resume
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses shell commands like
git checkout,cat, andgit logto retrieve experiment context from the local environment. It also executes a local Python script,setup_experiment.py, which is expected behavior for skill-based automation. - [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external files (
config.cfg,program.md,results.tsv) that record the history of an experiment. This creates a surface where previously generated content is fed back into the agent's context. - Ingestion points: Reads from the
.autoresearch/directory within the repository. - Boundary markers: No specific delimiters are used when reading these files into the context.
- Capability inventory: The skill can execute git commands, read files, and run a local Python script.
- Sanitization: The skill instructions do not specify any validation or sanitization for the content read from the experiment files.
Audit Metadata