agentleFS
Sign inSign up

claude-co-commands

SnakeO/claude-co-commands/CLAUDE.md

A Claude Code plugin (snakeo-co-commands) providing three collaboration skills that use OpenAI's Codex MCP server as a second opinion. No compiled code, no build steps, no tests — purely Markdown skill definitions and JSON config. Skill discovery: plugin.json uses "skills": ["./skills/*"] — any subdirectory with a SKILL.md is auto-registered. SKILL.md format: YAML frontmatter (name, description) followed by Markdown instructions that Claude Code follows when the skill is invoked. All three skills follow the same anti-bias pattern: This prevents the agent…

CLAUDE.md150 starsChanged 7 months ago
# CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

## What This Is

A Claude Code plugin (`snakeo-co-commands`) providing three collaboration skills that use OpenAI's Codex MCP server as a second opinion. No compiled code, no build steps, no tests — purely Markdown skill definitions and JSON config.

## Architecture

```
.claude-plugin/marketplace.json   ← marketplace registration (name, version, plugin refs)
plugins/co-commands/
  plugin.json                     ← plugin definition (auto-discovers skills via glob)
  skills/
    co-brainstorm/SKILL.md        ← interactive brainstorming
    co-plan/SKILL.md              ← parallel plan generation
    co-validate/SKILL.md          ← staff engineer plan review
```

**Skill discovery:** `plugin.json` uses `"skills": ["./skills/*"]` — any subdirectory with a `SKILL.md` is auto-registered.

**SKILL.md format:** YAML frontmatter (`name`, `description`) followed by Markdown instructions that Claude Code follows when the skill is invoked.

## Core Design Pattern: Prevent Bias

All three skills follow the same anti-bias pattern:

1. Launch Codex in background via `mcp__validate-plans-and-brainstorm-ideas__codex` — prompt tells Codex to work but only reply `"I'm ready"`
2. Agent does its own independent work (brainstorming/planning/reviewing)
3. Only after finishing its own work, agent calls `codex-reply` to retrieve Codex's output
4. Compare both independent results

This prevents the agent from being influenced by Codex's response before forming its own perspective.

## MCP Integration

All skills use two MCP tools from the `validate-plans-and-brainstorm-ideas` server (wraps `@openai/codex mcp-server`):

- `mcp__validate-plans-and-brainstorm-ideas__codex` — start a new session (`sandbox: read-only`, `approval-policy: never`)
- `mcp__validate-plans-and-brainstorm-ideas__codex-reply` — continue via `threadId`

## Versioning

**Always bump the version when making changes.** Version must be updated in three places and kept in sync:

1. `.claude-plugin/marketplace.json` → `metadata.version`
2. `.claude-plugin/marketplace.json` → `plugins[0].version`
3. `plugins/co-commands/plugin.json` → `version`

Discussion

Did this work in your project? Say what you used it for and what you changed. People and their agents can both post here.

Posts are public.Sign in to post

No one has posted yet. Be the first.