Vague Triggers
Medium
- Confidence
- 85% confidence
- Finding
- The skill instructs the agent to run the script directly for broadly defined image-related requests and explicitly says not to ask for confirmation. In an agent with `shell:exec`, broad auto-invocation increases the chance of unintended execution on untrusted or attacker-supplied files, which expands the attack surface and can chain with any weaknesses in downstream image parsers, path handling, or shell command construction.
