core-security-py-eval

Pack: core-security Default severity: BLOCKER Languages: Python Lifecycle: experimental Confidence: 0.95

What it catches

eval(...) and exec(...) calls — both interpret their string argument as Python code and are the canonical RCE sink.

Example — flagged

def parse(payload):
    return eval(payload)         # ← attacker-controlled payload = arbitrary code execution

def run(code):
    exec(code)                   # ← same risk

Example — not flagged

import json, ast

def parse_json(payload):
    return json.loads(payload)              # parse data, don't execute code

def parse_literal(payload):
    return ast.literal_eval(payload)        # rejects any non-literal — safe for ints/lists/dicts

def evaluator(formula):                     # local function named eval-like is fine
    return formula

Suggested fix

# (a) Parse data, don't execute code.
import json
data = json.loads(payload)

# (b) If you need to parse a literal Python expression:
import ast
data = ast.literal_eval(payload)

Implementation

Tree-sitter-python AST. Walks every call node whose callee dotted-path is eval or exec.

Source: `shared/packs/core-security/core-security-py-eval.ts` Fixtures: `corpus/rules/core-security-py-eval/`