Agent
Delegate a task to Antigravity CLI (agy) running headless in the current working directory. Best for broad-context review, alternate critique, second opinions, large-document/source comparison, and the Google-model perspective. Use for independent verification and consensus, or as the Google brain in a fan-out.…
Agent
Delegate a task to GPT (OpenAI Codex CLI) running headless in the current working directory. Best for implementation, writing/updating tests, structured code generation, debugging, code review, and evidence-backed repo changes — use as a second engine alongside Claude, for parallel fan-out, or for an independent…
Agent
Delegate a task to OpenCode running headless in the current working directory. Model-flexible (defaults to a free DeepSeek model; pass a specific provider/model when needed). Best for cheap parallel work, bulk/auxiliary tasks, or when you want a specific non-Claude/non-GPT model in the loop. Returns OpenCode's full…
Skill Claude CodeCodex
Write a structured handoff document so another agent (Claude, Codex, Antigravity) can continue the current work session without loss of context. Use when switching agents mid-task, closing a long session, or handing off to a fleet peer node.
Skill Claude CodeCodex
Build a contributor profile for a GitHub user — repos, contribution patterns, languages, notable PRs, activity timeline. Use when evaluating a contributor, understanding a codebase owner, or researching an open-source author.
Skill Claude CodeCodex
Deep review of a GitHub PR or commit — traces call paths, checks for regressions, security issues, data loss risks, and missing tests. More thorough than a surface diff review. Use when the operator wants a comprehensive pre-merge analysis.
Skill Claude CodeCodex
Triage open issues and PRs in a GitHub repository — label, prioritize, close stale, identify blocked items, and surface actionable work. Use when a repo's backlog needs organizing or a sprint planning pass is needed.
Skill Claude CodeCodex
Turn an open-ended or /goal objective into a bounded contract — interpretation, 3–5 verifiable completion criteria, and scope exclusions — announced in the first reply, then self-terminate when the criteria are met. Use whenever the operator gives a /goal, an outcome-only instruction ("〜しておいて", "ブラッシュアップして", "最高にして"…
Skill Claude CodeCodex
Run routine macOS maintenance on the local machine or a Tailscale fleet node — disk cleanup, log rotation, Homebrew update, Codex GC, launch daemon audit. Use when the operator asks to "clean up the Mac", "run maintenance", or reports disk/performance issues.
Skill Claude CodeCodex
Convert documents between formats — Markdown ↔ HTML, PDF, DOCX, plain text. Also cleans web-clipped markdown (removes navigation, ads, boilerplate) using defuddle or pandoc. Use when the operator needs a format conversion or a clean readable version of a document.
Skill Claude CodeCodex
Bootstrap a new repo into Katala OS agent-context style — AGENTS.md inheriting the global baseline, cross-OS .gitattributes, shared pre-commit hook via core.hooksPath, optional hash-locked CONSTITUTION + manifest.lock.json + verify.sh, NEEDSREVIEW.md. Use when the operator asks to "start a new repo", "scaffold an…
Skill Claude CodeCodex
Audit and clean up the /.claude/skills/ directory — detect broken symlinks, stale .bak directories, duplicate skills, and symlinks pointing to deleted repos. Use when skills are accumulating clutter or broken links are causing load errors.
Skill Claude CodeCodex
Download and clean transcripts from YouTube videos and other platforms using yt-dlp. Outputs clean Markdown suitable for LLM consumption. Use when the operator needs the text content of a video for analysis, summarization, or reference.