The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides explicit instructions for the agent to use the Bash tool to perform technical tasks. This includes executing a Python script (scripts/generate_schematic.py) to generate scientific figures and invoking LaTeX tools (xelatex, bibtex) to compile the final hypothesis reports.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to fetch and process data from external scientific sources like PubMed and general web searches. It lacks explicit sanitization or boundary markers to differentiate retrieved research content from agent instructions, which could allow maliciously crafted text in external papers to influence the agent's synthesis or report generation.
Ingestion points: SKILL.md workflow steps involving WebFetch and WebSearch for literature gathering.
Boundary markers: None identified; retrieved content is directly used in the synthesis workflow.
Capability inventory: Bash access (python, xelatex), file writing (Write), and file reading (Read).
Sanitization: No filtering or escaping is specified for content retrieved from external URLs.
[DYNAMIC_EXECUTION]: The skill dynamically generates and compiles documents at runtime. It populates a LaTeX template (assets/hypothesis_report_template.tex) with generated content and uses a custom style file (assets/hypothesis_generation.sty) before executing the xelatex compiler on the resulting file.

hypothesis-generation