maigo

This skill should be used when assembling git commits — deciding what to stage, how to scope a bash/agent command's working directory, whether to amend an unreleased commit or stack a new one, and how to size a diff against the correct baseline. Applies to the commit-assembly step of /maigo:go, /maigo:quick, /maigo:team, and /maigo:address-comments, and to the orchestrator generally.

This skill should be used when the maigo orchestrator, running in the Claude Code harness, decides whether to delegate work to a subagent instead of doing it on the main thread, needs to keep a long task's acceptance criteria from drifting after context compaction, or needs to keep verification independent from whoever produced the work. Scope note: 只在 Claude Code harness 下才適用——這裡的 orchestrator context 會被計費、可能被壓縮、且可 spawn subagent 分攤負載；沒有 subagent 能力的 harness 不適用此 skill。

This skill should be used by all maigo agents at startup, before beginning work, to load relevant cross-project memory entries with relevance-based ordering and a 10-entry cap. Consumers: Raana, Tomori, Soyo, the orchestrator itself, and any future agent that reads ~/.config/maigo/memory/.

This skill should be used by the maigo orchestrator on every /maigo command, alongside narration, to govern the conversational conduct of the main dialogue — AskUserQuestion widget discipline and Taiwanese Mandarin word-choice norms.

This skill should be used when performing code review on a diff (whether implemented by another agent or an external PR), enforcing a strict reviewer stance, applying a mandatory 9-item checklist, demanding evidence, and giving specific改法 instead of vague critique.

This skill should be used when orchestrating the full MyGO!!!!! teammate flow — Raana explores, Tomori plans, Anon implements, Soyo reviews, Taki validates. Applies to /maigo:go and /maigo:team (both sequential and parallel variants).

This skill should be used when working in the apache/airflow repository as a contributor (not as a Dag author). It surfaces the conventions Airflow's own AGENTS.md enforces — naming, Breeze/uv environment, Ruff/Mypy style, coding rules, pytest patterns, PR hygiene — so any task (review, quick-fix, refactor) in airflow follows them.

This skill should be used when drafting reply text for GitHub PR review threads or review comments — ensuring replies are terse, reference only committed symbols, are scoped per thread, and include the required attribution footer.

This skill should be used when reading, writing, or migrating `.maigo/board.md` — the single cross-session Work Board that tracks issues to triage, your own PRs, and PRs you're reviewing, bucketed by whose turn it is to act (🎯 你的球 / ⏳ 等別人 / ✅ merged-closed). Covers the line grammar, possession-判定 tables per item type, the upsert contract each writing command follows, the review-board.md migration path, and the checkbox → `--learn` memory gate.

This skill should be used when a maigo diff touches agents/, commands/, skills/ (including SKILL.md and references/), mkdocs.yml, or docs/ — in those cases validate_plugin + mkdocs --strict are the verification standard, not just pytest.

This skill should be used when handling failures in go-class command flows — Soyo blocking with NEEDS_CHANGES / BLOCKED, Taki test failures, subagent infrastructure overload / unavailability (e.g. 529), mid-task usage/session-limit interruption of a running subagent, and infinite-loop protection for repeated same must-fix or same test ID. Applies to /maigo:go, /maigo:quick, /maigo:team, and /maigo:address-comments step 5.

This skill should be used when drafting or rewriting a GitHub PR title and/or description. It produces a user-impact title (not conventional-commits formatted) and a Why / What / Test Plan body with optional Breaking changes / Related issues sections. Use both for fresh drafts from branch commits/diff (e.g., /maigo:describe-pr flow) and for ad-hoc rewrites of an existing PR body without re-fetching git context.

This skill should be used when the maigo orchestrator, running in the Claude Code harness (Agent tool available for spawning subagents), decides which model tier to dispatch a subagent task to, or judges an escalation / de-escalation after a subagent's result. Scope note: 只在 Claude Code harness 下才適用——這裡的 Agent tool 支援逐次 `model` override；其他 harness（例如 Codex 的 command-router 環境）沒有這個能力，此 skill 不適用。

Codex pseudo-command router for Maigo. Use when the user sends or discusses any `/maigo:*` or `maigo:*` command, including go, team, quick, review, board, remember, memory, retro, crystallize, describe-pr, address-comments, triage-issue, take-issue, doctor, or repo-audit; also use when the user asks why Maigo commands do not behave like native Codex slash commands.

This skill should be used when drafting a git commit message (new commit, `git commit --amend`, or a squash/rewrite message) from staged or proposed changes. It produces a user-impact subject and a concise body listing changed behaviour and breaking-change / newsfragment / related-PR pointers, deliberately avoiding motivation paragraphs (which belong in the PR description, not the commit log).

This skill should be used during /maigo:review when fetching or reusing PR context (title / body / diff / CI status / linked issues), caching the first fetch into review-rubric.md so subsequent re-review rounds skip re-fetching.

This skill should be used by the maigo orchestrator whenever a Soyo or Anon output contains a

This skill should be used whenever a maigo command produces a deliverable destined to be pasted elsewhere (a PR/issue comment, a reply draft, a commit message, a gh command draft). It mandates wrapping that content in a single fenced code block so the user gets raw, one-click-copyable markdown instead of a rendered-only version.

This skill should be used by the maigo orchestrator on every /maigo command, to frame the run with Doloris / Mortis narration at the opening, the closing, and stuck-point beats, and to enforce the emoji prefix on every mention of a maigo agent or narrator.

This skill should be used when triaging an inbound GitHub issue from a maintainer perspective — applying a 9-item checklist to decide a verdict (READY / NEEDS_INFO / DUP / CLOSE), surfacing classification disagreements with existing labels, and producing a draft response plus gh command suggestions. Parallel to strict-review but for issues, not code.

This skill should be used when writing or reviewing cross-file links inside Maigo source files (`agents/*.md`, `commands/*.md`, `skills/*/SKILL.md`). It enforces that cross-source links use absolute GitHub URLs — relative links break `mkdocs build --strict` because these files are dual-context (raw GitHub view + include-markdown shim into `docs/`).

This skill should be used when working in the `commitizen-tools/commitizen` repository as a contributor (not as a downstream `[tool.commitizen]` user). It surfaces the conventions from the repo's own AGENTS.md — uv + poe tasks, conventional commits enforced by commitizen itself, ruff/mypy linting, pytest, and PR guidelines — so any task (review, quick-fix, refactor) in commitizen follows them.