Write Skill
Author or update compact agent skills under skills/<name>/SKILL.md. Use when asked to write, add, or change a skill; not to audit a skill library.
Show Me
Answer with a compact visual — call tree, component tree, file tree, pseudocode, type signature, diff, or Mermaid — instead of a wall of prose. Use when the user says "show me", "draw", "sketch", "diagram", "visualize", or asks how something is wired, what shape an API should take, or what a change would look like. Do NOT use for whole-repo briefings (use grok-repo) or for charts and dashboards (…
Paper Search
Find and rank research papers by recency and field-adjusted impact. Use when searching literature, recent or superseding work, or evidence for AI/ML/systems claims. Do NOT use for general web search or GitHub repository evaluation.
Newcomer Lens Review
Review code for rationale, domain terms, constraints, and assumptions missing to a newcomer, and produce a documentation backlog. Use when asked for an onboarding review, what would confuse a new hire, what a handover is missing, or which parts of a codebase need documenting first. Do NOT use to find defects (use code-review) or to build your own understanding of unfamiliar code (use grok-repo).
Obsidian Capture
Append todos, logs, learnings, and reflections to Obsidian daily or weekly notes, and run the agent's own #agent-todo queue, via scripts/capture.py. Use when asked to add a todo, note something down, log this, capture an idea, set a due date, record a weekly reflection, or list/close the agent's assigned tasks ("看你的 todo", "干活", "what's on your todo list"). Do NOT use for searching notes (use obs…
Write Docs
Write or critique the prose that ships with a codebase — READMEs, CLI help and flag design, code comments and docstrings, AGENTS.md/CLAUDE.md agent guides, and commit or PR descriptions. Use when drafting, improving, or reviewing any of these; not for design docs or internal wikis.
Skill Curator
Audit a skill library for duplicated guidance, contradictions, bloat, dead references, and defects. Use when cleaning, consolidating, or de-duplicating a skills tree.
Obsidian Search
Search an Obsidian vault for topics, exact text, tags, recent notes, and tasks. Use when fuzzy, date-aware, due, or overdue note retrieval is needed.
Grok Repo
Explain an unfamiliar codebase through a repository briefing or scoped dataflow trace. Use when asked for a repo tour, how something works, where a value comes from, or design rationale. Do NOT use for repository health (use repo-eval) or diff review.
Questionnaire
Turn the open decisions from a long discussion into a self-contained offline HTML questionnaire the user fills at their own pace, whose answers come back as markdown keyed by stable question ids. Use when the user wants to finalize or confirm a design, align on accumulated decisions, review assumptions from notes/memory, or asks for a survey or questionnaire. Do NOT use for live one-question-at-a…
Hammerspoon
Operate macOS via Hammerspoon — run one-off `hs -c` Lua from the shell (launch/quit/focus apps, open/close/focus browser tabs, volume/wifi/caffeinate toggles, on-screen alerts, clipboard) or author persistent automations (hotkeys, watchers, menubar) in ~/.hammerspoon. Use when asked to open/close/focus an app or tab, control Mac system state, "keep my mac awake", notify on screen, or write a Hamm…
Test Writing
Write and repair tests that verify observable behavior, not implementation. Use when writing or reviewing tests, choosing test scope or mocks, fixing brittle tests, or deciding whether a failing test means the code is wrong or the test is.
Karpathy Planning
Plan non-trivial coding work by surfacing assumptions, minimizing scope, and defining success criteria and verification. Use when asked to plan or when implementation scope is ambiguous. Do NOT use for interactive interviews (use grill-me) or value-versus-complexity checks on existing code (use complexity-budget).
Code Review
Review a diff, package, or API through one of three lenses — necessity and layering (Russ Cox), invariant and cost honesty (BurntSushi), or product-versus-library fit (Mitsuhiko). Use when reviewing a change or dependency, auditing or polishing a package, or asking whether code is too complex, whether an API can panic or will age well, or whether a simpler version should ship. Do NOT use for gati…
Web Recap
Search browser history for URLs by topic or visit statistics. Use when asked about visited sites, browsing history, or recent online activity.
Complexity Budget
Gate changes to existing code by comparing added complexity with expected value before and after editing. Use when adding features, fixes, flags, branches, abstractions, or parallel paths. Do NOT use for greenfield work or finished-code review (use code-review).
Grill Me
Interview the user one question at a time until a plan is resolved, capturing durable decisions in CONTEXT.md or ADRs when the repo warrants it. Use when asked to "grill me," stress-test a design, or challenge assumptions before implementation.
Repo Eval
Evaluate a public GitHub repository's momentum and maintenance health. Use when asked whether a repository is healthy, growing, declining, abandoned, or well maintained, or when vetting dependencies and OSS projects. Do NOT use for code or architecture review.