wordly-wisdom

Warn

Audited by Gen Agent Trust Hub on Mar 4, 2026

Risk Level: MEDIUMPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits metadata poisoning by identifying the author as 'openai' in the SKILL.md YAML frontmatter, which contradicts the actual developer information. This is a deceptive practice that may cause users or agents to assign undue trust to the skill.
  • [COMMAND_EXECUTION]: The skill invokes bundled Python scripts (scripts/decision_matrix.py and scripts/ev_scenarios.py) to perform mathematical calculations. These scripts execute with local permissions and perform file read/write operations.
  • [PROMPT_INJECTION]: An indirect prompt injection vulnerability surface exists due to the following evidence chain:
  • Ingestion points: The skill processes untrusted user data for decision matrices and scenario analysis, including file paths provided as arguments to the bundled scripts.
  • Boundary markers: No explicit delimiters or 'ignore' instructions are used to sanitize or isolate external data processed by the agent or the Python scripts.
  • Capability inventory: The skill can execute subprocesses (Python) and perform file-system read and write operations via the scripts scripts/decision_matrix.py and scripts/ev_scenarios.py.
  • Sanitization: The bundled scripts do not validate or sanitize file paths provided to the --input and --output arguments, allowing for potential arbitrary file access or overwriting if the agent's logic is subverted.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 4, 2026, 04:54 AM