cuioss/python-security

Python Security

REFERENCE MODE: This skill provides reference material for Python security review and hardening. Load specific standards on-demand based on current task. Do not load all standards at once.

Enforcement

Execution mode: Reference library; load standards on-demand for Python security review and hardening tasks.

Prohibited actions:

• Do not pass untrusted input to subprocess with shell=True; use an argv list
• Do not pickle.load/pickle.loads untrusted bytes, nor yaml.load without a safe loader
• Do not eval/exec/compile externally-sourced strings; use ast.literal_eval for literals
• Do not interpolate user values into SQL query strings; use DB-API placeholders with a params tuple

Constraints:

• Every externally-sourced value (request data, file contents, environment, CLI args) is untrusted at stdlib boundaries
• User-supplied paths must be validated against a safe base to prevent traversal
• Reject untrusted input that fails a boundary check; never coerce it through

When to Use This Skill

Activate when:

• Running subprocesses — building argv, avoiding shell=True, quoting when a shell is unavoidable
• Deserializing external data — choosing yaml.safe_load/json over pickle/yaml.load
• Dynamic evaluation — choosing ast.literal_eval over eval/exec
• Building SQL — DB-API placeholders over string interpolation
• Handling user-supplied paths — traversal prevention with Path.resolve() / is_relative_to

Available Standards

Load progressively based on current task. Never load all standards at once.

Injection Sinks and Path Security

Read: ../python-core/standards/python-core.md

Load the Security (path traversal) and Injection and Unsafe Deserialization (subprocess / pickle / yaml / eval / SQL) sections. These cover the secure-by-default sink choices at each stdlib boundary:

| Sink | Insecure | Secure | |------|----------|--------| | Subprocess | shell=True with interpolation | argv list (no shell), or shlex.quote when shell is unavoidable | | Deserialization | pickle.loads, yaml.load | yaml.safe_load, json.loads | | Dynamic execution | eval/exec/compile | ast.literal_eval | | SQL | f-string interpolation | DB-API placeholder + params tuple | | Path handling | unchecked user path | Path.resolve().is_relative_to(safe_base) |

Surface Boundaries

| Surface | Home | |---------|------| | Injection sinks + path-traversal prevention | ../python-core/standards/python-core.md (Security / Injection sections) | | Cross-cutting OWASP / STRIDE / trust-boundary / secure-design foundations | Skill: plan-marshall:persona-security-expert |

Cross-Cutting Foundations (delegated upward)

This skill is a thin pointer: the Python sink mechanics above are genuinely language-specific (how to apply the secure choice at each stdlib boundary), but the conceptual why lives in the centralized plan-marshall:persona-security-expert sub-documents. Load the matching foundation, then return here for the Python mechanics — there is no content duplication:

| Python mechanic (here) | Centralized foundation (there) | |------------------------|-------------------------------| | Why every stdlib boundary value is untrusted; allow-list, canonicalize-before-validate, fail-closed (the general architecture behind every sink and path check) | input-validation-trust-boundaries.md | | The subprocess / SQL / eval sinks mapped to a recognized risk category (A03 Injection) | owasp-top-ten.md | | Why deserializing untrusted bytes is dangerous before any value inspection (the pickle/yaml.load avoidance is the Python mechanic) | input-validation-trust-boundaries.md — Unsafe Deserialization sub-section | | The principles behind secure-by-default sink choices (secure by default, fail securely, economy of mechanism) | secure-design-principles.md |

Note the deserialization sinks (pickle/yaml.load) also map to OWASP A08 Software and Data Integrity Failures — see owasp-top-ten.md.

Related Skills

• plan-marshall:persona-security-expert — Cross-cutting security review identity and authoritative home for OWASP Top 10, STRIDE, secrets, secure logging, trust boundaries, authn/authz, and secure-design principles
• pm-dev-python:python-core — Core Python development standards (the security/injection sections referenced above live under its standards/ directory)
• pm-dev-python:pytest-testing — Property-based / adversarial testing of security-relevant code