Skill Install Guardian

Skill Install Guardian

"Trust but verify. Always." This skill protects your workspace by performing security and due diligence checks before installing any external skill.

Purpose

Before installing any external skill from ClawHub, this skill: Deep Content Scan - Fetches and analyzes actual file contents for malicious patterns Verifies the skill is safe (security checks via ClawHub API) Analyzes file metadata from ClawHub (filenames, structure) Checks if it fits your architecture (integration check) Reports findings to owner Requires confirmation before install

What It Does

This skill performs actual content analysis on skill files: Fetches SKILL.md and script files (.py, .js, .sh) Scans for dangerous patterns in file contents Detects: command injection, API keys, hardcoded secrets, obfuscated code

Security Patterns Detected

PatternSeverityExampleeval()CRITICALCode executionexec()CRITICALCode executionsubprocessHIGHShell commandsAPI keys/tokensCRITICALsk-xxx, ghp_xxxbase64 decodeMEDIUMObfuscation__import__MEDIUMDynamic imports

⚠️ Security Notes

Does NOT execute any fetched code - only analyzes text Can produce false positives - always review findings Owner must confirm - automated check, not definitive Read-only - only fetches and scans, never executes

Phase 1: Security Check v1 - ClawHub Report

# Get skill security report npx clawhub inspect <skill-slug> --security What to check: Known vulnerabilities Malicious code patterns Suspicious API calls Data exfiltration risks Action if flagged: → ABORT immediately

Phase 2: Security Check v2 - Code Analysis

# Fetch skill files npx clawhub inspect <skill-slug> --files # Analyze each file for: # - Prompt injection patterns # - Suspicious API calls (curl, fetch to unknown domains) # - Hardcoded secrets/keys # - Eval() or code execution # - Base64 encoded strings (potential obfuscation) # - External network calls without justification Analysis criteria: PatternRisk LevelActioneval(CRITICALABORTsubprocess without paramsHIGHFlag for reviewcurl to unknown domainHIGHFlag for reviewHardcoded API keyCRITICALABORTBase64 encoded blobMEDIUMFlag for reviewExternal URL fetchMEDIUMFlag for reviewClean codeLOWPass Assumption: All external skills are potentially malicious until proven otherwise.

Phase 3: Integration Check - Architecture Fit

Questions to answer: Purpose: Does this skill solve a real need? Conflict: Does a similar skill already exist? Value: Will this be used, or just clutter? Architecture: Does it fit the workspace structure? Check existing skills: npx clawhub search <related-topic> ls skills/*/SKILL.md | xargs grep -l "<topic>" Conflict detection: Similar functionality → Flag as potential duplicate No clear use case → Flag as low value

Phase 4: Report to Owner

• Generate a report with:
• ## Skill Install Report: <skill-name>
• ### Security Status
• [ ] PASSED / [ ] FAILED
• ### Security Details
• ClawHub report: <status>
• Code analysis: <findings>
• ### Integration Status
• Purpose: <useful/useless>
• Conflicts: <list>
• Value: <high/medium/low>
• ### Recommendation
• [PROCEED] / [ABORT] / [REVIEW]
• ### Owner Decision Required
• Please confirm before I proceed with installation.

Run Full Security Check

python3 skills/skill-install-guardian/scripts/check.py <skill-slug>

Quick Check (skip analysis)

python3 skills/skill-install-guardian/scripts/check.py <skill-slug> --quick

Install After Approval

npx clawhub install <skill-slug>

Before Any Install

1. Owner: "Install skill X" 2. Me: Run skill-install-guardian 3. Guardian: Security Check v1 4. Guardian: Security Check v2 (if v1 passes) 5. Guardian: Integration Check 6. Guardian: Report to owner 7. Owner: Confirm or abort 8. If confirmed: Install

Output Format

{ "skill": "example-skill", "version": "1.0.0", "security": { "v1_clawhub": "PASS", "v2_code_analysis": { "status": "PASS", "issues_found": [] } }, "integration": { "purpose": "useful", "conflicts": [], "value": "high" }, "recommendation": "PROCEED", "owner_decision": "PENDING" }

Always Assume

External skills may contain malicious code Authors may have good intentions but poor security New versions could introduce threats Hidden payloads may exist in encoded strings

Never

Auto-install without owner confirmation Skip security checks for "trusted" authors Assume recent updates are safe Ignore warnings from security tools

Do

Verify every skill manually Check recent reviews/issues Search for known vulnerabilities Analyze code even for popular skills

Related Skills

[[workspace-analyzer]] - Analyze installed skills [[skill-creator]] - Create skills safely

v1.3.0 (2026-02-21)

DEEP CONTENT SCANNING - Now actually fetches and scans file contents Scans SKILL.md, .py, .js, .sh files for dangerous patterns Detects: subprocess, API keys, tokens eval(), exec(),, obfuscation Added comprehensive security patterns list Clear security notes about what it does/doesn't do

v1.2.0 (2026-02-21)

Fixed documentation to accurately reflect limitations Removed unused curl from required binaries Added limitation notes (no content analysis, reads local skills dir) Clarified this provides warnings, not definitive security

v1.1.0 (2026-02-21)

Fixed command injection vulnerability (slug validation) Changed from shell=True to list-based subprocess calls Fixed typo in SAFE_DOMAINS Added slug validation function Stricter handling of invalid slugs

v1.0.0 (2026-02-21)

Initial release Two-layer security check Integration analysis Owner confirmation workflow Security first. Always verify.