본문으로 건너뛰기
cccskills
무료GitHub에 공개

planning-commits

Triages an already-dirty Git working tree: breaks down every uncommitted change hunk by hunk, separates deliberate edits from tool-written churn, possibly accidental reversions, misplaced scope, and sensitive content, asks the user about the ambiguous cases, and proposes an ordered set of commits plus a list of hunks to hold back. Use when the user invokes `/pc` or `$pc`, asks what they should commit, wants their uncommitted changes broken down, or wants commits proposed for a dirty working tree or pile of accumulated local modifications. To plan commit boundaries for a change before or while implementing it, use `incremental-commits` instead.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

설치 방법 보기

포함된 파일(1)

  • SKILL.md5.2 KB

SKILL.md(원문)

설치하기 전에 에이전트에게 전달되는 지시 내용을 확인할 수 있습니다.

Planning Commits for a Dirty Working Tree

Turn an accumulation of uncommitted changes, often from several sessions, people, or tools, into a plan of commits the user can approve.

Boundary

Do not stage, commit, revert, discard, or delete anything until the user approves the plan. The working tree may be the only copy of some changes, and several of the classifications below are judgements the user must confirm.

1. Gather everything

  • git status --short --untracked-files=all
  • git diff --no-ext-diff and git diff --no-ext-diff --cached; keep already-staged changes distinct, since the index may be a boundary the user chose deliberately
  • the contents of untracked files; say which ones you skipped as build output or binaries

Do not truncate output carelessly: a hunk that was never read cannot be classified. For large diffs, save the output to a file and read it in parts.

2. Get context

  • Run git log on the touched paths to learn the commit-message conventions and to see whether a change continues earlier work. For example, a new hook entry of the same kind that a recent commit added for other events belongs with that theme, and may even be a fixup of an unpublished commit.
  • Read the repository's instructions (AGENTS.md, CLAUDE.md, CONTRIBUTING, and similar) for commit conventions and content rules.
  • Establish whether the repository is public, from its instructions or its remote (e.g. gh repo view --json visibility). If you cannot tell, treat it as public.

3. Classify every hunk

Classify hunks, not files: one config file often mixes a deliberate setting with tool noise.

  • Deliberate: a change the user made on purpose. Group these by logical concern.
  • Tool-written state or churn: an application rewriting its own config, such as auto-appended entries, trust or approval hashes, key reordering, version-specific path bumps, UI counters, "seen" or "done" flags, and timestamps. When a tool persists machine state into a version-controlled file, say so explicitly: it will recur until the root cause is fixed.
  • Possibly accidental: a setting silently removed or reverted to an older value, a duplicate key cleaned up, unrelated lines reformatted. Ask rather than assume either way.
  • Misplaced scope: content in the wrong file, such as project-specific settings landing in a global or user-level config, or machine-local paths in shared config. Propose where it belongs.
  • Sensitive or private: credentials, tokens, personal data, private paths or project names, conversation or session titles, and anything the repository's instructions forbid. Never propose committing these. Apply the strictest reading in a public repository, because history is permanent; a credential that was already pushed must be rotated, not merely removed.

When a hunk fits several categories, such as tool-written state that also leaks a private project name, report the most restrictive one.

4. Ask about the genuine decisions

Batch the decisions only the user can make into the interactive questionnaire tool (AskUserQuestion in Claude Code, or the equivalent in other harnesses), up to four questions per call, with the recommended option first. Do not bury questions in prose, and do not ask about anything the evidence already settles. Typical decisions: keep or revert a possibly accidental change, where misplaced content should live, whether churn should be reverted or left dirty, and whether two related concerns share a commit.

5. Present the plan

  1. Breakdown by file: each file's hunks with their classification.
  2. Proposed commits, in order: a message following the repository's conventions, and exactly which files and hunks each commit contains. Use incremental-commits for judgement on where to split and how to order.
  3. Hold back or revert: each excluded hunk with its reason, and whether to revert it, leave it uncommitted, or move it elsewhere.
  4. Follow-ups: for recurring churn, offer to investigate how to stop the tool writing it (a setting, an untracked local override file, an ignore rule) or to file an issue in the repository's tracker.

6. Execute after approval

Carry out the approved plan one logical commit at a time: stage exactly that commit's hunks with git-staging, commit with git-commit, then check that the remaining diff still matches the plan. Revert only the hunks the user approved reverting. If a file changes during execution, for example because a tool rewrote it again, stop and re-triage that file instead of committing the new content.

리뷰

아직 리뷰가 없습니다. 사용해 본 소감을 남겨 주세요.

댓글

로그인 없이 쓸 수 있습니다. 사용 후기, 질문, 작동하지 않은 환경 등을 편하게 남겨 주세요.

아직 댓글이 없습니다. 첫 댓글을 남겨 보세요.

보내면 바로 공개됩니다. 이메일 주소나 전화번호 같은 개인 정보는 쓰지 마세요. URL은 2개까지입니다.

0 / 1,000자

이 양식은 reCAPTCHA로 보호되며 Google 개인정보처리방침 및 서비스 약관이 적용됩니다.

같은 저장소의 스킬

Applies local reliability, viewport, tab-safety, stale-ref, widget, and browser-boundary lessons to every agent-browser automation. Use alongside the upstream agent-browser skill whenever navigating, clicking, filling, testing, extracting, taking screenshots, or debugging any website with agent-browser.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

aspiers/ai-config152026년 10월 9일 업데이트

Size the agent-browser Chromium window to fill its current screen, leaving a configurable bottom margin for the desktop panel. Use before any agent-browser automation that needs the full screen visible without elements overlapping or scrolling off (e.g. Xero, dense web apps).

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

aspiers/ai-config152026년 10월 9일 업데이트

Create Claude Code slash commands, OpenCode command files, Pi prompt templates, and Codex custom prompts that delegate to the right subagent or skill. Use when creating new commands, porting commands across agent platforms, or refactoring existing commands to follow platform conventions.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

aspiers/ai-config152026년 10월 9일 업데이트

Adds or changes allowed commands in AI agent permission configs for OpenCode and Claude Code. Use when a command needs adding to an agent's allow list, when permission settings for existing commands need changing, when setting up a new AI tool that requires command permissions, or when the user asks to stop being prompted for a particular command.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

aspiers/ai-config152026년 10월 9일 업데이트

bc

무료

Creates a Beads issue for a described problem or piece of work without doing the work itself. Use when the user invokes `/bc` or `$bc`, or asks to file, log, or record a bead for something. To also carry out the work, use the `bx` skill instead.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

aspiers/ai-config152026년 10월 9일 업데이트

bdr

무료

Stores a persistent memory or learning with `bd remember` so it survives across sessions. Use when the user invokes `/bdr` or `$bdr`, or asks to remember, note, or persist an insight in a Beads workspace.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

aspiers/ai-config152026년 10월 9일 업데이트

aspiers의 스킬 모두 보기

이 스킬의 문제 신고