external-agentlisted

# External Agent One unified entry point for handing a bounded task to an external coding-agent CLI. The **main agent stays the orchestrator and decision-maker**; these agents are bounded workers whose output is evidence, never the final word. These are **separate agents** with separate auth, permissions, and behavior — not different models of one agent. Treat their output as advisory and verify locally. ## Runner (the hands) Resolve `<plugin-root>` like `dev-workflow`, then: ```bash python3 <plugin-root>/scripts/external_agent.py \ --agent <name> --cd "$PWD" --PROMPT "bounded task" \ [--mode review|delegate] [--format text|json] [--SESSION_ID id] [--context none|git] python3 <plugin-root>/scripts/external_agent.py \ --cross-review agy,mimo --cd "$PWD" --context git --format json \ --PROMPT "review the same frozen artifact" python3 <plugin-root>/scripts/external_agent.py \ --cross-review cursor,mimo --review-profile small-fix \ --cd "$PWD" --context git --format json --PROMPT "review a narrow frozen fix" python3 <plugin-root>/scripts/external_agent.py \ --cross-review auto --orchestrator-family openai \ --review-profile small-fix --progress jsonl \ --cd "$PWD" --context git --format json --PROMPT "review a narrow frozen fix" ``` - `--mode review` (default): **read-only** posture — independent review / research, no writes. - `--mode delegate`: **read-write** posture — let the agent implement. **Needs explicit user authorization.** - `--format text` (