Token导航 LogoToken导航TokenDH.com
研究检索只读github未标认证来源可访问许可证需确认审计通过

patchpatch 搜索

Agent Skill

patch 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

总安装

618

周安装

26

GitHub Stars

53

下载量

216
CodexClaudeCursorGemini CLI

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

unknown

最后核验

2026-05-01

来源状态

来源可访问

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:patch(patch 搜索)
来源仓库:https://github.com/tkersey/dotfiles
仓库路径:skills/patch
安装命令:
npx skills add https://github.com/tkersey/dotfiles --skill patch
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/tkersey/dotfiles --skill patch

简介

用于查找、检索和筛选相关信息。patch 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

  • 适合根据关键词或任务场景快速定位候选结果。
  • 可结合来源仓库和原始 README 核验具体用法。
  • 安装前建议确认权限范围和维护状态。适用宿主包括 Codex、Claude、Cursor、Gemini CLI,接入前应确认版本、权限和运行环境要求。
  • 注意是否会触发联网或文件读写操作。

SKILL.md

Patch

Intent

Carve changes into surgical git patch files: one coherent change, minimal blast radius, and at least one feedback signal before exporting.

This skill treats a patch as a transport artifact (a context-sensitive diff), not a magical commutative "merge atom". The goal is to make patches small, reviewable, and maximally likely to apply cleanly.

Mental model (why this is harder than it looks)

  • A git patch file is a *procedure over a base*: it is not commutative, not associative, and not idempotent in general.
  • Patch-based VCSes (Darcs/Pijul) work hard to make patch application order-independent; git-style patches don't promise that.
  • Therefore: always be explicit about the intended base (usually HEAD), keep patches tiny, and validate against a clean base.

When to use

  • "Split this into micro patches."
  • "Stage only the minimal change and make a patch file."
  • "Keep the patch tiny, keep checks passing."

Output formats (choose intentionally)

  • Default: git apply-style patch from the index.

- Use when you want "here are the staged changes" and the consumer will apply with git apply.

  • Optional: git am/email-style patch (requires commits).

- Use when the consumer needs commit message/author metadata and will apply with git am. - This skill prefers the non-commit path unless the user explicitly asks for format-patch.

Workflow (Surgeon's principle)

1) Scope the incision

  • Identify the smallest change that can stand alone.
  • Isolate unrelated edits; avoid drive-by refactors/formatting unless required for correctness.

2) Stage surgically (non-interactive-first)

Inspect:

  • git status -sb
  • git diff

Stage what you intend (prefer file-level staging in non-interactive harnesses):

  • git add <paths...>
  • git restore --staged <paths...>

Verify:

  • git diff --cached matches the intended incision.

If you truly need hunk-level staging but your environment can't do interactive staging, ask the user to stage hunks locally or provide a patch you can apply.

3) Validate the micro scope

  • Optional helper: scripts/micro_scope.py (staged vs unstaged size).
  • If the staged diff is multi-concern, split it before running checks.

Heuristics that reduce downstream conflicts:

  • Prefer "one file or one conceptual area" per patch.
  • Avoid adjacent edits that increase diff ambiguity (e.g. unrelated edits near each other).
  • Avoid formatting-only noise unless required; it increases patch context fragility.

4) Close the loop (required)

  • Select the tightest available signal and run it.
  • If the repo's test/check command is not discoverable, ask for the preferred command.
  • Reference: references/loop-detection.md.

5) Export patch

  • Choose an output path, e.g. patches/0001-<slug>.patch.
  • Before exporting, sanity-check for secrets (tokens, creds, private keys). If likely, stop and ask.
  • Export the staged change as a binary-safe patch (deterministic diff output):

- git diff --cached --binary --no-ext-diff --no-textconv --no-color > <patch-path>

  • Verify the patch is non-empty:

- test -s <patch-path>

  • Verify patch correctness (two levels):

- Fast local consistency check (works even if the changes are already present in the working tree): - git apply --check --reverse --binary <patch-path> - Strong check against a clean base (recommended when sharing externally): - git worktree add --detach <tmp-dir> HEAD - in <tmp-dir>: git apply --check --binary <patch-path> - git worktree remove <tmp-dir>

  • Keep staged changes intact unless the user asks to unstage or reset.

6) Repeat

Repeat until the working tree is clean or remaining changes are intentionally deferred.

Guardrails

  • Don't widen scope without asking.
  • Prefer the smallest check that meaningfully exercises the change.
  • Don't claim completion without a passing signal and at least one patch verification step (reverse-check or clean-base check).
  • Don't export or share secrets; if in doubt, ask before writing the patch file.
  • Don't overwrite existing patch files unless the user asks.

Notes on patch identity (optional)

  • Patch identity matters for deduplication ("have we already applied this logical change?").
  • In git ecosystems you can approximate this locally with git patch-id --stable (best-effort, not cryptographic).
  • Patch identity is a major reason patch-based merge can outperform diff3-style merges; see references/patch-theory.md.

Resources

  • scripts/micro_scope.py
  • references/loop-detection.md
  • references/patch-theory.md

适合场景

01

用户想查找某类 Agent Skill 时

02

需要根据任务场景推荐可安装能力包时

03

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

Codex

36.57%
按下载量换算79

Claude

31.44%
按下载量换算68

Cursor

19.48%
按下载量换算42

Gemini CLI

8.54%
按下载量换算18

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

只读

该 Skill 主要提供规则、说明或参考内容,本身偏只读;真正读写文件、联网或执行命令仍取决于宿主 Agent 的任务。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills