Token导航 LogoToken导航

Agent Skill · 研究检索

working-tree-governor(工作树调控器)

working-tree-governor 用于查找、检索和筛选相关信息,适合在 OpenClaw 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。

working-tree-governor 是研究检索类 Agent Skill,可用宿主为 OpenClaw,通过ClawHub获取,权限为需要联网;许可证为 MIT-0,第三方扫描已通过,按安装量在研究检索类的 73,675 个 Skill 里排第 34,152。

总安装 2,258周安装 96GitHub Stars 公开资料未说明下载 791核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
MIT-0
版本
1.0.0

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:working-tree-governor(工作树调控器)
来源仓库:https://github.com/lobsterquant/working-tree-governor
安装命令:
openclaw skills install working-tree-governor
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 OpenClaw 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。

ClawHubOpenClaw
openclaw skills install working-tree-governor

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

工作树调控器用于管理肮脏的 git 工作树,支持噪音分类和暂存控制。适合需要整理代码变更或清理开发环境的场景。提供运行时噪音识别和暂存范围验证功能。working-tree-governor 属于研究检索类 Skill,可作为该场景下的辅助能力补充。安装前需确认是否会触发文件操作或网络访问。

建议参考原始 README 了解具体命令和限制条件。

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 OpenClaw;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 openclaw skills install working-tree-governor;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它需要需要联网,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

working-tree-governor 怎么安装?

在终端执行 openclaw skills install working-tree-governor。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

working-tree-governor 支持哪些 AI 工具?

公开资料标注支持 OpenClaw。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

working-tree-governor 需要什么权限?

权限级别是需要联网。该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

working-tree-governor 是免费的吗?

来源标注的许可证是 MIT-0。本站不代收任何费用,也不托管安装包;能否商用、要不要署名,以许可证原文为准。

working-tree-governor 用的人多吗?

公开数据是累计安装 2,258 次,近一周安装 96 次,下载 791 次。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

working-tree-governor 安全吗?

第三方扫描结果为通过,但这只覆盖已知的静态检查项,不能替代自己读一遍源码和安装命令。

值得选的地方和要先想清楚的地方

值得选的地方

  • 专注 OpenClaw 一种宿主,触发方式和使用习惯不用来回切换。
  • 许可证明确为 MIT-0,商用和分发边界好判断。
  • 第三方扫描已通过,常见的高风险行为没有命中。

先想清楚的地方

  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 OpenClaw,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。
  • 需要许可证清楚、并且有第三方扫描结果的团队项目。

不太适合

  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
working-tree-governor

工作树调控器 · 本页

2,258—需要联网MIT-0通过

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

name
working-tree-governor
description
Govern dirty git working trees by classifying runtime noise vs real source changes, defaulting to selective staging, verifying staged scope, and asking the operator when scope is ambiguous.
version
1.0.0
author
Hermes Agent
license
MIT

Working-tree-governor

Use when a repo is dirty and you need to commit safely without dragging runtime noise, logs, cache, state files, or generated artifacts into history.

This skill exists for the common Hermes failure mode where git status becomes unreadable because code, tests, logs, runtime state, forensics, cached binaries, and generated outputs are all mixed together.

The goal is not cleanliness theater. The goal is to preserve signal and produce small, reviewable commits.

When to trigger

Load this skill when any of these are true:

  • the user asks to commit while the repo is dirty
  • git status --short shows mixed code plus logs/state/cache/output noise
  • more than ~25 changed or untracked files
  • more than ~8 source or test files are modified
  • the next action is risky: restart, deploy, cutover, refactor, or live-sensitive change
  • runtime-critical files are dirty in a tree already full of noise

Treat these as high-risk triggers:

  • more than 100 changed/untracked files
  • live runtime files dirty and HEAD stale
  • planned restart/deploy while mixed dirt is still present
  • tracked state/log/output files continuously mutating and obscuring code changes

Core principle

Never use git add . in this situation. Never default to stash for this situation. Default to selective staging. If confidence is low, ask the operator before commit.

Discovery workflow

Start read-only.

Run:

git status --short
git diff --name-only
git diff --cached --name-only
git rev-parse HEAD
git log -1 --format='%H %ad %s' --date=iso

Optional when useful:

git diff --stat
git diff -- <path>

Then bucket every changed path.

Buckets

Bucket A — core code and tests

Examples:

  • src/**/*.py
  • tests/**/*.py
  • service scripts
  • core config templates
  • startup scripts
  • bridge/runtime code

Default action: include_now

Bucket B — docs and plans

Examples:

  • docs/**
  • selected *.md
  • intentionally authored design notes

Default action: usually needs_operator_decision unless clearly part of the requested commit

Bucket C — runtime state

Examples:

  • shared/risk/*.json
  • shared/reports/strategy_pnl.json
  • shared/risk/open_positions.json
  • hermes_state.json

Default action: preserve_but_exclude

Bucket D — generated outputs and artifacts

Examples:

  • results/**
  • backtest outputs
  • parquet / duckdb
  • generated summaries

Default action: preserve_but_exclude

Bucket E — logs

Examples:

  • *.log
  • *.jsonl
  • streamer.log*
  • monitor logs
  • cron logs

Default action: ignore_for_now or preserve_but_exclude

Bucket F — caches and binaries

Examples:

  • __pycache__/
  • *.pyc
  • .metaapi/*.bin

Default action: ignore_for_now

Bucket G — forensics and snapshots

Examples:

  • shared/forensics/**
  • *.bak-*
  • incident snapshots

Default action: preserve_but_exclude

Bucket H — unknown

Anything not confidently classifiable.

