Token导航 LogoToken导航

Agent Skill · 研究检索

project-harness(项目线束)

project-harness 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。

project-harness 是研究检索类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为操作浏览器;许可证为 MIT,暂无第三方扫描结果,按安装量在研究检索类的 73,675 个 Skill 里排第 50,882。

总安装 605周安装 26GitHub Stars 公开资料未说明下载 212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

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:project-harness(项目线束)
来源仓库:https://github.com/srstomp/pokayokay
仓库路径:skills/project-harness
安装命令:
npx skills add srstomp/pokayokay --skill "project-harness"
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。

AgentSkills.tonpx skills
npx skills add srstomp/pokayokay --skill "project-harness"

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。

project-harness 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。适用于研究检索类任务,支持基于关键词和来源的信息筛选。

可通过 npx skills add srstomp/pokayokay --skill "project-harness" 安装使用。安装前建议确认权限范围和维护状态,注意可能涉及联网或文件读写。建议结合原始仓库文档核验具体用法和功能边界。

来源仓库内路径:skills/project-harness

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 Codex、Claude、Cursor、Gemini CLI;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 npx skills add srstomp/pokayokay --skill "project-harness";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

    它需要操作浏览器,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

project-harness 怎么安装?

在终端执行 npx skills add srstomp/pokayokay --skill "project-harness"。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

project-harness 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。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。

project-harness 需要什么权限?

权限级别是操作浏览器。该 Skill 可能涉及浏览器控制能力,使用时可能读取或操作网页内容,需要在受控环境中确认权限边界。

project-harness 是免费的吗?

来源标注的许可证是 MIT。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。

project-harness 用的人多吗?

公开数据是累计安装 605 次,近一周安装 26 次,下载 212 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。

project-harness 安全吗?

没有查到第三方扫描结果,安全性无法从公开资料判断,安装前请自行审查源码、依赖和命令行为。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。
  • 许可证明确为 MIT,商用和分发边界好判断。

先想清楚的地方

  • 没有第三方扫描结果,安全性只能靠自己读源码判断。
  • 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 Codex,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。

不太适合

  • 要求有安全扫描结论才肯安装的团队。
  • 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
project-harness

项目线束 · 本页

605—操作浏览器MIT未收录

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
project-harness
description
Orchestrates long-running AI development sessions with human checkpoint control. Uses ohno for task management, manages progress tracking, routes work to appropriate skills, and implements supervised/semi-auto/autonomous modes. Use this skill when starting work sessions, resuming interrupted work, or managing multi-session projects.

Project Harness

Orchestrate AI-assisted development with configurable human control.

Core Concept

This skill bridges the gap between fully manual Claude Code sessions and runaway autonomous agents. It provides structured handoffs between sessions while giving you control over when to intervene.

Integrated with ohno for task management via MCP.

┌─────────────────────────────────────────────────────────────┐
│                    SESSION START                            │
│                          │                                  │
│                          ▼                                  │
│              ┌──────────────────────┐                       │
│              │ ohno: get_session_   │                       │
│              │       context()      │                       │
│              └──────────┬───────────┘                       │
│                          │                                  │
│              ┌──────────────────────┐                       │
│              │ ohno serve           │◄── Browser access     │
│              └──────────┬───────────┘                       │
│                          │                                  │
│              ┌──────────────────────┐                       │
│              │ ohno: get_next_task()│                       │
│              └──────────┬───────────┘                       │
│                          │                                  │
│              ┌──────────────────────┐                       │
│              │ Route to skill       │                       │
│              └──────────┬───────────┘                       │
│                          │                                  │
│              ┌──────────────────────┐                       │
│              │ CHECKPOINT (by mode) │◄── Human decision     │
│              └──────────┬───────────┘                       │
│                          │                                  │
│              ┌──────────────────────┐                       │
│              │ ohno: done + sync    │                       │
│              │ Git commit           │                       │
│              └──────────────────────┘                       │
└─────────────────────────────────────────────────────────────┘

Subagent Execution Model

Tasks are implemented by fresh subagents, not inline in the coordinator session.

Why Subagents?

