本文へ移動
cccskills
無料GitHub で公開

mas-skills

Use when a task needs AUTO-MAS engineering conventions across frontend, UI, code style, schema naming, module boundaries, function design, API contracts, data modeling, script adapters (MAA, SRC, MaaEnd, General, ok-script family such as Okww and OkNte, multi-engine HSR), E2E flows, plan schedules, or game community sign-in.

日本語の概要は準備中です。原文の説明を表示しています。

AUTO-MAS-Project/AUTO-MASスター 757(GitHub 757 + cccskills 0)
連携ツールGitHub

実行スコア 未検証

インストール方法を見る

含まれるファイル(2)

  • SKILL.md15.7 KB
  • agents/openai.yaml298 B

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

MAS Skills

Objective

Provide one routing entrypoint for MAS engineering standards and keep implementation decisions consistent with current project maintainer review preferences.

Repository Authority

AUTO-MAS splits responsibilities across repositories:

  1. AUTO-MAS-Project/AUTO-MAS keeps application code and a minimal AGENTS.md entrypoint.
  2. AUTO-MAS-Project/AUTO-MAS-docs / https://doc.auto-mas.top/developer/ owns branch, contribution, commit, version, Issue, and PR writing rules.
  3. Local .agents/skills owns the project-affiliated Agent Skill behavior and engineering routing rules for this checkout.

When these areas overlap, use this skill for engineering decisions and use the docs site for contribution-process decisions.

Project Rule Areas

Apply these project rules directly when they overlap with a task:

  1. Cherry-picks to an already released release/{version} branch must be small pure-backend fixes that touch no frontend code, because released apps hot-update only backend code from that branch while the frontend ships inside the installer. Cherry-picking a commit that carries frontend logic is a violation: close the related PR and revert the related commit. Judge by the files the commit actually touches, not by the type / scope in its message. Never help prepare or push such a cherry-pick.

  2. API work follows the schema-first route flow: define app/models/schema.py, wire the matching app/api/ module, then regenerate frontend API clients.

  3. Config work follows ConfigBase and ConfigItem ownership: declare fields before super().__init__(), use validators as auto-correction behavior, and register multi-config classes in the owning collection.

  4. Script adaptation must be complete across config, schema, API, task dispatch, task folder, and frontend entry points, and must respect the black-box boundary: lower the configuration barrier before filling any gap, then decide capability ownership: work the script owns (in-game actions, task execution and verdicts, script config semantics) may only be reused through upstream entry points; when upstream lacks such work, consider contributing it upstream first and only fill in inside MAS as a temporary measure (marked for removal once upstream ships it) when that is not viable; MAS-owned domains (accounts, scheduling, plans, notifications, statistics, emulator lifecycle, cross-script orchestration) may be implemented in MAS but must not read or infer upstream internal state; upstream private formats may only be passed through, never modeled on, and no self-invented verdicts for upstream-executed tasks. Once upstream ships an entry point for such work, existing implementations become violations and must be replaced by reuse. After this skill is loaded, self-check the task and, on a hit, report "This may violate the MAS development norms" with file:line evidence and an alternative; the warning does not block work. Read .agents/skills/mas-script-specialized-adapter/references/blackbox-boundary.md before designing.

  5. Frontend work follows the dedicated frontend skills in this directory.

  6. Contribution style follows Conventional Commits, Google-style backend docstrings where useful, config-item comments, and keyword arguments for booleans or multi-argument calls.

  7. CI must pin every external reference to immutable content: GitHub Actions uses entries use a full 40-character commit SHA with a # <tag> version comment (resolve annotated tags to the peeled commit), and container images use @sha256:<digest>. Apply this across GitHub workflows, composite actions, and CNB pipelines; local actions such as ./.github/actions/... are the only exception. Never leave a mutable tag, branch, or short hash. When changing CI, scan every uses: and image: reference and verify that all external references remain pinned.

  8. Start every branch from the latest upstream dev (git fetch first; in a fork, branch off upstream dev), never from a stale local dev: a branch built on an old baseline silently overwrites other people's merged work when it is synced, with no conflict and green type checks. Branch rules: https://doc.auto-mas.top/developer/development-specifications.html.

Sub-Skills

Use these skills as needed:

  1. mas-frontend-standards: Vue 3, TypeScript, Vite, Electron renderer, routing, API composables, state, styles, forms, validation, and frontend verification.
  2. mas-frontend-ui: Ant Design Vue UI, desktop business layout, visual tokens, forms, tables, dialogs, feedback, drag interactions, dark mode, and UX constraints.
  3. mas-code-standards: project code standards derived from representative dev commits and existing AUTO-MAS modules.
  4. mas-schema-naming: canonical naming for shared schema semantics in future domain work.
  5. mas-module-boundary: module ownership and dependency direction across backend layers.
  6. mas-function-design: function-level design for responsibility, signatures, side effects, and errors.
  7. mas-api-contract: endpoint contract standards for HTTP/WS request-response behavior.
  8. mas-data-model: modeling standards for schema/config/task layers and compatibility evolution.
  9. mas-script-specialized-adapter: specialized script integration by script frontend architecture line; requires intake before implementation.
  10. mas-plan-schedule: plan schedule type registration, backend/frontend plan dispatch, plan combobox consumers, and per-type table integration.
  11. mas-e2e: author-local real script/game/account runs, isolated browser flows, PR-specific selection, sanitized evidence, the command matrix, and the post-PR hand-off that offers those commands to the developer.
  12. mas-game-sign: game community sign-in providers, credential encryption and login routes, sign-in locks and trigger paths, and result/notification contracts.
  13. mas-ci-checks: every gate that runs on a PR to dev (format, lint, typecheck, changelog fragment, uv.lock, revert hints, CNB mirror, unsigned Windows build) and the exact local command for each.

Global Constraints

Apply these constraints before selecting or combining sub-skills.

  1. Make minimal necessary changes first; avoid broad refactors unless explicitly requested.
  2. Align with current code style and existing project conventions in the touched module.
  3. Avoid over-engineering, over-abstraction, and defensive programming that does not match existing code patterns.
  4. Study similar existing implementations deeply before coding and follow established local patterns.
  5. Never edit files generated by OpenAPI code generation under any circumstance. If regeneration is needed, explicitly tell the developer to run the generation process manually.
  6. For frontend work, load mas-frontend-standards; for UI or user-facing component behavior, also load mas-frontend-ui.
  7. When local patterns and generalized guidance differ, prefer concrete maintainer review comments and fold them back into the selected sub-skills.
  8. Before completing a user-visible feature or fix, add exactly one changelog fragment under changelog.d/ (<PR number or branch>.<feat|change|fix|breaking|remove|security|dev>.md; first line project: <key> from the project table in changelog.d/README.md — 14 adapters plus home / scheduler / emulator / notify / tools / settings / update / runtime, no catch-all, pick the nearest per the README; only dev fragments may omit it; then one user-facing sentence of at most 50 characters without project prefix, PR number or signature; python scripts/changelog.py add <type> <project> "<sentence>" creates it, with - as the project for dev). Before writing a fix or change fragment, decide which release introduced the feature you touch: a feature absent from the previous stable release's notes was introduced in the current X.Y.0 cycle (for 5.5.0: Runtime initialization, virtual display, Emulator 2.0, MFW, BetterGI, BAAH, ZZZ-OD, config restore, operator cultivation). Every later fix, maintenance or supplementary change to such a feature (bug fixes, turning something into a prompt, extra hints, default tweaks, layout changes, i18n wiring, cleanup) is beta-only: if the feature's entry is still in the top ## [未发布] section or its fragment is still in changelog.d/, add no fragment and label the PR skip-changelog (describe any wording change in the PR for a maintainer to apply); if it shipped in an earlier beta of the cycle, add the header line beta-only: true so the entry is dropped from the stable roll-up. Sub-features added to it later are beta-only as well: in the stable notes a new adapter is exactly one line (【bgi】新增 bgi 专项) and a new feature is its single introduction entry; fixes and changes to features that shipped in the previous stable are normal. Project keys use the short names bgi (BetterGI) and end (MaaEnd). A PR with several human authors lists them all in the header line author: a, b (the script never reads Co-authored-by). Maintainers may add highlight: true to route an entry into 「本次亮点」 as part of the implementation; do not merely remind the developer. Condense all changes of the PR into that one sentence. Never edit CHANGELOG.md, res/version.json or any version number: they are written only by the release PR. Do not create a commit unless the user explicitly asks for one.

Routing Rules

Choose sub-skills by task intent.

  1. Task mentions field names, shared terms, schema key consistency: Use mas-schema-naming.
  2. Task touches frontend, src/views, src/components, src/composables, src/router, src/types, src/utils, Vue, TypeScript, Vite, Electron renderer, API composables, forms, validation, or frontend verification: Use mas-frontend-standards.
  3. Task touches UI, layout, style, Ant Design Vue, components, forms, tables, modals, drawers, feedback, loading, empty, error states, drag interactions, dark mode, or visual polish: Use mas-frontend-ui with mas-frontend-standards.
  4. Task implements, fixes, refactors, or reviews repository code: Use mas-code-standards as the baseline, then add the minimum domain-specific mas-* skills. Skip it for read-only work unless code conventions or commit wording are requested.
  5. Task mentions layer ownership, imports, coupling, or where code should live: Use mas-module-boundary.
  6. Task mentions function splitting, signature quality, return/error behavior: Use mas-function-design.
  7. Task mentions endpoint payloads, response model, error contract, websocket payloads: Use mas-api-contract.
  8. Task mentions model structure, typing/defaults/constraints, migration of model fields: Use mas-data-model.
  9. Task mentions adding a new script, script-specific adaptation, task lifecycle, or a specific adapter such as MAA / SRC / MaaEnd / General / Okww / OkNte(异环)/ HSR (M9A is not an adapter: it is a MaaFW flavor, see app/task/M9A/AGENTS.md): Use mas-script-specialized-adapter first, then combine mas-module-boundary, mas-data-model, mas-function-design, and mas-api-contract as needed.
  10. Task mentions E2E, Playwright, real game runs, emulator/account scheduling evidence, PR-specific test selection, screenshot/trace evidence, or a PR is open and awaiting review while the change still needs E2E proof: Use mas-e2e; combine mas-script-specialized-adapter when the flow depends on a particular script line.
  11. Task mentions commit messages, docstrings, config comments, or project contribution style: Use mas-code-standards for code-style decisions; use the docs site for contribution-process wording.
  12. Task mentions plan schedules, schedule types, PlanConfig, PLAN_BOOK, plan comboboxes, or adding a new plan table: Use mas-plan-schedule.
  13. Task mentions game community sign-in, app/tools/game_sign*, a sign-in platform such as Skland/Miyoushe/Kuro/Taygedo, sign-in credentials or QR login, or sign-in result and notification behavior: Use mas-game-sign.
  14. Task mentions cherry-picking a fix to a release/{version} branch, backporting to a released version, or hot-updating released backend code: Apply the cherry-pick rule in Project Rule Areas before touching git; confirm the commit is small and pure backend by inspecting the files it actually touches.
  15. Task mentions a red PR, a CI failure, GitHub Actions, format / lint / typecheck gates, the skip-changelog or changelog-maintenance labels, or editing .github/workflows, .github/actions, scripts/changelog.py, scripts/check_reverts.py: Use mas-ci-checks.

Combined Execution Order

When multiple concerns appear, apply this order:

  1. mas-frontend-standards, when the task touches frontend code or docs.
  2. mas-frontend-ui, when the task touches UI or user-facing component behavior.
  3. mas-code-standards
  4. mas-module-boundary
  5. mas-data-model
  6. mas-schema-naming
  7. mas-function-design
  8. mas-api-contract
  9. mas-e2e, when the task requires an E2E flow or evidence.
  10. mas-script-specialized-adapter after architecture intake, when the task is a specialized adapter.
  11. mas-plan-schedule
  12. mas-game-sign
  13. mas-ci-checks, when the work ends by making a PR pass CI.

Reason: frontend tasks need their engineering and UI constraints loaded before implementation decisions; then establish local conventions, place code correctly, stabilize model structure, then naming, function behavior, and transport contract. Specialized adapters add a mandatory architecture-intake step, plan schedule rules apply when the task touches scheduler registration, and game sign rules apply last among the implementation skills because they constrain credential handling and external request behavior inside an already-placed module. CI checks come after everything else because they only push finished work through the PR gates.

