SentinelOps — Self-Healing OpenClaw Agent Ops (Runnable MVP)

A minimal ops agent that checks disk/log risk in an OpenClaw workspace and emits a daily markdown report.

• Track: Agent Track
• No browser automation
• Safe by default (read-only unless --heal)

What it does (MVP)

• Disk guardrail: reports disk usage and compares to a threshold (default 85%)
• Log growth guardrail: lists the largest files under the workspace
• Cron verifier (best-effort): checks whether expected substrings appear in crontab -l
• Self-heal (optional): if usage is above threshold, can safely truncate log-like files under the workspace

Quickstart (Demo)

cd /home/ubuntu/.openclaw/workspace/projects/moltiverse-hackathon/entries/05-sentinelops
python3 sentinelops.py --dry-run --workspace /home/ubuntu/.openclaw/workspace --top 15 --expect-cron "openclaw,gateway"
sed -n '1,200p' ops-report.md

Optional: self-heal (only if needed)

This only triggers if disk usage is above --threshold.

# still safe: shows what would be truncated/deleted
python3 sentinelops.py --heal --dry-run

# live: will truncate a few biggest log-like files under the workspace
python3 sentinelops.py --heal --max-heal 3

Output artifacts

• ops-report.md — daily ops report with disk usage, largest files, cron check, and actions taken

Proof (what to capture for submission)

• Terminal output showing the command run and OK
• ops-report.md (Disk section + Largest files table)
• If you used --heal: the Self-heal actions list in ops-report.md
• Optional: df -h output before/after (if healing)

Safety notes

• Healing only touches files under --workspace and only for log-like extensions: .log, .trace, .jsonl, .txt.
• It may remove __pycache__ directories.
• It does not delete arbitrary files.