Agent Workflow Skills

Classifies changed files into review treatment tiers before any reviewer decides SKIP. Prevents reviewers from misclassifying LLM prompt engineering artifacts (agents/*.md, skills/*/SKILL.md, CLAUDE.md, rules/*.md) as documentation-only. Load this skill when reviewing any diff that contains markdown or prose files. Authoritative definition is verdict-schema.md §2.5.

Quantitative cost measurement for technical research — token injection costs, payload sizes, context window consumption, and file-level counts from actual repo files. Use when a technical-researcher orchestrator needs the cost dimension of adding or changing something: how many tokens will it inject, how big are the artifacts, what is the context window impact? This is NOT blast-radius analysis (which files break) — it covers size, tokens, and performance cost only.

Spawn and manage persistent interactive Claude Code CLI sessions with bidirectional communication via tmux. Provides spawn, send, read, status, list, and kill subcommands for orchestrating parallel peer sessions. Uses built-in --worktree and --tmux flags. Sessions stay alive for multi-turn steering. Triggers on "spawn claude session", "launch separate claude", "peer session", "inter-session communication", "shadow clone", "kage bunshin".

Entry point for refreshing the DH workflow graph after changes to skills, agents, Mermaid flowcharts, MCP tools, or artifact flows. Use after any structural workflow change to keep docs/dh-workflow-graph.json current.

Use when creating or updating the project skill discovery config — generates or regenerates .dh/skill_discovery.yaml by scanning the repo to infer tech stack, inventorying installed skills via npx skills list, loading candidate skill content before suggesting, and writing a config-driven skill injection file. Triggers on /dh:setup-skill-discovery invocations and programmatic --auto calls from add-new-feature Phase 3.

Use when executing a SAM task — claims the task via MCP to set it IN PROGRESS, writes active-task context for hooks, loads task-level skills, implements against acceptance criteria, and marks complete via --complete flag. Triggers on task execution within the implement-feature loop or when an agent picks up a specific task from a plan file.

Global contract for all specialist subagents — enforces role boundaries, scope discipline, and DONE/BLOCKED status signaling. Use when loading any agent that should operate as a bounded specialist following supervisor delegation patterns.

Use when starting, stopping, or checking the dot-dash live session dashboard — a real-time browser UI that monitors all active Claude Code sessions, streams transcripts, and supports prompt injection

Query and invoke tools on MCP servers using fastmcp list and fastmcp call. Use when you need to discover what tools a server offers, call tools, or integrate MCP servers into workflows.

Read The Fucking Prompt — finds the strongest user reaction to an AI instruction-following failure in a chosen session, reconstructs what the assistant did wrong, and renders a shareable terminal-style PNG. Use when asked to find rage moments, generate a rage receipt, or capture a frustration incident from a session.

Orchestrator delegation workflows for linting. Guides orchestrators on when and how to delegate to linting-root-cause-resolver and post-linting-architecture-reviewer agents. Use when orchestrating linting tasks, delegating quality checks, or reading linting resolution reports.

Orchestrator discipline plugin documentation index. Load when needing to read about orchestration patterns, delegation rules, or context window management.

Orchestrator context window discipline enforcement. Prevents the orchestrator from reading source files it will not edit, running diagnostic commands that waste context, and rationalizing delegation bypasses. Use when setting up orchestrator guardrails, reviewing delegation discipline, or diagnosing context window waste in multi-agent workflows. Activates PreToolUse hooks that surface decision points before source file reads and diagnostic command execution.

Runs the description-drift experiment — spawns all Claude Code agents simultaneously to collect self-reported capabilities, then compares them against static frontmatter descriptions to reveal how reliable orchestrator routing based on descriptions actually is. Use when measuring description drift across the agent fleet, re-running the capability collection experiment, analyzing a specific agent's self-reported capabilities, or auditing whether frontmatter descriptions accurately reflect agent behavior.

Create high-quality Claude Code agents from scratch or by adapting existing agents as templates. Use when the user wants to create a new agent, modify agent configurations, build specialized subagents, or design agent architectures. Guides through requirements gathering, template selection, and agent file generation following Anthropic best practices (v2.1.63+).

