Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
What This Skill Does
Auto-improve runs a persistent learning loop across sessions, loading past mistakes and project context at session start, then saving patterns or logging errors after each task. It maintains a `.antigravity.md` file per project to carry forward specific lessons. Metrics track whether repeated mistakes and verification failures decrease over time.
Unlike session-scoped context, this persists lessons to disk and reloads them automatically, so the agent doesn't repeat documented mistakes.
When to use it
- Avoiding the same coding mistake in a new session
- Getting a warning before touching code that caused past bugs
- Building a project-specific pattern library from successful approaches
- Reviewing what went wrong in a previous refactor before starting a new one
- Tracking whether first-attempt success rate improves over weeks
Example Workflow
Here's how your AI assistant might use this skill in practice.
User starts a new coding session on a project
- 1Load `.antigravity.md` from project root
- 2Recall tagged mistake memories for this project
- 3Warn about any known issues relevant to the current task
- 4Execute the task, then run verification (tests + lint)
- 5Save successful approach as a reusable pattern, or log the error with root cause and fix to `.antigravity.md`
Session ends with updated `.antigravity.md` and new memory entries; next session starts with those lessons pre-loaded