wdm0006/python-skills/security-audit

# Run all four scanners (gates CI): exit 1 on a blocking finding, exit 2 when a
# requested scanner could not run, exit 0 only when every scanner ran clean:
uv run python scripts/security_scan.py .
 
# Tolerate scanners that are missing or hung (restores exit 0 in that case):
uv run python scripts/security_scan.py . --allow-scanner-failure
 
# Or individually:
uvx bandit -r src/ -ll                       # High-severity static analysis
uvx pip-audit .                              # This project's dependencies
uvx semgrep --config auto src/               # Pattern-based SAST
uvx detect-secrets scan > .secrets.baseline  # Secrets detection

exclude_dirs: [tests/, docs/, .venv/]
skips: [B101]  # assert_used - OK in tests

uvx pip-audit -r requirements.txt     # Scan requirements
uvx pip-audit --fix                   # Auto-fix vulnerabilities

# SQL - Parameterized query
conn.execute("SELECT * FROM users WHERE id = ?", (user_id,))
# Commands - No shell
subprocess.run(["cat", filename], check=True)
# Secrets - Environment
API_KEY = os.environ.get("API_KEY")
# Paths - Validate
base = Path("/data").resolve()
file_path = (base / filename).resolve()
if not file_path.is_relative_to(base):
    raise ValueError("Invalid path")

from rich.traceback import install
install(show_locals=False)

import sys
def test_unhandled_traceback_does_not_expose_locals(capsys):
    sentinel = "sentinel-secret-that-must-not-appear"
    try:
        raise RuntimeError("boom")
    except RuntimeError:
        exc_type, exc, traceback = sys.exc_info()
        sys.excepthook(exc_type, exc, traceback)
    output = capsys.readouterr()
    rendered = output.out + output.err
    assert "RuntimeError: boom" in rendered  # proves the hook rendered
    assert sentinel not in rendered

# .github/workflows/security.yml — full workflow in CI_SECURITY.md
- uses: astral-sh/setup-uv@v5
- run: uv run python scripts/security_scan.py . --output security-report.json

Code:
- [ ] No SQL injection (parameterized queries)
- [ ] No command injection (no shell=True)
- [ ] No hardcoded secrets
- [ ] Exception and logging configuration cannot render frame locals containing secrets
- [ ] No weak crypto (MD5/SHA1)
- [ ] Input validation on external data
- [ ] Path traversal prevention
- [ ] SSRF fetches connect to the validated IP on every redirect hop (DNS rebinding safe)
- [ ] SSRF deny policy covers IPv4/IPv6 and CGNAT (`100.64.0.0/10`)
 
Dependencies:
- [ ] pip-audit clean
- [ ] Minimal dependencies
- [ ] From trusted sources
 
CI:
- [ ] Security scan on every PR
- [ ] Weekly dependency scan