Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
Version History
No user-visible changes in this version. - No file changes were detected between versions 2.3.0 and 2.3.1. - Documentation, features, and security summary remain the same. - Behavior and configuration are unchanged.
What This Skill Does
Upgrades OpenClaw's agentic loop with persistent plan state, approval gates for risky operations, automatic retry with error diagnosis, context summarization, session checkpointing, and knowledge graph memory injection. A Mode dashboard provides visual configuration without editing JSON.
Plans survive across conversation turns and checkpoints, so multi-hour tasks can be interrupted and resumed without losing state or replaying completed steps.
When to use it
- Resuming a multi-step coding task after closing the agent mid-session
- Pausing before destructive shell commands to review and approve them
- Running parallel subtasks within a single agent goal
- Automatically retrying a failed file write with elevated permissions
- Keeping context window manageable during long refactor sessions
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: Build a REST API with authentication from scratch
- 1Detects planning intent and generates a step-by-step plan stored to ~/.openclaw/agent-state/
- 2Injects plan status into the system prompt before each turn
- 3Pauses at the git push step and requests human approval with a 15-second timeout
- 4Retries a failed dependency install by diagnosing the error and applying a modified command
- 5Creates an automatic checkpoint after each completed step
Completed plan with all steps marked done; checkpoint file available for restore if the session is interrupted