ProblemSubagent Solution
Context degradation in long sessionsFresh context per task
Accumulated assumptionsEach task starts clean
Token limit issuesContext discarded after task

Coordinator vs Implementer

Coordinator (this session)Implementer (subagent)
Reads ohno tasksReceives task from coordinator
Extracts contextWorks with provided context
Dispatches subagentsImplements, tests, commits
Processes resultsReports back
Updates ohnoNo ohno access
Triggers hooksNo hook access

See references/subagent-dispatch.md for details.

Quick Start

1. Initialize ohno

npx @stevestomp/ohno-cli init

2. Create Project Context (optional)

Create .claude/PROJECT.md for shared project context:

# Project Name

## Overview
Brief description of the project.

## Tech Stack
- Framework: Next.js 14
- Database: PostgreSQL
- Styling: Tailwind CSS

## Conventions
- Use TypeScript strict mode
- Follow existing patterns in codebase

3. Start Session

Use ohno MCP tools or CLI:

# Get context from previous sessions
npx @stevestomp/ohno-cli context

# See all tasks
npx @stevestomp/ohno-cli tasks

# Get recommended next task
npx @stevestomp/ohno-cli next

# Start kanban board
npx @stevestomp/ohno-cli serve

Operating Modes

SUPERVISED Mode (Default)

Human reviews after every task. Maximum control, slower pace.

Checkpoint behavior:

  • Task complete → PAUSE
  • Story complete → PAUSE
  • Epic complete → PAUSE

Use when: Starting new projects, unfamiliar domains, critical code.

SEMI-AUTO Mode

Human reviews at story/epic boundaries. Good balance.

Checkpoint behavior:

  • Task complete → Log and continue
  • Story complete → PAUSE
  • Epic complete → PAUSE

Use when: Established patterns, routine implementation.

AUTONOMOUS Mode

Human reviews at epic boundaries only. Maximum speed.

Checkpoint behavior:

  • Task complete → Skip
  • Story complete → Log and continue
  • Epic complete → PAUSE

Use when: Well-defined specs, trusted patterns, time pressure.


Session Protocol

Starting a Session

## Session Start Checklist

1. [ ] Get session context: `ohno context` or MCP get_session_context()
2. [ ] Read .claude/PROJECT.md if exists
3. [ ] Check task list: `ohno tasks`
4. [ ] Start kanban: `ohno serve`
5. [ ] Check git status - clean working directory?
6. [ ] Get next task: `ohno next`
7. [ ] Announce plan - tell human what you'll do

Using ohno MCP:

1. get_session_context() → Understand previous session
2. get_tasks() → See all work
3. get_next_task() → Pick what to do

Hook Integration

Hooks execute automatically at lifecycle points. Do not call manually.

Work Loop with Hooks

[pre-session hooks]  <- Verify clean state

WHILE tasks remain:
  [pre-task hooks]   <- Check blockers

  1. Get next task (ohno next)
  2. Start task (ohno start <id>)
  3. Extract task context
  4. Dispatch subagent        <- Fresh context per task
  5. Process subagent result

  [post-task hooks]  <- GUARANTEED: sync, commit

  6. CHECKPOINT based on mode

[post-session hooks] <- Final sync, summary

Hook Execution Output

When hooks run, you'll see:

## Hooks: post-task

| Action | Status | Time |
|--------|--------|------|
| sync | ✓ | 0.3s |
| commit | ✓ | 0.5s |

Continuing...

Mode-Specific Behavior

Modepost-task hooks
supervisedsync only
semi-autosync, commit
autonomoussync, commit, quick-test

Hook Failures

Hooks are fail-safe:

  • Warnings don't stop the session
  • Critical failures pause for human input
  • All results logged for review

See hooks/HOOKS.md for configuration.


During Work

For each task:

## Task: [task-id] Create grid component

### Plan
- Create responsive grid using CSS Grid
- Support 1-4 column layouts
- Add to component library

### Implementation
[Work happens here]

### Verification
- [ ] Component renders
- [ ] Responsive breakpoints work
- [ ] Exported from index

### Post-Task
Hooks handle: sync, commit (mode-dependent)
Checkpoint triggers based on mode

Ending a Session

## Session End Checklist

1. [ ] Session notes logged via ohno
2. [ ] No broken code left
3. [ ] Clear summary of what was done
4. [ ] Clear next steps documented

*Note: post-session hooks handle final sync and summary automatically.*

Checkpoint Protocol

PAUSE Checkpoint

Agent stops completely and waits for human input.

