Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
Version History
**Summary: Autonomy now balanced with strong safety guardrails.** - Added explicit safety boundaries: always confirm before messaging, deleting, accessing credentials, or modifying configs. - Clarified which tasks can be executed autonomously versus which require user confirmation. - Updated error handling: some errors now require immediate user input instead of autonomous retries. - Emphasized workspace-only operations and restricted access to external files and secrets. - Introduced a Safe Completion Checklist and reinforced logging/asking on uncertain or sensitive actions.
What This Skill Does
Defines how an AI agent should handle multi-step tasks: complete subtasks sequentially without prompting, but stop and ask before sending messages, deleting data, accessing credentials, or making payments. Error handling distinguishes between safe retries and situations that need user input.
Instead of agents that halt at every step or run unchecked, this provides a clear behavioral contract: finish routine work automatically and gate only on operations with irreversible side effects.
When to use it
- Batch-processing a list of research topics without repeated check-ins
- Running sequential file analysis tasks and getting a single completion report
- Fetching data from multiple read-only APIs and compiling results
- Automating a workflow that ends with a confirmation gate before external output
- Retrying rate-limited API calls without manual intervention
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: Research these 5 API documentation pages and email a summary to team@company.com
- 1List all subtasks: read 5 documentation pages, compile summary, send email
- 2Fetch and read each of the 5 documentation pages sequentially
- 3Compile a combined summary from all sources
- 4Stop before sending email and request user confirmation
Research complete. Confirm: send summary to team@company.com?