The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources (PubMed via WebFetch and general web content via WebSearch) to synthesize evidence and generate hypotheses. This creates a vulnerability surface where instructions embedded in external research papers or websites could influence the agent's logic or output.
Ingestion points: PubMed literature via WebFetch, web search results via WebSearch, and user-provided observations.
Boundary markers: Absent. The instructions do not define delimiters or specific markers to prevent the agent from obeying instructions embedded in the fetched text.
Capability inventory: Use of Bash for executing local scripts and LaTeX compilation, and Write for file generation.
Sanitization: Absent. No filtering or validation of external content is mentioned before it is processed by the model.
[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using the Bash tool to perform task-specific operations.
Evidence: Use of python scripts/generate_schematic.py for image generation and xelatex/bibtex for document compilation.
Context: These are functional requirements of the skill's purpose (generating scientific reports) but represent an execution vector for local scripts provided within the skill's assets.

hypothesis-generation