My heavily opinionated configurations, instructions, policies, and tooling for the agent harnesses I use.
A sanitized, inert-by-default snapshot of how I configure and instruct the agent harnesses I use. It covers global instructions, execution policies, permissions, harness-specific configuration, reusable agent assets, and validation tooling. It is published for reference and experimentation—not as a drop-in security baseline.
The public repository is intentionally incomplete. My private dotfiles remain the source of truth; configurations, instructions, scripts, and tooling are added here only after they have been reviewed and sanitized for publication.
This repository is derived from the private layer of my dotfiles repository, which I started tracking in late 2024 when I began experimenting with LLMs for coding.
Labels:
✅: Published🚧: Exists privately / sanitization pending-: Not applicable or not used
| Usage Rank | Layer | Configs | Base Instructions | Instruction Rules | Custom Agents | Exec Policy | Sandbox Policy | MCPs | Skills | Plugins | Hooks |
|---|---|---|---|---|---|---|---|---|---|---|---|
| - | .agents/ |
- | ✅ | 🚧 | - | - | - | - | 🚧 | 🚧 | - |
| 1️⃣ | Codex | ✅ | ✅ | - | 🚧 | ✅ | ✅ | 🚧 | 🚧 | ✅ | 🚧 |
| 2️⃣ | OpenCode | ✅ | ✅ | - | 🚧 | ✅ | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 |
| - | Claude Code | 🚧 | ✅ | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 |
| - | Qwen Code | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 |
| - | Hermes | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 |
| - | Pi | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 |
| - | ZCode | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 |
| - | DeepSeek | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 |
| - | Crush | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 |
| - | Goose | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 |
| - | RooCode | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 |
| - | Cline | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 |
| - | Cursor | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 |
| - | GitHub Copilot | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 |
| - | Google Gemini | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 |
| - | G Antigravity | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 | 🚧 |
agents/: My portable.agents/configuration, installed at~/.agents/and used as a shared, vendor-agnostic layer across harnesses.codex/: The global user scope~/.codex/directory.claude/: The global user scope~/.claude/directory.qwen/: The global user scope~/.qwen/directory. (🚧)opencode/: The global user scope~/.config/opencode/directory.vendor/: Independently maintained or vendored agent skills, plugins, and supporting packages. (🚧)
I prefix files or directories that would be automatically loaded as harness
configurations or harness instructions with _.
This keeps the repository inert when cloned: merely checking out the repository must not cause an agent harness to interpret the examples as active instructions. Any other files or directories follow their officially documented naming conventions.
For example:
AGENTS.md→_AGENTS.mdCLAUDE.md→_CLAUDE.mdQWEN.md→_QWEN.md