research-experiment-plan
Pass
Audited by Gen Agent Trust Hub on Aug 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests research-related documents such as research-brief.md, novelty-report.md, and literature reviews to generate experiment plans. This represents an attack surface where malicious instructions in processed data could influence the agent's behavior. However, the skill includes explicit instructions for falsification tests and independence requirements to maintain scientific rigor.\n
- Ingestion points: Reads from local project files including research-brief.md, novelty-report.md, literature-review/*.review.md, and REVIEW_STATE.json.\n
- Boundary markers: Uses structured markdown templates and JSON schemas for output files like claim-map.json and run-blocks.json.\n
- Capability inventory: Includes local Python scripts scripts/init_experiment_pack.py (file writing) and scripts/validate_experiment_pack.py (file reading and validation).\n
- Sanitization: Relies on structural validation of JSON and markdown files during the workflow.\n- [COMMAND_EXECUTION]: The test suite (tests/test_validate_experiment_pack.py) uses subprocess.run to execute the skill's local validation and initialization scripts. These executions are restricted to the local environment and use paths derived from the skill's internal directory structure.
Audit Metadata