Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
Version History
- Skill renamed from "moltbot, openclaw-best-practices" to "moltbot-best-practices" - Description and keywords updated for clarity and broader tool coverage (now mentions ChatGPT, Copilot, GitHub Copilot, developer/dev tools, TypeScript, LLM) - Minor text and formatting tweaks for conciseness and consistency in SKILL.md - Version bumped to 1.1.3
What This Skill Does
A set of behavioral rules for AI coding agents derived from documented real-world failures. Covers confirmation loops, draft-before-publish flows, failure escalation, and stop-signal handling. Applies to agents running in Cursor, Claude Code, ChatGPT, and GitHub Copilot.
Instead of generic prompt-engineering advice, these rules map directly to failure modes observed in live agent sessions, making them easier to apply and validate.
When to use it
- Preventing an agent from publishing a social post without showing a draft first
- Stopping an agent from retrying broken browser automation indefinitely
- Getting an agent to ask whether a task should run in background or foreground
- Enforcing that an agent reads all queued user messages before acting
- Configuring session memory so an agent retains context across compaction
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: Write a long form post about our product launch and publish it
- 1Reads all queued messages before acting to check for corrections or cancellations
- 2Asks clarifying question: 'Long form post — do you mean an X Article or a thread?'
- 3Confirms scope: 'You want an X Article about the product launch. I'll draft it and show you before posting. Correct?'
- 4Drafts the content and presents it to the user for review
- 5Waits for explicit approval before taking any publish action
Draft delivered to user for review; no publish action taken until user confirms