The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process large amounts of untrusted data from external scientific sources, creating a surface for indirect instructions to influence agent behavior.
Ingestion points: Data enters the agent's context through literature searches using WebFetch (PubMed) and WebSearch (general scientific web), as detailed in SKILL.md and references/literature_search_strategies.md.
Boundary markers: The instructions do not define explicit delimiters or warnings to treat ingested literature content as data rather than instructions.
Capability inventory: The skill has access to the Bash tool for LaTeX compilation and local script execution, and Write/Edit tools for file system modifications.
Sanitization: There are no mentioned filters, sanitization logic, or validation steps for the content retrieved from external scientific papers.
[EXTERNAL_DOWNLOADS]: The assets/FORMATTING_GUIDE.md file provides instructions for installing missing LaTeX dependencies using the tlmgr package manager (e.g., tlmgr install tcolorbox xcolor). This involves fetching and installing code from the external TeX Live repository system.
[COMMAND_EXECUTION]: The skill uses the Bash tool to perform routine operational tasks, including compiling LaTeX documents with xelatex and bibtex, and executing a local Python script for diagram generation (python scripts/generate_schematic.py).
[DYNAMIC_EXECUTION]: The workflow involves generating structured LaTeX source code based on experimental observations and external evidence, which is subsequently compiled into a binary PDF format via shell execution. This script-generation-and-execution pattern is necessary for the skill's purpose but represents a dynamic execution flow.

hypothesis-generation