, "find the truth", "what's really true", "epistemological analysis", "truth-seek this", "check this", "is this true", "sanity check".
Improve one existing shipped skill from concrete feedback or a failed run via bounded PDCA. Use when skill instructions or package behavior look wrong and the fix belongs in that one package; otherwise route to the owning workflow. Triggers: tune-skill.
Uninstall community skills installed through this hub. Confirms before deleting files, refuses to touch first-party plugin skills, and logs every action. Use when the user wants to fully remove a community skill ("uninstall [skill]," "remove this skill") rather than just disable it.
Master orchestration workflow for reverse engineering complex Web/JS sites. Covers reversing sign/token/cookie/header/body/websocket fields, heavy obfuscation, junk instructions, control flow flattening, JSVMP, worker/wasm, browser vs. Node differentials, environment patching, local reproduction and regression, and case leads such as Akamai/Kasada/PX/reese84/Tongdun/a_bogus/Tencent captcha/Alibaba captcha/JSVMP/227/226/wasm/protobuf/rid/fuid/fs/bx-pp/run_js/storage.estimate/animationend. By default it coordinates the three MCPs jshook + js-reverse + chrome-devtools-mcp for debugging and analysis, switching to specialized skills at the locate, recover, runtime, env-patch and replay stages.
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.