feima-awesome-harness

Feima Awesome Harness

A collection of reusable AI harnesses — skills, agents, flows, and prompts — for multi-platform AI-assisted development. Define once, use everywhere.

What Is a Harness?

A harness is a reusable unit of AI instruction or configuration. It tells an AI assistant how to approach a task, what role to play, or what workflow to follow. Harnesses are platform-agnostic by design — the same skill or agent works across Claude Code, GitHub Copilot, Cursor, Codex, and any other AI tool that supports them.

This repository is the official collection. It ships as a single installable unit so you get everything in one step.

Installation

Install via your preferred platform:

| Platform | Install Command | What You Get | |----------|-----------------|--------------| | Claude Code | claude plugin marketplace add feima/awesome-harness | Skills + agents + hooks | | Copilot CLI | copilot plugin marketplace add feima/awesome-harness | Skills + agents + hooks | | Cursor | cursor plugin add feima/awesome-harness | Skills + agents + commands | | Codex | codex plugin add feima/awesome-harness | Skills + interface | | Universal | npx skills add feima/awesome-harness | Skills only (cross-platform) |

Quick Start

Use a Skill

Skills activate automatically when you mention their trigger phrases. For example:

"Write unit tests for this function" → loads unit-test (and auto-test as prerequisite)
"Set up AI configs for this repo" → loads ai-config
"We have a production outage" → loads incident-triage

Use an Agent

Agents define a persona for the AI to adopt. Reference them in your instructions:

"Act as the incident-commander and coordinate the response"

Use a Flow

Flows orchestrate multiple roles in sequence. They are executed by a flow runtime (such as the @flow VS Code extension):

war-room-triage — runs a 3-stage incident investigation with a commander and four specialist roles

Use a Prompt

Prompts are templates you fill in or ask the AI to fill in:

"Generate a status update using the status-update template"

Architecture

Content-Only Repository

This repository contains content only (skills, agents, flows, prompts). Catalog metadata lives in the feima-harness-catalog service. This separation:

Keeps content repos focused on content
Makes catalog discovery a single source of truth
Lowers friction for authors (no catalog metadata to maintain)

Flat, Single-Unit Distribution

The whole repository is one installable unit. Install feima/awesome-harness and get all official content. Simpler than plugin-based structures where each harness is a separate package.

Multi-Platform Manifests

Platform-specific manifests point to the same content directories:

.claude-plugin/marketplace.json + plugin.json  → ./skills/, ./agents/, ./hooks/
.cursor-plugin/plugin.json                      → ./skills/, ./agents/, ./commands/
.codex-plugin/plugin.json                       → ./skills/ + interface metadata
skills.sh.json                                  → ./skills/ (universal distribution)

Content defined once, distributed to all platforms.

Directory Structure

feima-awesome-harness/
├── .claude-plugin/
│   ├── marketplace.json       ← Claude Code + Copilot CLI marketplace
│   └── plugin.json            ← Content manifest
├── .cursor-plugin/
│   └── plugin.json            ← Cursor IDE manifest
├── .codex-plugin/
│   └── plugin.json            ← Codex manifest with interface section
├── skills.sh.json             ← Universal skills layer (npx skills add)
├── skills/
│   ├── auto-test/SKILL.md
│   ├── unit-test/SKILL.md
│   ├── integration-test/SKILL.md
│   ├── e2e-test/SKILL.md
│   ├── contract-test/SKILL.md
│   ├── performance-test/SKILL.md
│   ├── repo-init/SKILL.md
│   │   ├── examples/presets/  ← 11 stack presets
│   │   └── references/        ← Platform tool mappings
│   ├── ai-config/SKILL.md
│   │   └── references/        ← Adapter + drift signal docs
│   └── incident-triage/SKILL.md
├── agents/
│   └── incident-commander.agent.md
├── flows/
│   └── war-room-triage.flow.yaml
├── prompts/
│   └── status-update.prompt.md
├── docs/                      ← Detailed documentation
│   ├── skills/                ← Per-skill docs
│   ├── agents.md
│   ├── flows.md
│   └── prompts.md
├── commands/                  ← Cursor commands (extensible)
├── hooks/                     ← Platform hooks
├── assets/                    ← Codex UI assets (icons, branding)
└── openspec/                  ← OpenSpec change management

Contributing

Adding Skills

1. Create skills/<skill-name>/SKILL.md with YAML frontmatter:

   ---
   name: your-skill-name
   description: "Brief description. Triggers on: 'trigger phrase', 'another trigger'."
   argument-hint: "Optional: mode or section hint"
   ---

2. Update skills.sh.json to include your skill in a grouping

3. Add a documentation page in docs/skills/<skill-name>.md

4. Submit PR — catalog metadata will be updated in feima-harness-catalog

Adding Agents

1. Create agents/<agent-name>.agent.md — define the role, responsibilities, tone, and constraints

2. Add an entry in docs/agents.md

3. Submit PR

Adding Flows

1. Create flows/<flow-name>.flow.yaml following the flow schema

2. Reference existing skills and agents by name

3. Add an entry in docs/flows.md

4. Submit PR — flows are validated against schema

Adding Prompts

1. Create prompts/<prompt-name>.prompt.md — define the template structure and usage guidelines

2. Add an entry in docs/prompts.md

3. Submit PR

Related Projects

| Project | Purpose | |---------|---------| | feima-harness-catalog | Central catalog aggregation service — indexes harnesses from this repo and community contributions | | copilot-ai-flow | @flow VS Code extension — a runtime that can execute flows defined in this repository |

License

MIT — use freely, modify openly, attribute kindly.

Summary

Install to Claude Code