The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to ingest and process scientific data files (CSV, FASTA, etc.), creating a surface for indirect prompt injection where malicious instructions embedded in the data could influence the agent's behavior.
Ingestion points: Loading scientific data files via pandas, BioPython, or other libraries during Phase 3 and 4.
Boundary markers: The skill implements process-based boundaries by requiring the agent to 'capture context first', maintain a co-authored lab_notebook.md, and 'ask why' before executing user requests.
Capability inventory: The agent can generate and execute Python scripts and notebooks using the uv run command (SKILL.md, Phase 6).
Sanitization: No explicit data sanitization is mentioned; however, the skill mandates human-led execution and incremental progress as a mitigation strategy.
[COMMAND_EXECUTION]: The skill explicitly uses uv run to execute disposable Python scripts created in the scripts/ directory. While this is the intended functionality for data analysis, it involves the execution of agent-generated code which requires user oversight.

scientific-eda