Agent Skills Open Standard reference (agentskills.io). Use when creating portable skills for Claude Code, Cursor, Gemini CLI, OpenAI Codex, VS Code, Roo Code, and 20+ compatible agents. Covers frontmatter schema, naming rules, directory structure, progressive disclosure, validation, and authoring. Load before creating cross-agent skills.

Audit agent lifecycle — validates agent execution capability against configuration. Accepts plugin path, runs 8 semantic audits (capability vs config alignment, skill loading correctness, inter-agent contracts, prompt contradictions, tool sufficiency, dead agents, scriptable patterns, pattern learning), writes reports to .plugin-creator/audits/. Use when auditing agent lifecycle, checking agent capabilities, verifying tool access, finding dead agents, validating agent contract alignment, or confirming agents can execute workflows.

Evaluate a single skill's quality against its stated purpose. Classifies the skill's purpose type, then scores applicable quality categories — universal dimensions always apply; structural dimensions (scripts, references, assets) are scored only when warranted by the skill's purpose. A focused 20-line behavioral skill with no bundled resources scores fully on all applicable categories. Use when auditing skill quality, checking marketplace readiness, evaluating skill completeness, performing pre-publication evaluation.

Audit skill lifecycle by tracing call chains, detecting circular dependencies, finding instruction contradictions, identifying duplicated datasets, analyzing bidirectional coherence, discovering scriptable sequences, and learning patterns. Use when checking skill coherence, validating skill workflow, finding semantic gaps in plugin structure, or auditing plugin before marketplace submission. Generates audit reports to .plugin-creator/audits/ with findings by dimension.

Complete reference for Claude Code plugins system (April 2026). Use when creating plugins, understanding plugin.json schema, marketplace configuration, bundling skills/commands/agents/hooks/MCP/LSP/monitor servers, plugin caching, validation, or distribution. Covers plugin components, directory structure, installation scopes, environment variables, CLI commands, debugging, and enterprise features.

Claude Code session log schema — JSONL record types, message structure, tool call/result pairing, subagent file locations, team session layout, task/plan/team configuration paths. Use when parsing ~/.claude/projects/**/*.jsonl files, writing PostToolUse hooks that measure response sizes, building session analyzers, or any agent that reads or queries Claude session transcripts.

Reference guide for Claude Code skills system (March 2026). Use when creating, modifying, or understanding skills, SKILL.md format, frontmatter fields, hooks, context fork, or skill best practices.

Reference spec and schema for Claude Code subagents — covers built-in agents (Explore, Plan, general-purpose), all frontmatter fields (name, description, tools, disallowedTools, model, permissionMode, maxTurns, skills, mcpServers, hooks, memory, background, effort, isolation, color, initialPrompt), scope and file locations, invocation patterns (@-mention, --agent, --agents CLI), tool restrictions, and example subagent definitions. Use when creating, configuring, or debugging a subagent definition file, looking up a frontmatter field, choosing between subagents and agent teams, or understanding what loads at subagent startup.

Create and configure slash commands for Claude Code — the legacy .claude/commands/ format. Use when asked to "create a slash command", "add a command", "write a custom command", "define command arguments", "use command frontmatter", "organize commands", "create command with file references", "interactive command", "use AskUserQuestion in command", or for guidance on slash command structure, YAML frontmatter fields, dynamic arguments, bash execution in commands, user interaction patterns, or command development best practices. Commands are the legacy format superseded by skills — for new development prefer /plugin-creator:skill-creator

Decide which plugin component type to use and how to organize components at scale. Covers the component lifecycle (discovery and activation phases), decision framework for choosing between commands, skills, agents, hooks, and MCP servers, and organization patterns for each component type. Use when asking "which component type should I use", "command vs skill vs agent", "when to use a hook vs MCP server", "component lifecycle", "how to organize plugin components", "plugin structure patterns", or "scale a plugin with many components".

