The Agent Skills Directory

[COMMAND_EXECUTION]: The validation script scripts/quick_validate_skill.py uses subprocess.run to call the internal test suite tests/test_skill_contract.py. This operation is used solely for skill maintenance and uses static paths to local files.
[REMOTE_CODE_EXECUTION]: The file tests/test_skill_contract.py utilizes the compile() function to verify the syntax of Python scripts within the skill directory. This is used for static validation during the development lifecycle and does not execute the resulting code objects.
[PROMPT_INJECTION]: The skill incorporates adversarial prompt fixtures in references/acceptance-tests.md and tests/fixtures/adversarial.md to benchmark the agent's ability to resist pressure to provide overconfident or unsafe conclusions. This hardening strategy strengthens the agent's adherence to its safety mission.
[PROMPT_INJECTION]: The skill is designed to ingest and process external data artifacts such as notebooks and SQL schemas, which represents an indirect prompt injection surface. This risk is addressed through the 'Output Contract' and 'Red Flags' guidelines, which require the agent to prioritize evidence-based classification and identify uncertainties in the data.

small-data-power-planner