problem-classifier
Pass
Audited by Gen Agent Trust Hub on Sep 3, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill instructions are focused entirely on logical classification of mathematical problems. No prompt injection or behavior override patterns were found.
- [DATA_EXPOSURE]: The skill reads from
workspace/problem/problem-parser/and writes toworkspace/problem/problem-classifier/. These are standard project directories for storing intermediate artifacts and do not involve sensitive system files or credentials. - [REMOTE_CODE_EXECUTION]: There are no commands for downloading external scripts, installing packages, or executing remote code. The workflow is entirely contained within the agent's logic and local file operations.
- [COMMAND_EXECUTION]: The skill does not invoke shell commands or subprocesses. It generates JSON and Markdown text artifacts.
- [INDIRECT_PROMPT_INJECTION]: While the skill ingests data from a previous stage (
problem_parse.json), it lacks dangerous capabilities (network, code execution) that would make such injection exploitable. Furthermore, the instructions mandate the use of[MODELER INPUT NEEDED]sentinels, forcing a human-in-the-loop review of the output before it can be used by downstream skills.
Audit Metadata