Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
Version History
IBT 2.9.2: remove private data, make examples generic, remove proprietary homepage, clarify USER.md is agent workspace
What This Skill Does
IBT is an execution framework that structures agent behavior around a repeating control loop: Observe, Parse, Plan, Commit, Act, Verify, Update, Stop. It adds trust calibration, approval gates, error classification, and preference learning to help agents act with initiative while avoiding reckless or irreversible mistakes.
Combines a single structured loop with trust and safety rules, so an agent can move fast on routine tasks and automatically slow down for high-impact or ambiguous ones without needing separate frameworks for each concern.
When to use it
- Pausing before sending an email draft until the user approves
- Stopping cleanly mid-task when a user says halt or wait
- Retrying a failed network call with explicit limits before surfacing the error
- Learning a user's preferred response length to skip repeated clarifying questions
- Previewing file deletions and stating scope before executing
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: Delete all log files older than 30 days from the production server
- 1Observe: flags destructive action on a production system
- 2Parse: confirms goal is log cleanup scoped to files older than 30 days
- 3Plan: list matching files first to verify scope before touching anything
- 4Commit: show the file list and deletion count, request explicit approval
- 5Act: delete only after the user confirms
Logs deleted with a count and path summary confirmed against the original criteria