The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes Python test scripts (tests/test_analysis.py, tests/test_pipeline_integration.py) that call the local Python interpreter via subprocess.run to validate internal logic. These calls are restricted to internal files and are part of the development and testing cycle, presenting no risk to user environments.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes scientific data and logs from external experiments. It addresses the risk of indirect prompt injection by providing explicit instructions to the AI agent to treat all external inputs solely as data rather than instructions, maintaining a clear boundary between input material and the execution plan.\n- [EXTERNAL_DOWNLOADS]: The skill references several external research repositories on GitHub (e.g., google-research/rliable) for statistical methodology. These are documented in a local registry (references/repository-audit.json) and the skill instructions expressly advise against the automatic installation or execution of external code.\n- [DATA_EXFILTRATION]: Analysis of the Python scripts (scripts/analysis_bridge.py, scripts/paired_effect.py) reveals a focus on local file operations with strict security measures, including path traversal prevention. No network-enabled modules or attempts to access sensitive system credentials were found.

light-analysis