Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
Version History
Security fix: agentType path traversal prevention (allowlist sanitization + path.resolve confinement), input length limits, control char stripping on correction content
What This Skill Does
Correction Memory makes agent corrections persistent across sessions. When an agent repeats a mistake you've already fixed, this skill stores your override and prepends it automatically the next time that agent type is spawned. Corrections expire after 30 days to prevent stale rules from accumulating.
Unlike one-off system prompt edits, corrections are stored per agent type and replayed automatically, so you never have to re-teach the same rule in a new session.
When to use it
- Stopping a CoderAgent from repeatedly using ESM imports instead of require()
- Teaching a writing agent to maintain a specific narrative voice across sessions
- Logging corrections via natural language without touching code
- Auditing all past corrections per agent type with a single shell command
- Sharing correction history across team members working in the same workspace
Example Workflow
Here's how your AI assistant might use this skill in practice.
User tells the main agent: 'Note that CoderAgent: used ESM import instead of require() — always use require() for Node.js stdlib modules'
- 1Main agent calls logCorrection() with agent type, issue description, and correct behavior
- 2correction-tracker.js writes the entry to memory/corrections/CoderAgent.jsonl
- 3Next time a task matching CoderAgent keywords is spawned, agent-context-loader detects the agent type
- 4Loader reads CoderAgent.jsonl and prepends a corrections preamble to the task
- 5CoderAgent receives the correction context before executing
CoderAgent uses require() consistently in all subsequent sessions without being reminded