SecurityClaw (Skill Scanner)

Install auto-scan service (recommended)

Install directly from public GitHub repo via npm:

npx github:mallen-lbx/SecurityClaw install

Run installer:

python3 scripts/install_securityclaw.py --skills-dir ~/.openclaw/skills --notify-config ~/.openclaw/securityclaw-notify.json

What installer does:

macOS: installs launchd agent for continuous watch scanning
Linux: installs systemd --user service for continuous watch scanning

Linux behavior when systemd is missing:

warns during installation
prints a concrete install command for detected package manager
offers to run install command automatically

Use the scanner script

Run the scanner (read-only by default):

python3 scripts/securityclaw_scan.py --skills-dir ~/.openclaw/skills

Quarantine anything suspicious (moves folders, no deletion):

python3 scripts/securityclaw_scan.py --skills-dir ~/.openclaw/skills --quarantine-dir ~/.openclaw/skills-quarantine --quarantine

Use an allowlist for known-safe exceptions (skill + hash + rules):

python3 scripts/securityclaw_scan.py --skills-dir ~/.openclaw/skills --allowlist ~/.openclaw/securityclaw-allowlist.json

Reports are saved by default to:

~/.openclaw/SecurityClaw_Scans
file format: Security_Scan-(MM)-(DD)-(YYYY)-(scan number) (example: Security_Scan-02-06-2026-001.json)
when quarantine candidates exist, an ELI5 summary is also written: Security_Scan-...-ELI5.md

The scanner also prints a user-facing summary with severity counts, top findings, confidence indicators, and the owner action menu.

If quarantine candidates are found, the Markdown report includes a dedicated proof section with 4 concrete findings per quarantined skill. The scanner tells the user where the ELI5 summary file was written.

Notifications

Configure notifications at ~/.openclaw/securityclaw-notify.json (see references/notification.example.json).

Then run:

python3 scripts/securityclaw_scan.py --skills-dir ~/.openclaw/skills --notify-config ~/.openclaw/securityclaw-notify.json --notify-on quarantine

Supported channels:

telegram (bot token + chat ID)
webhook (POST JSON payload)
stdout (local terminal output)

Auto-scan new skills

Run watch mode to automatically scan when a skill is added or changed:

python3 scripts/securityclaw_scan.py --skills-dir ~/.openclaw/skills --watch --watch-scan-on-start

How it works:

keeps a hash snapshot at ~/.openclaw/SecurityClaw_Scans/watch-state.json
polls ~/.openclaw/skills on an interval (--watch-interval, default 30s)
triggers a new scan when new/changed skills are detected
in watch mode, report files are created only when findings require review/quarantine
when a new skill is scanned, notifications are always sent

Scan log

SecurityClaw appends every scan (manual and auto) to a monthly log file:

directory: ~/.openclaw/SecurityClaw_Scans/Scan_Logs
file: <Month>.log (example: April.log)
line format per scan: scan completed 04-06-26 12:00:00

What to do when findings exist

If the report recommends quarantine for any skill:

Do not execute the skill.
Quarantine the skill folder.
Notify the owner with:
- skill name
- top reasons + file/line locations
- recommended action
Await owner instruction:
- Delete: remove quarantined skill
- Report: prepare public report / IOCs (no secrets)
- Allow: add allowlist entry and restore
- Scan all: deep scan everything

Optional: sandbox/dynamic checks (advanced)

Dynamic checks are optional and should run only after owner approval.

Prefer running unknown code with:
- no network egress
- read-only filesystem except a temp workspace
- no access to OpenClaw config/secrets

See references/sandboxing.md.

Files

scripts/securityclaw_scan.py — main scanner + quarantine
scripts/install_securityclaw.py — cross-platform scheduler installer (launchd/systemd)
references/rules.md — rule catalog (what we flag and why)
references/sandboxing.md — safe sandbox strategy + what to avoid
references/allowlist.example.json — allowlist template (skill/hash/rule suppression)
references/notification.example.json — notification channels template

securityclaw-skill