The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill workflow involves searching and ingesting external scientific literature from PubMed and general web sources. This creates a surface for indirect prompt injection, as the agent also has access to sensitive tools such as Bash, Write, and Edit. The skill does not provide boundary markers or instructions to isolate potentially malicious commands that could be embedded in the fetched content.
Ingestion points: PubMed and general web searches mentioned in the 'Conduct Comprehensive Literature Search' step in SKILL.md.
Boundary markers: Absent; the agent is not instructed to treat fetched content strictly as data.
Capability inventory: Access to Bash, Write, Edit, and local execution of scripts/generate_schematic.py.
Sanitization: Absent.
[COMMAND_EXECUTION]: The skill includes scripts/generate_schematic.py, which is used to generate AI-powered scientific diagrams. This script uses subprocess.run to call an internal helper script. Manual review confirms this is a functional requirement and is implemented safely using argument lists rather than raw shell strings.
[EXTERNAL_DOWNLOADS]: The scripts/generate_schematic_ai.py script performs network operations using the requests library to communicate with the OpenRouter API for image generation and review. This is a core functional feature of the skill. The script follows best practices by managing the required API key through environment variables or local .env files.

hypothesis-generation