The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted text from user-provided documents (e.g., weekly reports, meeting minutes, technical schemes) to identify innovation points. Ingestion points: Documentation processing logic in SKILL.md. Boundary markers: The instructions lack specific delimiters or boundary markers to separate user-provided text from agent instructions, which could allow instructions embedded in reports to influence agent behavior. Capability inventory: The skill has capabilities to read local files, execute external patent search tools (patsnap_search, patsnap_fetch), and write HTML reports to the session directory. Sanitization: There is no evidence of sanitization or escaping of the user-derived content before it is rendered into the final HTML template, creating a potential vector for report manipulation.

innovation-radar