Guide for creating Claude Code plugin hooks — language selection heuristic (Node.js default, match project runtime when obvious), hooks.json configuration, event selection, prompt-based vs command hooks, ${CLAUDE_PLUGIN_ROOT} paths, stdio suppression, timeout sizing, and testing. Use when adding hooks to a plugin, creating PreToolUse/PostToolUse/Stop/SubagentStop/SessionStart/UserPromptSubmit hooks, or wiring hook scripts to hooks.json.

Hook recipes and working examples — plugin hooks, frontmatter hooks in skills/agents/commands, prompt-based LLM hooks, and complete code examples in Python and Node.js. Use when building hook scripts, integrating hooks into plugins, implementing prompt-based hooks, or looking for hook configuration patterns.

Use when checking skill quality, validating frontmatter before commit, or diagnosing validator warnings. Runs the plugin validator on a skill, agent, or plugin directory — reports token complexity, broken links, frontmatter issues, and structural problems. Pass the path as an argument.

Configure and manage Claude Code persistent memory (CLAUDE.md, auto memory, rules) across sessions. Use when setting up project memory, organizing .claude/rules/, managing auto memory files, creating CLAUDE.md with imports, debugging memory loading, or advising on memory hierarchy and best practices.

Optimize existing CLAUDE.md, SKILL.md, agent definitions, and other AI-facing files for Claude comprehension and economy. Scope: optimization of existing content only — not upstream sync, not read-only auditing. Measures baseline metrics, delegates to @ai-doc-optimizer agent with file-type-specific context, runs independent verification via second agent, measures post-optimization metrics, and presents comprehensive before/after report. Supports iterative mode for large targets. Use when improving prompt effectiveness, reducing token waste, or rewriting instructions for LLM consumption. Invoke with /optimize-claude-md <file-or-directory>.

Use when writing delegation instructions to subagents, authoring CLAUDE.md files, rules, skills, or agent definitions, or auditing existing AI-facing content for bloat. Activates on "write a rule for", "add to CLAUDE.md", "create an agent", "update memory", or any request to author AI-facing instruction content. Removes discoverable data, explained-away knowledge, invented constraints, and stale cached facts.

Plugin creator documentation index. Load when needing to read about plugin validation error codes, skill creation, agent creation, or plugin structure.

Use when creating a new Claude Code plugin from scratch — orchestrates prerequisite check, user discussion, parallel research, design with verification, atomic implementation, multi-layer validation, documentation, and final verification. For existing plugin improvement, use /plugin-creator:plugin-lifecycle instead.

Optimize CLAUDE.md files and Skills for Claude Code CLI. Use when reviewing, creating, or improving system prompts, CLAUDE.md configurations, or Skill files. Transforms negative instructions into positive patterns following Anthropic's official best practices.

Configure Claude Code permissions — tool approval rules, permission modes, managed policies, and sandboxing. Use when setting up permission rules, configuring allow/deny/ask policies, debugging permission prompts, deploying managed settings for organizations, or controlling Bash/Read/Edit/WebFetch/MCP/Agent tool access.

Use when creating a new skill or updating an existing skill that extends Claude's capabilities with specialized knowledge, workflows, or tool integrations. Activates on "create a new skill", "add a skill to plugin", or "update existing skill".

Sync a skill's content against the upstream library it covers — check whether the API descriptions, version references, and SOURCE: citations inside SKILL.md and its reference files still reflect what the library actually does today. Use when a library has released updates that may have changed the APIs the skill covers, when a skill references an old library version, when asked to sync, refresh, update, or check a skill against upstream docs, when the skill's content may have drifted from current library behavior, or when SOURCE: citations are stale. Accepts a SKILL.md path, skill directory, or plugin directory.

Analysis criteria, transformation patterns, output format, and validation checklist for refactoring Claude Code agent prompt files. Load this skill when preparing to run the subagent-refactorer agent or when reviewing agent prompt files for structural, model optimization, or instruction quality improvements.

Write or rewrite frontmatter description fields for Claude Code skills and agents. Use when creating new skills/agents, description exceeds 1024 characters, description uses forbidden YAML multiline indicators (>-, |-), description lacks trigger keywords, or when optimizing descriptions for AI tool selection. Ensures descriptions are single-line, complete, informative, third-person, front-loaded with trigger conditions.

