dotfiles

Create or modify a Claude Code subagent — author a new agent .md with valid frontmatter at the user-chosen path, or make targeted edits to an existing one, then lint it. Use when the user says "create an agent", "create a subagent", "scaffold an agent", "new agent definition", "modify this agent", "edit the agent", "iterate on this subagent", "change the agent's routing".

Add an MCP server to the appropriate config (.mcp.json or ~/.claude.json). Handles transport, env vars, headers, OAuth, and scope selection. Use when the user says "add MCP server", "connect to <service>", "register MCP", "set up Linear/Slack/...".

Add a Claude Code hook — scaffold the script and wire the event/matcher/command into settings.json correctly. Surfaces the exit-code contract. Use when the user says "add a hook", "create a hook", "fire X after Y", "block Z on edit".

Promote local .claude/ artifacts (skills, agents, hooks, MCP servers) into a proper plugin layout with .claude-plugin/plugin.json. Validates structure and runs 'claude plugin validate' if available. Use when the user says "make a plugin", "bundle into a plugin", "promote to plugin", "publish skill".

Read-only audit of the user's Claude Code configuration. Scans skills, agents, hooks, MCP servers, settings.json, and CLAUDE.md for documented gotchas, size warnings, deprecated patterns, and misconfigurations. Writes nothing. Use when the user says "audit my Claude Code config", "what's wrong with my setup", "doctor", "check my .claude/".

Guide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK).

Make a guarded edit to settings.json — validate the schema, refuse managed-only keys at user/project scope, and show a diff before applying. Use when the user says "change settings.json", "set <key>", "update settings", "add permission rule", "allow X", "deny X", "stop allowing X", "move permission to user|project", or "set env var". Preferred over the built-in /update-config skill for any change where managed-key refusal, scope-silent-ignore detection, or programmatic lint matters. Hooks authoring goes to `create-hook`, not here.

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.

Structured debugging session for a reported error or unexpected behavior. Use when given a traceback, error message, or symptom to investigate. Gathers context, forms hypotheses, and proposes a fix without making changes until confirmed.

Summarize the current PR's diff and flag risks. Use when asked to review changes, summarize a PR, write a PR description, or check what changed.

Bootstrap a new Python project with uv: pyproject.toml, ruff config, .python-version, and optional src layout. Use when starting a new Python package or script project.