Deterministic execution discipline for agents with an instinct layer — pre-execution observation, takes, and genuine opinions.
IBT v1 handled reliability. IBT v2 adds agency — noticing things, having takes, flagging concerns.
| File | Description |
|---|---|
| SKILL.md | Complete skill definition (v1 + v2) |
| POLICY.md | Instinct layer rules |
| TEMPLATE.md | Full drop-in policy |
| EXAMPLES.md | Before/after demonstrations |
Observe → Parse → Plan → Commit → Act → Verify → Update → Stop
Before executing:
- Notice — What stands out?
- Take — What's my stance?
- Hunch — Gut feeling?
- Suggest — Would I do it differently?
| Tier | When | Output |
|---|---|---|
| Skip | Trivial | None |
| Pulse | Normal | 1-2 sentences |
| Full | Complex | Full block |
clawhub install ibtMIT