Routes Python engineering tasks to specialist skills by matching trigger patterns before any architecture, plan, or code is written. Use when working with Typer CLI frameworks, Rich or Textual terminal UIs, CLI UI/UX design, questionary prompts, FastMCP/MCP servers, ty type checker, uv package manager, Hatchling build backend, TOML editing, pre-commit/prek hooks, async Python, PyPI packaging, complex linting, technical debt modernization, testing workflows, feature development, or stdlib-only scripting.

Scan Claude Code session transcripts to find the strongest user reactions to assistant instruction-following failures, reconstruct the triggering assistant output, and render a shareable terminal-style PNG artifact. Use when you want to surface and share a moment where the assistant completely missed what was asked — captures what they were doing, what Claude said, and how the user reacted. Triggers on: "rtfp", "read the fucking prompt", "find my worst AI moment", "make a rage screenshot from this session".

Design and run controlled experiments using the experiment-registry MCP server — domain-agnostic, pluggable, mechanically enforced. Use when you need evidence that a change actually improves behaviour.

Rules for orchestrators handling sub-agent results without lossy re-summarization. Activates when receiving agent output, relaying research results, passing data between agents, or combining agent findings. Prevents the failure mode where 'not found' becomes 'doesn't exist' and counts get dropped.

Use when auditing docs vs code drift, syncing docs after changes, optimizing CLAUDE.md or SKILL.md for AI consumption, validating GLFM and Markdown, or summarizing files/URLs/images — routes each task to the correct specialist agent via /rwr:audit, /rwr:optimize, or /rwr:author

> Rigorously evaluate an Agent Skill end-to-end across ANY coding-agent CLI — verify its scripts emit the documented numbers (deterministic checks), test whether its description triggers on the right prompts, and measure whether an agent following the SKILL.md beats a no-skill baseline (with/without pass-rate delta, mean ± stddev, benchmarked). Use whenever you need to test, benchmark, validate, grade, or quantify a skill's quality, check if a skill "actually works," compare two skill versions, optimize a skill's triggering, or set up an eval suite — even if the user just says "is this skill any good," "does my skill work," or "benchmark this skill." Drives Claude Code, OpenAI Codex, Antigravity (agy), Cursor, GitHub Copilot, Amp, opencode, or Grok in headless mode.

>- Use this skill to run the genesis maintainer-side eval suite against genesis PR, when changing the genesis catalogue (architectural-patterns, primitives, design-patterns, refactor-patterns, composition-substrate, pattern-tradeoffs, SKILL.md), or when the operator asks to "run evals" or "regenerate the eval matrix". This skill orchestrates parallel cold sub-agent spawns via the harness's task tool, scores deterministically, and converges P>=0.8 / N>=0.8 / R==1.0 within max 3 iteration loops. This skill is contributor-only -- it lives under dev/skills/ (OUTSIDE .apm/) and is NOT shipped inside the user-facing skills/genesis/ bundle (BUNDLE LEAKAGE discipline). See "Why this lives outside .apm/" below.

>- Use this skill BEFORE drafting any agentic primitive module (skill, persona scoping file, scope-attached rule file, orchestrator workflow) or when refactoring an existing one. Activate whenever the task asks to design, restructure, or critique an agentic module across any agent harness (Claude Code, Copilot, Cursor, OpenCode, Codex), or when the task asks to make a workflow cost-effective, route model calls by capability, design for cache discipline, prune token spend, or pick a model class. This skill drives an 8-step disciplined design process whose output is mermaid diagrams + an interface sketch + a persisted plan (including a cost projection) that the calling thread (or a coder persona it loads) then turns into natural-language modules. Do not skip to natural-language drafting before the design artifacts exist.

| Build autonomous AI agents with Claude Agent SDK. TypeScript v0.2.141 | Python v0.2.82. structured outputs, and sessions. permissions/hooks, using structured outputs, troubleshooting SDK errors, or working with subagents.