ask-lukas

Pass

Audited by Gen Agent Trust Hub on Nov 2, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes internal documentation (SKILL.md files) and an evaluation dataset (evals/evals.json) to provide routing recommendations and perform self-validation.
  • Ingestion points: The skill logic reads the contents of evals/evals.json and referenced SKILL.md files from the catalog.
  • Boundary markers: Not explicitly implemented in the provided scripts; the skill relies on model capability to distinguish evaluation data from active instructions.
  • Capability inventory: Capabilities are limited to providing text recommendations and executing local validation scripts.
  • Sanitization: The scripts/validate-evaluations.py script performs structural and type validation on input JSON data to ensure compliance with the expected schema.
  • [DYNAMIC_EXECUTION]: The test runner tests/run-validation.py uses the importlib.util module to dynamically load and execute functions from the scripts/validate-evaluations.py file. This is a standard and safe technique for Python test harnesses to maintain modularity and manage relative file paths within a project directory.
  • [COMMAND_EXECUTION]: The skill documentation includes instructions for running bundled Python scripts (tests/run-validation.py and scripts/validate-evaluations.py) to validate model decisions. These scripts are self-contained, use only Python standard library components, and the skill explicitly notes that they should only be run with explicit user or test authorization.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 2, 2026, 11:20 AM