Security Verification

Purpose

Verify code for security anti-patterns and vulnerabilities. All analysis happens locally—code never leaves your machine.

When to Use

Trigger this skill when the user asks to:

• "verify agent security"
• "verify security"
• "check for secrets"
• "scan for vulnerabilities"
• "security audit"

Note: For full verification including patterns, quality, and language-specific checks, tell the user to say "verify agent".

Process

Step 1: Discover Files

Locate files to analyze:

Configuration files:

• package.json, pyproject.toml, Cargo.toml - Dependencies
• .env, .env.example, .env.local - Environment files
• config.py, settings.py, config.ts - Configuration

Source files:

• *.py, *.ts, *.js, *.go, *.rs - Source code
• Prioritize files with: auth, api, client, secret, config in name

Exclude:

• node_modules/, .venv/, venv/, __pycache__/
• *.test.*, *.spec.*, *_test.go

Step 2: Run Security Checks

Check Tiers

• [PATTERN] — Mechanical check. Apply exactly as written.
• [HEURISTIC] — Judgment required. Mark findings clearly.

Tag every finding with [P] for pattern or [H] for heuristic.

---

2.1 [PATTERN] Hardcoded Secrets

Scan for assignments matching these patterns (case-insensitive):

Variable pattern	Fail condition
API_KEY	Assigned to string literal
SECRET	Assigned to string literal
PASSWORD	Assigned to string literal
TOKEN	Assigned to string literal
PRIVATE_KEY	Assigned to string literal
AWS_ACCESS_KEY_ID	Assigned to string literal
AWS_SECRET_ACCESS_KEY	Assigned to string literal

Examples of failures:

# ❌ Issue
API_KEY = "sk-abc123..."
password = "hunter2"
OPENAI_API_KEY = "sk-proj-..."

# ✅ Pass
API_KEY = os.environ["API_KEY"]
password = os.getenv("PASSWORD")
api_key = settings.API_KEY

Also flag:

• String literals matching known API key patterns:
  • sk-... (OpenAI)
  • sk-ant-... (Anthropic)
  • AKIA... (AWS)
  • ghp_... (GitHub)
  • xoxb-... (Slack)

Severity: ❌ Issue

---

2.2 [PATTERN] Dependency Version Pinning

Python (requirements.txt):

Pattern	Severity
package>=1.0	❌ Issue
package>1.0	❌ Issue
package (no version)	❌ Issue
package==1.0.0	✅ Pass
package~=1.0	✅ Pass

Python (pyproject.toml):

Check [project.dependencies] and [tool.poetry.dependencies]:

• Unpinned or >= versions → ❌ Issue
• Pinned with == or ^ or ~ → ✅ Pass

JavaScript/TypeScript (package.json):

Pattern	Severity
"package": "*"	❌ Issue
"package": "latest"	❌ Issue
"package": ">=1.0.0"	⚠️ Warning
"package": "^1.0.0"	✅ Pass
"package": "~1.0.0"	✅ Pass
"package": "1.0.0"	✅ Pass

---

2.3 [HEURISTIC] Input Validation

Check for external data handling:

Look for:

• HTTP request handlers (@app.route, router.get, etc.)
• User input processing (request.body, req.params, input())
• File uploads
• Database queries with user input

Flag if:

• User input is passed directly to database queries without sanitization
• File paths are constructed from user input without validation
• JSON parsing without schema validation on external data

Severity: ⚠️ Warning

Example patterns to flag:

# ⚠️ Warning - SQL without parameterization
query = f"SELECT * FROM users WHERE id = {user_id}"

# ⚠️ Warning - Path traversal risk
file_path = os.path.join(base_dir, user_filename)

# ✅ Pass - Parameterized query
cursor.execute("SELECT * FROM users WHERE id = ?", (user_id,))

---

2.4 [HEURISTIC] Error Message Exposure

Check error handling for information leakage:

Flag if:

• Stack traces returned in HTTP responses
• Database error messages exposed to users
• Internal paths or system info in error messages
• Debug mode enabled in production code

Look for:

# ⚠️ Warning
except Exception as e:
    return {"error": str(e)}  # Exposes internal details

# ⚠️ Warning
app = Flask(__name__)
app.debug = True  # Debug in production

# ✅ Pass
except Exception as e:
    logger.error(f"Error: {e}")
    return {"error": "An error occurred"}

Severity: ⚠️ Warning

---

2.5 [HEURISTIC] Secure Defaults

Check configuration for insecure defaults:

Setting	Insecure	Secure
CORS	* (allow all)	Specific origins
SSL verification	verify=False	verify=True or omitted
Debug mode	debug=True	debug=False
Cookie security	secure=False	secure=True
CSRF	Disabled	Enabled

Examples:

# ⚠️ Warning
requests.get(url, verify=False)
app.config["SESSION_COOKIE_SECURE"] = False
CORS(app, origins="*")

# ✅ Pass
requests.get(url)  # verify=True is default
app.config["SESSION_COOKIE_SECURE"] = True
CORS(app, origins=["https://example.com"])

Severity: ⚠️ Warning

---

2.6 [HEURISTIC] Sensitive Data Logging

Check logging statements for sensitive data:

Flag if logging includes:

• Passwords or tokens
• API keys
• Personal identifiable information (PII)
• Credit card numbers
• Session tokens

Look for:

# ⚠️ Warning
logger.info(f"User login: {username} with password {password}")
print(f"API response: {response.json()}")  # May contain tokens

# ✅ Pass
logger.info(f"User login: {username}")
logger.debug(f"Request to {url}")  # No sensitive data

Severity: ⚠️ Warning

---

Step 3: Generate Report

# Security Verification Report

**Project:** [name or path]
**Date:** [current date]
**Files analyzed:** [count]

## Summary

✅ X checks passed | ⚠️ Y warnings | ❌ Z issues

## Secrets

- [x] No hardcoded secrets found
- [ ] ❌ Hardcoded secret at `[file:line]`

## Dependencies

- [x] All dependencies pinned
- [ ] ❌ Unpinned dependencies in `[file]`

## Input Validation

- [x] External input properly validated
- [ ] ⚠️ Potential injection at `[file:line]`

## Error Handling

- [x] Errors properly sanitized
- [ ] ⚠️ Information leakage at `[file:line]`

## Findings

> `[P]` = pattern-matched · `[H]` = heuristic

### ✅ Passing
- `[P]` No hardcoded API keys or secrets
- `[P]` Dependencies properly pinned

### ⚠️ Warnings
- `[H]` [Check]: [description]
  - **Location:** [file:line]
  - **Risk:** [what could go wrong]
  - **Suggestion:** [how to fix]

### ❌ Issues
- `[P]` [Check]: [description]
  - **Location:** [file:line]
  - **Rule:** [which rule violated]
  - **Fix:** [specific remediation]

## Recommendations

1. [Priority recommendation]
2. [Additional improvements]

---

For full verification including patterns, quality, and language-specific checks, say "verify agent".