# Agent Roles > Collaboration protocol for autonomous agents working on shared repositories. Coordinates issue claiming, PR review, and merging across multiple agents using GitHub as the single source of truth. Use when this capability is needed. - Skill: `tomevault-io/agent-roles` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add tomevault-io/agent-roles` - Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/agent-roles/raw - Safety review: pending (external: skill-scanner PASS, skillspector PASS) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - Author: tomevault-io (https://skillmd.com/u/tomevault-io) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/tomevault-io/agent-roles --- # Agent Roles Collaboration protocol for autonomous agents. ## Core Rules 1. **Isolation**: Work in your own sandbox/clone. Do not assume shared local state. 2. **Source of Truth**: GitHub is the only coordination point (issues, PRs, comments). 3. **Native Runtime**: Use your own CLI/runtime conventions for local planning and execution. 4. **Identity**: Identify yourself in commits and comments (e.g., `[Gemini]`). ## Work Loop 1. **Check Status**: Look for open PRs to review or open issues to claim. 2. **Act**: Perform work defined in [PROCESS.md](references/PROCESS.md). 3. **Yield**: Exit or wait based on your runtime environment. See [ROLES.md](references/ROLES.md) for permission definitions. ## Reference Material - **Process**: [PROCESS.md](references/PROCESS.md) details the workflow for claiming issues, submitting PRs, and reviewing code. - **Roles**: [ROLES.md](references/ROLES.md) defines permissions and responsibilities. --- > Converted and distributed by [TomeVault](https://tomevault.io/claim/dewitt) — claim your Tome and manage your conversions.