\r \r

Security Review Skill\r

\r

Trigger\r

Run this skill BEFORE installing ANY external package, tool, CLI, npm module, Python library, browser extension, or third-party integration.\r \r No exceptions. "Open source" is not a security clearance.\r \r

What This Skill Does\r

Spawns a security review sub-agent that performs a world-class, current-intelligence security assessment and produces a GO / NO-GO / CONDITIONAL verdict.\r \r

How to Use\r

\r

1. Spawn the review agent\r

sessions_spawn with model: anthropic/claude-sonnet-4-6, task: [security review prompt below]\r
```\r
\r
### 2. Standard Review Prompt Template\r
Fill in [PACKAGE NAME], [INSTALL COMMAND], [DESCRIPTION], [SOURCE URL]:\r
\r
```\r
You are a world-class security analyst. Perform a comprehensive security review before installation.\r
\r
PACKAGE: [PACKAGE NAME]\r
Source: [GitHub URL or npm/pypi link]\r
Install: [INSTALL COMMAND]\r
Description: [what it claims to do]\r
\r
Cover ALL 7 sections:\r
\r
1. LEGITIMACY & TRUST SIGNALS\r
   - Author/maintainer: GitHub profile, history, reputation\r
   - Stars, forks, contributors, last commit\r
   - Red flags: new account, copied code, suspicious activity\r
\r
2. LATEST SECURITY INTELLIGENCE (USE web_search)\r
   Search: "[name] security vulnerability", "[name] malware", "[name] CVE",\r
   "[name] data exfiltration", "[author] security issues", HackerNews/Reddit discussions\r
\r
3. SOURCE CODE ANALYSIS (USE web_fetch on raw GitHub files)\r
   - What does the entry point actually do?\r
   - Network calls — to where?\r
   - File system access beyond documented scope?\r
   - Obfuscated code?\r
   - npm/pip dependencies — any known-bad?\r
\r
4. DATA FLOW ANALYSIS\r
   - What data does it access? (conversations, files, env vars, API keys)\r
   - Where does data go? Local only or external?\r
   - Telemetry/analytics present?\r
   - Exfiltration risk for workspace content?\r
\r
5. PERMISSION SCOPE\r
   - System access required\r
   - Network access?\r
   - Touches env vars or config files?\r
   - Can it interfere with other processes?\r
\r
6. DEPENDENCY RISK\r
   - Full dependency list\r
   - Any known-vulnerable deps?\r
   - Recent supply chain attacks on dependencies?\r
\r
7. VERDICT\r
   GO / NO-GO / CONDITIONAL\r
   - Confidence: High/Medium/Low\r
   - Top 3 specific risks\r
   - If CONDITIONAL: exact conditions required before install\r
\r
Write complete review to:\r
C:\Users\hhx-sandbox2\.openclaw\workspace\logs\security-review-[package-name]-[YYYY-MM-DD].md\r
\r
Verify file exists and is complete before reporting done.\r
```\r
\r
### 3. Read the output\r
After the sub-agent completes, read the file and surface the verdict to Marcus.\r
\r
### 4. Marcus decides\r
Present the verdict clearly. Never install without explicit Marcus approval after a GREEN or CONDITIONAL review.\r
\r
## Standing Rules\r
\r
- **NO install without a completed security review** — not even "quick" ones\r
- **Web search is mandatory** — latest CVEs and community reports must be checked\r
- **Source code must be read** — not just the README\r
- **All reviews logged** to `logs/security-review-[package]-[date].md`\r
- **MEMORY.md updated** with verdict after each review\r
\r
## Review Log Index\r
| Date | Package | Verdict | File |\r
|------|---------|---------|------|\r
| 2026-03-04 | claude-subconscious | Pending | logs/security-review-claude-subconscious-2026-03-04.md |\r