Output Requirements

When using this hub:

  1. State which sub-skills are selected.
  2. Explain why each selected sub-skill is needed.
  3. Apply only the minimum set required by the task.
  4. Keep compatibility-first decisions for legacy modules unless explicitly asked to refactor broadly.
  5. In review tasks, call out where findings follow known maintainer preferences rather than only generic engineering taste.
  6. For Issue/PR body writing, follow the docs site instead of inventing repository-specific text here; the drafting techniques live in the pr skill.
  7. For frontend tasks, state whether mas-frontend-standards and mas-frontend-ui were selected and why.

Review Checklist

  1. Selected sub-skills match the user request scope.
  2. Frontend tasks loaded the correct frontend skill pair.
  3. No layer-boundary violations are introduced.
  4. Shared schema semantics remain canonical.
  5. Function behavior and API contract stay consistent after changes.
  6. Relevant project rules were considered for API, config, script task, frontend task, and contribution-style changes.
  7. Contribution-process details were not duplicated from the docs site except as links or brief reminders.
  8. Every user-visible feature or fix includes exactly one changelog fragment under changelog.d/, and leaves CHANGELOG.md, res/version.json and the version numbers untouched.
  9. Any cherry-pick to a released branch was verified to touch backend files only.
  10. Every external CI action and container image is pinned to a full commit SHA or digest, with an updater-readable version comment where applicable; no mutable reference remains.
  11. CI failures were diagnosed from the run log first, and reproduced locally with the uv.lock-pinned toolchain rather than a system-installed one.

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

コメント

ログインなしで書き込めます。使ってみた感想、質問、動かなかった環境などを気軽にどうぞ。

まだコメントはありません。最初のコメントを書いてみませんか。

送信するとすぐに公開されます。メールアドレスや電話番号などの個人情報は書かないでください。URL は 2 つまでです。

0 / 1000 文字

このフォームは reCAPTCHA で保護されており、Google のプライバシーポリシーと利用規約が適用されます。

同じリポジトリのスキル

从固定点(commit、branch、tag 或 merge-base)开始,沿两条轴线审查变更——规范(代码是否遵循仓库文档化的编码规范?)和规格(代码是否与原始 issue/PRD 的要求一致?)。两条审查线在并行子 agent 中运行,并以并排方式报告结果。当用户想审查一个分支、PR、进行中的变更,或要求"从 X 开始审查"时使用。

日本語の概要は準備中です。原文の説明を表示しています。

AUTO-MAS-Project/AUTO-MAS7572026年10月26日 更新

grill-me

無料

一场无休止的盘问,用来打磨你的计划或设计。

日本語の概要は準備中です。原文の説明を表示しています。

AUTO-MAS-Project/AUTO-MAS7572026年10月26日 更新

grilling

無料

对用户的计划、决策或想法进行层层盘问。当用户想要压力测试自己的思考,或使用任何 'grill' 触发短语时使用。

日本語の概要は準備中です。原文の説明を表示しています。

AUTO-MAS-Project/AUTO-MAS7572026年10月26日 更新

Define backend API contract standards for FastAPI services. Use when adding or refactoring HTTP/WebSocket endpoints in app/api, designing request/response schemas in app/models/schema.py, standardizing status/error contracts, and maintaining backward compatibility for clients.

日本語の概要は準備中です。原文の説明を表示しています。

AUTO-MAS-Project/AUTO-MAS7572026年10月26日 更新

Use when an AUTO-MAS PR is red in CI, when reproducing a dev-PR gate locally before pushing, when a workflow needs a skip-changelog or changelog-maintenance label, or when editing .github/workflows, .github/actions, scripts/changelog.py or scripts/check_reverts.py. Covers every gate that runs on a PR to dev and the exact local command for each.

日本語の概要は準備中です。原文の説明を表示しています。

AUTO-MAS-Project/AUTO-MAS7572026年10月26日 更新

Use when implementing, fixing, refactoring, or reviewing non-generated AUTO-MAS code, or when preparing code-style guidance, comments, docstrings, version notes, or Conventional Commit wording. Skip read-only diagnosis, explanation, exploration, and planning unless code conventions are requested.

日本語の概要は準備中です。原文の説明を表示しています。

AUTO-MAS-Project/AUTO-MAS7572026年10月26日 更新

AUTO-MAS-Project のスキルをすべて見る

このスキルの問題を報告する