The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses a potential attack surface for indirect prompt injection due to its core function of processing untrusted external research data.
Ingestion points: The system is designed to ingest full paper PDFs, abstracts, figure screenshots, and clinical data provided by users or retrieved via external identifiers (DOI/PMID).
Boundary markers: The skill uses a Mandatory Output Template to structure results, though it does not explicitly define unique delimiters or 'ignore' instructions for the raw research content itself.
Capability inventory: The skill instructions imply the use of tools for literature retrieval and the generation of structured project deliverables such as mind maps and replication starters.
Sanitization: The risk is mitigated by comprehensive behavioral rules that prohibit fabrication, mandate clear labeling of unverified sections, and strictly enforce interpretation boundaries (e.g., explicitly forbidding the characterization of association as causation).
[SAFE]: The skill demonstrates high standards of integrity and security throughout its logic.
Scientific Safeguards: The 'Minimum Viable Input' protocol prevents the agent from hallucinating findings when data is missing, and 'Overclaim Prohibitions' ensure evidence is not inflated.
No Malicious Patterns: A thorough scan of the instructions and the included reference modules (tracks.md, reporting_style.md, etc.) found no evidence of unauthorized command execution, credential harvesting, or persistence mechanisms.
Metadata Integrity: The skill's metadata and descriptions are consistent with its clinical and bioinformatics research purpose.

medical-research-literature-reader-pro