The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: Research automation pipelines in data-collection-automation/SKILL.md and research-workflow-automation/SKILL.md ingest data from external URLs using pd.read_csv and curl. This creates a surface for indirect prompt injection if the ingested content is subsequently processed by the agent without validation. Evidence Ingestion points: research_etl_pipeline function and Snakemake download_data rule. Boundary markers: Absent. Capability inventory: Data transformation, local file writing, and shell execution. Sanitization: No sanitization of external data was detected.
[EXTERNAL_DOWNLOADS]: The skills instruct the agent to download or clone research tools from external repositories, including Microsoft's RD-Agent and Sakana AI's AI-Scientist-v2. These are legitimate project-specific dependencies for the research discovery use case.
[COMMAND_EXECUTION]: The skill includes instructions for managing data pipelines via snakemake and make, and mentions curl for automated data retrieval. These commands are typical for research automation workflows but involve interacting with the shell and network.

automation-skills