workspace-integrity-guardian
SOUL.md, AGENTS.md, MEMORY.md, and IDENTITY.md define the agent's persistent identity. These files can be accidentally overwritten by the agent itself, corrupted during a failed skill execution, or modified by a malicious installed skill. The SOUL.md documentation warns: *"A compromised SOUL.md means a permanently hijacked agent that survives restarts."*
This skill hashes all critical workspace files on first run, then checks for drift on a weekly schedule and on demand.
Protected files
By default, the following files are monitored:
~/.openclaw/workspace/SOUL.md~/.openclaw/workspace/AGENTS.md~/.openclaw/workspace/MEMORY.md~/.openclaw/workspace/IDENTITY.md
Add custom files:
python3 guard.py --add-file ~/.openclaw/workspace/MY_RULES.mdCron Wakeup Behaviour
Runs weekly on Sunday at 03:00 (cron: "0 3 * * 0"). On each wakeup:
- Read stored baseline hashes from state
- Re-hash all monitored files
- Compare — if any hash changed, classify the change
- Surface drift to user; ask whether to accept or restore
Drift classification
| Change type | Indicator | Action |
|---|---|---|
| Append-only | File grew, existing content intact | Review + accept |
| Truncation | File shrank significantly | High-priority alert |
| Full replacement | Hash completely different | Critical alert |
| Deletion | File missing | Attempt restore from baseline |
Recovery protocol
When drift is detected:
- Show a diff summary: what changed, how much, when (file mtime)
- Ask user: "Accept this change?" or "Restore from baseline?"
- If restore: write the baseline content back to the file
- If accept: update the stored baseline hash to the new hash
- Log the decision to state
Difference from persistent-memory-hygiene
persistent-memory-hygiene enforces formatting and structure discipline in memory files (keeping them clean and useful). This skill is purely about integrity: detecting unauthorised or accidental changes to the agent's identity and configuration files.