lgbarn/shipyard/security-audit

Security Audit

<activation>

When to Use

• Working with any code that handles user input, authentication, or authorization
• Adding or updating dependencies
• Reviewing infrastructure-as-code (Terraform, Ansible, Docker, CloudFormation)
• Before claiming security posture is adequate
• When conversation mentions: security, vulnerability, CVE, OWASP, secrets, hardening

Natural Language Triggers

• "security check", "is this secure", "vulnerability scan", "audit security", "check for secrets"

</activation>

Core principle: Assume every change introduces risk until proven otherwise.

<instructions>

OWASP Top 10 Quick Checks

For every code change, verify:

• [ ] Injection: All user input parameterized/escaped -- no string concatenation in queries or commands
• [ ] Broken Auth: Passwords hashed properly (bcrypt/argon2), session tokens random, rate limiting on auth endpoints
• [ ] Data Exposure: Encryption at rest and transit, no sensitive data in logs/errors, proper key management
• [ ] XXE: XML parsers disable external entities
• [ ] Access Control: Authorization on every request (not just UI), default deny, restrictive CORS
• [ ] Misconfiguration: Debug mode off in production, default credentials changed, security headers set
• [ ] XSS: Output encoding on all user data, CSP headers, no dangerouslySetInnerHTML without sanitization
• [ ] Deserialization: No deserializing untrusted data without allowlists
• [ ] Vulnerable Components: Dependencies pinned, no known CVEs, lock files committed
• [ ] Insufficient Logging: Auth events logged, failures logged, logs don't contain secrets

Secrets Detection

Flag these patterns in ANY file (code, config, IaC, docs, tests):

Where secrets hide: .env files in git, Docker build args, Terraform tfvars, CI configs, test fixtures, comments.

Prevention: Environment variables or secret managers. Add .env, *.tfvars, *.pem to .gitignore.

Dependency Security

1. Check for known CVEs: npm audit / pip-audit / cargo audit / govulncheck
2. Verify exact version pins (not ranges) and lock files committed
3. Minimize dependency footprint -- is this package necessary?

IaC Security Quick Checks

</instructions>

<rules>

Finding Severity

Non-Negotiables

• Always run the OWASP Top 10 checklist for every code change -- no exceptions
• Flag secrets in ANY file type (code, config, IaC, docs, tests, comments)
• Never skip dependency audit when dependencies are added or updated
• Security-Critical findings must be fixed before merge -- no deferral

</rules>

<examples>

Finding Report Examples

Good Critical/Important Finding -- specific, evidenced, actionable

**[C1] SQL Injection in user search endpoint**
- **Location:** src/routes/users.py:42
- **Description:** User-supplied `q` parameter is interpolated directly into a SQL query
  via f-string: `cursor.execute(f"SELECT * FROM users WHERE name = '{request.args['q']}'")`
- **Impact:** Attacker can execute arbitrary SQL via the `q` query parameter, potentially
  exfiltrating the entire user database or escalating privileges.
- **Remediation:** Use parameterized query:
  `cursor.execute("SELECT * FROM users WHERE name = %s", (request.args['q'],))`
- **Evidence:** `cursor.execute(f"SELECT * FROM users WHERE name = '{request.args['q']}'")`

Good Advisory Finding -- bulleted, concise

- Missing rate limiting on `/api/login` (src/routes/auth.py:15) — add express-rate-limit middleware
- Debug logging enabled in production config (config/prod.yml:8) — set `debug: false`

Good Executive Summary -- plain English, prioritized

Two API endpoints accept user input directly in SQL queries, creating injection
vulnerabilities that could expose the entire user database. An API key committed
to test fixtures should be rotated immediately. The remaining findings are
low-risk code quality improvements. Fix the SQL injection first — it's the most
dangerous and affects the most-used endpoints.

Bad Finding -- vague, no evidence, not actionable

**Security Issue: Possible injection**
 
The code might have injection vulnerabilities. Consider reviewing input handling.

</examples>

Integration

Referenced by: shipyard:auditor agent (comprehensive scans), shipyard:builder (awareness during implementation)

Pairs with: shipyard:infrastructure-validation (IaC tool workflows), shipyard:shipyard-verification (security claims need evidence)