https://github.com/samzong/agent-brains
Personal agent brain catalog for coding agents: AGENTS.md, SOUL.md, MEMORY.md, skills, workflows, and a bundled loader.
https://github.com/samzong/agent-brains
agent-brains agent-skills agents-md ai-agents claude-code codex coding-agents cursor
Last synced: 3 days ago
JSON representation
Personal agent brain catalog for coding agents: AGENTS.md, SOUL.md, MEMORY.md, skills, workflows, and a bundled loader.
- Host: GitHub
- URL: https://github.com/samzong/agent-brains
- Owner: samzong
- License: mit
- Created: 2026-06-08T14:33:24.000Z (about 1 month ago)
- Default Branch: main
- Last Pushed: 2026-06-09T03:59:36.000Z (about 1 month ago)
- Last Synced: 2026-06-09T05:09:27.149Z (about 1 month ago)
- Topics: agent-brains, agent-skills, agents-md, ai-agents, claude-code, codex, coding-agents, cursor
- Language: Python
- Size: 48.8 KB
- Stars: 1
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
- Agents: AGENTS.md
Awesome Lists containing this project
README
# agent-brains
Portable **domain agent brains** for coding agents (Codex, Claude Code, Cursor, and others that read markdown context).
An agent brain is a long-lived context pack for a recurring domain. It keeps identity, judgment criteria, style, durable memory, domain methods, and stable SOPs in one loadable structure so the agent does not start from a blank context every session.
See [docs/AGENT_BRAIN_SPEC.md](./docs/AGENT_BRAIN_SPEC.md) for the full specification.
**Skill vs workflow vs brain** (see spec for detail):
- `skill` — reusable method or tool capability (**how**)
- `workflow` — end-to-end SOP for stable business behavior (**what process**)
- `agent brain` — identity, memory, and domain judgment (**who**, on **what history and standards**)
## Create brains
| Name | Purpose |
|------|---------|
| `agent-creator` | Create, audit, and evolve agent brains |
## Bundled skills
| Name | Purpose |
|------|---------|
| `agent-loader` | List and load repo-scoped and user-scoped agent brains |
Built-in loader: [`skills/agent-loader/`](./skills/agent-loader/). Copy or symlink it into your agent client's global skills directory to use it outside this repo.
Discovery checks repo-local `.agents/agents` directories from the current working directory up to the repository root, then `$HOME/.agents/agents`. It reads only `AGENTS.md` frontmatter and reports `name`, `description`, `scope`, and `path`. On load, `agent-loader` resolves the agent path, then reads `AGENTS.md`, `SOUL.md`, and `MEMORY.md` first. `memories/`, `skills/`, and `workflows/` are lazy-loaded only when the task needs them or core files point to them.
## Optional local routing
For Codex users, [`codex-agents-local`](https://github.com/samzong/codex-agents-local) can add a private `AGENTS.local.md` prompt that nudges Codex to choose an agent brain only when useful:
```md
When a task is domain-specific or recurring enough to benefit from a personal agent brain, proactively use `agent-loader` to list/resolve candidates, load the single best-matching agent core before working, ask only if the match is ambiguous, and keep skills/workflows/memories lazy-loaded.
```
## Brain structure
Each brain lives under `agents/`:
```text
agents/
└── agent-name/
├── AGENTS.md
├── SOUL.md
├── MEMORY.md
├── memories/
├── skills/
│ └── SKILL_NAME/
│ └── SKILL.md
└── workflows/
└── WORKFLOW_NAME/
└── WORKFLOW.md
```
- `AGENTS.md` — identity, responsibilities, rules, boundaries (`name`, `description` frontmatter required)
- `SOUL.md` — style, judgment, taste, and tone
- `MEMORY.md` — durable main memory
- `memories/` — optional focused sub-memories (by date or topic)
- `skills//SKILL.md` — optional reusable methods and tools
- `workflows//WORKFLOW.md` — optional end-to-end SOPs
Keep brains narrow. ***A useful brain knows one domain well instead of trying to manage every task.***
## Related projects
- [AGENTS.md](https://agents.md) — open format for agent-facing project instructions
- [OpenClaw workspace](https://docs.openclaw.ai/concepts/agent) — runtime-injected workspace files (similar file semantics, different loading model)
## License
MIT