The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external drug and patent databases, as well as user-uploaded modification documents, which are used to synthesize research reports. Maliciously crafted content in these inputs could potentially influence the agent's behavior or include malicious scripts in the generated reports.
Ingestion points: External patent and drug data retrieved from search tools and user-supplied revision documents (.md/.docx) read via files.cat (SKILL.md, Phase 5.4).
Boundary markers: No explicit delimiters or 'ignore' instructions are defined for the interpolation of external data into the agent's workflow.
Capability inventory: The skill uses files.begin_write, files.append, and files.finish_write to generate HTML reports from collected data.
Sanitization: There is no specification for sanitizing or escaping the ingested content before it is included in the generated HTML report files.
[DYNAMIC_EXECUTION]: The skill dynamically assembles HTML reports at runtime by concatenating data retrieved from external tools and user input.
Pattern: The workflow uses files.append to build HTML files in multiple chunks (SKILL.md, Phase 5.3). While this is the intended method for report generation, the lack of input validation for the appended content poses a risk of script injection if viewed in a web browser.

biotech-pipeline-landscape