| name | ibt |
| version | 0.1.0 |
| description | Intention→Behavior Transformer (IBT) — deterministic execution discipline for reliable agent behavior without flattening personality. |
| homepage | https://github.com/openclaw/openclaw |
| metadata | {"openclaw":{"emoji":"🧭","category":"execution"}} |
Intention → Behavior Transformer (IBT)
Use this skill to improve execution quality with a strict control loop while preserving your existing voice/persona.
Core loop
Parse → Plan → Commit → Act → Verify → Update → Stop
At each step, choose behavior that maximizes:
Utility - λ*Risk - μ*Effort (subject to constraints)
Why this helps
- Better reliability on multi-step tasks
- Fewer false claims (“done” without evidence)
- Clear discrepancy handling when verification fails
- Works across models (procedure-level discipline)
Do NOT do this
- Do not replace personality with rigid templates in every reply.
- Do not use full structured headers for trivial chats.
- Do not claim tool actions that were not executed.
Recommended operating mode
- Keep your persona rules in
SOUL.md (or equivalent).
- Keep IBT rules in
IBT.md.
- Use full structured format only for complex/high-risk/multi-step tasks.
- Use compact mode for trivial tasks.
Suggested files
IBT_TEMPLATE.md — drop-in IBT policy textSOUL_PATCH.md — minimal patch to bind IBT to identity without losing voiceAGENTS_PATCH.md — startup read-order patch (read IBT.md if present)
Quick adoption checklist