## CHECKPOINT: Task Complete

**Completed**: task-abc123 - Create grid component
**Status**: Awaiting your review
**Kanban**: http://localhost:3456 (run `ohno serve`)

### What I Did
- Created `GridLayout.tsx` component
- Added responsive breakpoints (sm, md, lg, xl)
- Exported from components/index.ts

### What I'd Do Next
- task-def456: Create dashboard header

### Your Options
1. **Continue** - Proceed to next task
2. **Modify** - Change approach before continuing
3. **Pause** - Stop session here
4. **Switch** - Work on different task

Waiting for your decision...

REVIEW Checkpoint

Agent continues but flags work for later review.

## REVIEW FLAG: Story Complete

**Completed**: Dashboard Layout (3 tasks)
**Continuing to**: Dashboard Widgets
**Kanban**: Synced ✓

Flagged for review.

NOTIFY Checkpoint

Agent logs and continues without stopping.

## ✓ Task Complete: task-abc123 - Create grid component
Kanban synced → Continuing to next task...

Skill Routing

Based on task type, load relevant skill for domain knowledge:

Task TypeSkill
API designapi-design
API testsapi-testing
UI componentsaesthetic-ui-designer
User flowsux-design
User researchpersona-creation
Accessibilityaccessibility-auditor
Architecturearchitecture-review
SDK/packagesdk-development
Marketingmarketing-website

Skill Invocation

When routing to a skill:

## Invoking Skill: api-design

**Context**: Feature - API Endpoints
**Task**: task-xyz - Design user endpoints

Reading skill documentation...
[Skill takes over for this task]

ohno Integration

MCP Tools Available

Query Tools:

  • get_session_context() - Previous session notes, blockers, in-progress tasks
  • get_project_status() - Overall project statistics
  • get_tasks() - List all tasks
  • get_task(id) - Get specific task details
  • get_next_task() - Recommended next task
  • get_blocked_tasks() - Tasks with blockers

Update Tools:

  • start_task(id) - Mark task in-progress
  • complete_task(id, notes) - Mark task done
  • log_activity(message) - Log session activity
  • set_blocker(id, reason) - Block a task
  • resolve_blocker(id) - Unblock a task

CLI Commands

# Session management
ohno context              # Get session context
ohno status               # Project statistics

# Task management
ohno tasks                # List all tasks
ohno next                 # Get recommended next task
ohno start <id>           # Start working on task
ohno done <id> --notes    # Complete task with notes
ohno block <id> <reason>  # Set blocker
ohno unblock <id>         # Resolve blocker

# Kanban
ohno serve                # Start kanban server
ohno sync                 # Sync kanban HTML

Error Recovery

Build Failures

## ⚠️ Build Failed

**Error**: TypeScript compilation error in Dashboard.tsx
**Line 47**: Property 'user' does not exist on type '{}'

### Recovery Plan
1. Check recent changes (git diff)
2. Identify breaking change
3. Fix type error
4. Verify build passes
5. Block task if needed: `ohno block <id> "Build failure"`
6. Continue or escalate

Proceeding with recovery...

Blocked Tasks

# Block a task
ohno block task-abc123 "Waiting for API spec"

# View blocked tasks
ohno tasks --status blocked

# Resolve blocker
ohno unblock task-abc123

Anti-Patterns

Anti-PatternProblemFix
Skipping verificationStart on broken codeAlways verify first
No git commitsCan't recover from errorsCommit every task
No kanban syncStale visual stateRun ohno sync after changes
Giant tasksLose progress on failureKeep tasks ≤8 hours
Ignoring checkpointsLose human controlRespect mode settings
No session contextNext session confusedUse ohno context
Autonomous on new projectBad patterns amplifiedStart supervised

References

平台分布

Claude Code

29.09%
按下载量换算62

Codex

22.15%
按下载量换算47

Antigravity

17.41%
按下载量换算37

windsurf

13.31%
按下载量换算28

trae

9.39%
按下载量换算20

github-copilot

3.3%
按下载量换算7

安全审计

暂无安全审计结果可展示。

权限和风险

操作浏览器

该 Skill 可能涉及浏览器控制能力,使用时可能读取或操作网页内容,需要在受控环境中确认权限边界。

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