Docs for Agent (Coding Agent Usage)

If you want Codex / Claude Code to use Agently correctly, follow these steps.

---

1. Download the agent docs pack

• Online: /docs/agent_docs.zip

After unzip, you should see:

agent_docs/
  AGENT.md                # Agent instructions
  docs/                    # Scenario → capability → operations
  skills/                  # Task skills
  examples/                # Runnable examples
  outputs/                 # Real outputs
  model_configs/           # Model config references

---

2. Codex setup

Codex reads AGENTS.md at the project root.

Recommended steps:

1. Create AGENTS.md at the repo root.
2. Copy the content of agent_docs/AGENT.md into it.
3. Put agent_docs/ in your repo (root or docs/agent_docs/).
4. In AGENTS.md, point to the doc pack path, e.g.:

Agently Docs Pack is at: /agent_docs
Please read /agent_docs/docs/agent-systems/overview.md first.

---

3. Claude Code setup

Claude Code reads CLAUDE.md (supports multiple files in sub‑folders).

Recommended steps:

1. Create CLAUDE.md at the repo root.
2. Copy agent_docs/AGENT.md into it.
3. Put agent_docs/ in your repo and reference the path inside CLAUDE.md.

---

4. Recommended layout (generic)

<your_repo>/
  AGENTS.md  (Codex)
  CLAUDE.md  (Claude Code)
  agent_docs/
    docs/
    skills/
    examples/
    outputs/

---

5. Why this matters

• Ensures the agent follows Scenario → Capability → Operations.
• Avoids misuse of output() + instant + ensure_keys.
• Forces test + expected output per delivery batch.

For the human‑developer playbook, see: Agent Systems Playbook.