πŸ•ΈοΈ Ada Research Browser

skill-audit-2026-02-28.md
← Back

Skill Security Audit β€” 2026-02-28

⚠️ Immediate Attention: NONE

No malicious or suspicious skills detected. All installed skills are clean.


Summary

Threat Intelligence

Cross-referenced against ClawHavoc campaign (discovered Feb 2026): - 341β†’1,184 malicious skills found on ClawHub marketplace - Attack pattern: fake "Prerequisites" sections directing users to download AMOS (Atomic macOS Stealer) via password-protected ZIPs or obfuscated glot.io scripts - IOC domains: glot.io snippets, app-distribution.net, IP 91.92.242.30 - Target categories: crypto/Solana wallets, Polymarket bots, YouTube tools, ClawHub typosquats

None of these IOCs or patterns were found in any installed skill.

Scan Results

NPM-Bundled Skills (~/.npm-global/lib/node_modules/openclaw/skills/)

Skill Status Notes
1password βœ… Clean
apple-notes βœ… Clean References pip install (legitimate)
apple-reminders βœ… Clean
bear-notes βœ… Clean
blogwatcher βœ… Clean
blucli βœ… Clean
bluebubbles βœ… Clean
camsnap βœ… Clean
canvas βœ… Clean References fetch (documentation only)
clawhub βœ… Clean
coding-agent βœ… Clean References npm install for pi agent (documented)
discord βœ… Clean
eightctl βœ… Clean
gemini βœ… Clean
gh-issues βœ… Clean
gifgrep βœ… Clean
github βœ… Clean
gog βœ… Clean
goplaces βœ… Clean
healthcheck βœ… Clean
himalaya βœ… Clean
imsg βœ… Clean
mcporter βœ… Clean
model-usage βœ… Clean
nano-banana-pro βœ… Clean base64 usage for image data (legitimate)
nano-pdf βœ… Clean
notion βœ… Clean
obsidian βœ… Clean
openai-image-gen βœ… Clean base64 for image decode (legitimate)
openai-whisper βœ… Clean
openai-whisper-api βœ… Clean curl for API calls (legitimate)
openhue βœ… Clean
oracle βœ… Clean
ordercli βœ… Clean
peekaboo βœ… Clean
sag βœ… Clean
session-logs βœ… Clean
sherpa-onnx-tts βœ… Clean
skill-creator βœ… Clean
slack βœ… Clean
songsee βœ… Clean
sonoscli βœ… Clean
spotify-player βœ… Clean
summarize βœ… Clean
things-mac βœ… Clean
tmux βœ… Clean
trello βœ… Clean
video-frames βœ… Clean
voice-call βœ… Clean
wacli βœ… Clean
weather βœ… Clean
xurl βœ… Clean

Workspace Skills (~/.openclaw/workspace/skills/)

Skill Status Notes
agent-autonomy-kit βœ… Clean
api-gateway βœ… Clean
auto-updater βœ… Clean
automation-workflows βœ… Clean
blogwatcher βœ… Clean
deep-research-pro βœ… Clean
find-skills βœ… Clean
frontend-design βœ… Clean
github βœ… Clean
gmail βœ… Clean
home-assistant βœ… Clean curl for HA API (legitimate, uses HA_TOKEN)
humanizer βœ… Clean
mcporter βœ… Clean
n8n βœ… Clean Has Python scripts for n8n API (legitimate)
nano-banana-pro βœ… Clean
obsidian βœ… Clean
openai-whisper βœ… Clean
outlook-api βœ… Clean
self-improving-agent βœ… Clean References SOUL.md/AGENTS.md in documentation context only (suggests where to promote learnings)
summarize βœ… Clean
telegram βœ… Clean
youtube-watcher βœ… Clean

What Was Checked

  1. Malicious URLs/domains: grep for curl, wget, fetch to external domains β€” only legitimate API calls found
  2. System file tampering: grep for SOUL.md/MEMORY.md/AGENTS.md write/modify β€” only documentation references found (self-improving-agent suggests edits, doesn't perform them autonomously)
  3. Obfuscated code: grep for base64, eval, atob, encoded hex β€” only legitimate image processing (base64 decode of API image responses)
  4. Binary downloads: grep for npm install, pip install, chmod +x, /tmp β€” only documented tool installation instructions
  5. ClawHavoc IOCs: grep for glot.io, openclaw-agent, app-distribution.net, 91.92.242.30 β€” zero matches
  6. Typosquat names: checked for crypto/solana/phantom/polymarket/wallet/trading skill names β€” zero matches

Recommendations

  1. No immediate action needed β€” all skills are clean
  2. Consider pinning skill versions and reviewing diffs on updates (ClawHavoc has been expanding)
  3. The auto-updater skill auto-updates from ClawHub β€” monitor its update summaries for new skill installs from unknown authors
  4. Re-run this audit periodically, especially after installing new skills from ClawHub