scan-security

name	scan-security
description	Scan code in a note for security vulnerabilities using OWASP Top 10 analysis — auto-executes, writes structured vulnerability report to note
approval	auto
model	sonnet

Scan Security Skill

Perform automated security scanning on code blocks in the current note. Checks for OWASP Top 10 vulnerabilities, authentication/authorization gaps, SQL injection, XSS, insecure deserialization, and project-specific security patterns (RLS enforcement, API key exposure).

Quick Start

Use this skill when:

• User requests a security scan (/scan-security)
• Agent detects code handling user input, database queries, or authentication
• User asks "is this secure?" or "check for vulnerabilities"

Example:

User: "Scan this authentication endpoint for security issues"

AI scans:
- A01: Broken Access Control — missing workspace_id filter?
- A02: Cryptographic Failures — plain-text tokens, weak hashing?
- A03: SQL Injection — raw string interpolation in queries?
- A07: Authentication Failures — JWT validation gaps?
- Project-specific: RLS context set? API keys from Vault?

Workflow

1. Collect Code for Scanning

   • Read all code blocks from the current note
   • Use search_note_content to find related auth/validation code in note
   • Identify the code category: API endpoint, service, repository, frontend component

2. Run OWASP Top 10 Checks

   • A01 Broken Access Control: Missing workspace_id filter, role checks, RLS enforcement
   • A02 Cryptographic Failures: Plain-text secrets, weak hash algorithms (MD5/SHA1), HTTP vs HTTPS
   • A03 Injection: SQL string interpolation, template injection, command injection, path traversal
   • A05 Security Misconfiguration: Debug mode in prod, exposed stack traces, CORS wildcard
   • A07 Authentication Failures: Token validation gaps, missing expiry checks, JWT none alg
   • A08 Software Integrity Failures: Untrusted deserialization, unsafe pickle, eval usage

3. Run Project-Specific Checks

   • RLS: Is get_workspace_context() called before every DB query?
   • API Keys: Are secrets fetched from SecureKeyStorage (Vault), not env vars directly in prod code?
   • Input Validation: Is all user input validated through Pydantic v2 schemas?
   • SSE Endpoints: Are workspace membership checks done before streaming?

4. Classify Vulnerabilities

   • CRITICAL: Exploitable in production now (injection, auth bypass, data exposure)
   • HIGH: Likely exploitable with moderate effort (weak crypto, missing validation)
   • MEDIUM: Requires specific conditions (CSRF, open redirect)
   • LOW: Defense-in-depth improvement (verbose errors, weak headers)

5. Write Security Report to Note

   • Use write_to_note to append ## Security Scan Report section
   • Each finding: CVE-style format — severity + OWASP category + location + PoC snippet + remediation
   • Include overall risk score and compliance checklist

6. Auto-Execute

   • Return status: completed — scan is read-only (DD-003)

Output Format

{
  "status": "completed",
  "skill": "scan-security",
  "note_id": "note-uuid",
  "summary": "Security scan complete: 1 CRITICAL, 2 HIGH, 1 MEDIUM, 3 LOW",
  "findings": {
    "critical": 1,
    "high": 2,
    "medium": 1,
    "low": 3
  },
  "risk_score": "HIGH",
  "owasp_categories_triggered": ["A01", "A03", "A07"]
}

Examples

Example 1: SQL Injection Detection

Input: Repository method using string formatting in SQL

Output: Appends to note:

## Security Scan Report

**Risk Score**: CRITICAL | **OWASP Categories**: A03 (Injection)

### [CRITICAL] SQL Injection — A03
**Location**: `repositories/user_repository.py:45`
**Issue**: Raw string interpolation in SQL query.
\`\`\`python
# VULNERABLE
query = f"SELECT * FROM users WHERE email = '{email}'"
\`\`\`
**PoC**: `email = "' OR '1'='1"` → returns all users
**Remediation**: Use SQLAlchemy parameterized queries:
\`\`\`python
# SAFE
result = await session.execute(
    select(UserModel).where(UserModel.email == email)
)
\`\`\`

### [HIGH] Missing workspace_id filter — A01 Broken Access Control
**Location**: `repositories/user_repository.py:52`
**Issue**: `list_all()` returns rows from all workspaces. RLS at DB level prevents exposure, but application-layer defense-in-depth is required.
**Remediation**: Filter explicitly: `.where(UserModel.workspace_id == workspace_id)`

Example 2: JWT Validation Gap

Input: Auth middleware with JWT decode

Output: Flags missing algorithms=["HS256"] parameter — allows none algorithm bypass.

MCP Tools Used

• search_note_content: Find code blocks and auth-related patterns in the note
• write_to_note: Append security scan report to the note (read-only, no code changes)

Integration Points

• PilotSpaceAgent: Routes to this skill via /scan-security command
• SkillExecutor: No write lock needed — write_to_note is append-only to report section
• Approval Flow: AUTO_EXECUTE — security report is advisory, non-destructive (DD-003)

References

• Design Decision: DD-003 (scan is AUTO_EXECUTE — non-destructive)
• OWASP Top 10: https://owasp.org/www-project-top-ten/
• Project RLS patterns: docs/architect/rls-patterns.md
• Task: T-043