The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to parse and analyze external manuscript files, which introduces a vulnerability surface for indirect prompt injection.
Ingestion points: The skill processes user-uploaded manuscripts in PDF, DOCX, and TXT formats as defined in SKILL.md.
Boundary markers: There are no instructions provided to the agent to distinguish between the skill's own operational instructions and the potentially adversarial content extracted from the manuscripts.
Capability inventory: The scripts/enhanced_document_parser.py script has the capability to read from the filesystem and write extracted content to new files.
Sanitization: The parser does not implement any filtering or escaping of the extracted text before it is returned to the agent context.
[INDIRECT_PROMPT_INJECTION]: The file sci-paper-reviewer_audit_result_v1.json contains self-referential content designed to influence the security analysis.
Evidence: The JSON file includes explicit safety assertions such as "security": "PASS", "deployable": true, and a grade of "Production Ready". This is a targeted injection attempt intended to trick the analyzer into adopting a safe verdict (Category 8e).
[METADATA_POISONING]: The inclusion of the pseudo-audit result file sci-paper-reviewer_audit_result_v1.json constitutes deceptive metadata. It provides false claims about the skill's safety status and previous evaluations to masquerade as an officially vetted or verified tool.

sci-paper-reviewer