Default action: needs_operator_decision

Heuristics

High-confidence exclude

Exclude by default when path matches patterns like:

  • **/__pycache__/**
  • **/*.pyc
  • shared/hermes-acp/*.log
  • shared/hermes-acp/*_log.jsonl
  • shared/reports/*.jsonl
  • shared/risk/*.json
  • results/*.json
  • memory/**/*.md
  • src/.metaapi/*.bin
  • shared/forensics/**
  • *.bak
  • rotated logs / archives

High-confidence include

Include by default when path matches patterns like:

  • src/**/*.py
  • tests/**/*.py
  • skill files
  • fixtures under tests/fixtures/**

Important exception

Do not exclude real code just because it lives beside noisy runtime files. For example, shared/hermes-acp/*.py may be real source code and must be inspected, not auto-excluded.

Commit-governor flow

1. Summarize the tree

Report:

  • total dirty file count
  • bucket counts
  • runtime-critical files dirty
  • likely risk level: low / moderate / high / critical

2. Infer requested commit scope

Use the user’s requested commit intent to infer the include set. Examples:

  • execution-truth work → shared/hermes-acp/execution_truth*.py, matching tests, matching fixtures
  • streamer fix → exact src/hermes/streamer.py plus targeted tests

Do not include unrelated source changes automatically just because they are code.

3. Stage selectively

Stage only the inferred relevant files. Never stage all dirty files.

Use:

git add <explicit paths>

4. Verify staged scope before commit

Always run:

git diff --cached --name-only
git diff --cached --stat
git status --short

Verify explicitly:

  • no logs staged
  • no runtime state staged
  • no cache/binaries staged
  • no unrelated subsystem staged
  • staged paths match the requested commit intent

If suspicious files appear, unstage them or stop.

5. Conventional commit message

Prefer these prefixes:

  • feat:
  • fix:
  • refactor:
  • test:
  • docs:
  • chore:

If the user’s message lacks a valid prefix, suggest a corrected version before commit.

6. Operator approval gate

Do not auto-commit if any of these are true:

  • ambiguous file group present
  • multiple unrelated subsystems staged
  • runtime-sensitive file included unexpectedly
  • a Bucket H file would be committed
  • the staged set is larger than expected for the request

In those cases, ask a short scope question and wait.

7. Commit and post-commit verification

After commit, run:

git show --stat --oneline HEAD
git diff --cached --name-only
git status --short

Report:

  • commit hash
  • files included
  • notable dirty files intentionally left out
  • whether runtime noise remains uncommitted

Risk policy

Low risk

Small tree, mostly one category, no runtime-critical dirt. Proceed normally but still summarize buckets.

Moderate risk

Mixed categories, some core code dirty, risky action coming. Recommend selective baseline before continuing.

High risk

Large mixed tree, core runtime dirty, noise obscures real work. Switch to hygiene-first mode before more feature work.

Critical risk

Huge mixed tree plus stale HEAD plus runtime-critical dirt plus pending restart/deploy. Strongly recommend immediate selective preservation before any risky next step.

Edge cases

Real code in noisy directories

Example:

  • shared/hermes-acp/execution_truth_slice.py

This is real code. Inspect it. Do not auto-exclude it because of the directory.

Fixtures that look like logs

Example:

  • tests/fixtures/execution_truth_streamer_2026_04_10_sample.log

This is a committed test fixture, not runtime noise.

Runtime-sensitive source files

Example:

  • src/hermes/streamer.py

These require stricter operator confirmation before commit.

Already-staged junk

If bad paths are already staged, verify and unstage before commit instead of proceeding.

Fail-safe rule

If confidence is not high, do not commit. Show:

  • candidate include list
  • suspicious/ambiguous list
  • recommended commit message
  • one short operator question

The fail-safe is simple: when uncertain, ask before commit.

Output style

When using this skill, report in this order:

  1. dirty-tree summary
  2. bucket classification
  3. recommended include-now list
  4. recommended exclude/preserve list
  5. staged verification
  6. commit recommendation or operator question

What success looks like

The skill is successful when Hermes consistently does this:

  • detects dirty tree risk early
  • separates runtime noise from real changes
  • stages only the requested source/test/config files
  • verifies staged scope before commit
  • uses conventional commit messages
  • avoids dragging runtime junk into history
  • asks the operator whenever scope is not obvious

平台分布

OpenClaw

73.8%
按下载量换算584

安全审计

VirusTotal

通过

ClawScan

通过

Static analysis

通过

权限和风险

需要联网

该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

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。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills