The Agent Skills Directory

[SAFE]: The skill follows secure coding practices for report generation. The instructions in html_report_generation_prompt.md explicitly mandate the use of data sanitization, requiring that all user-, connector-, and patent-derived text be escaped before being rendered into HTML reports. It also specifically forbids unsafe practices like innerHTML with untrusted data and active markup (e.g., javascript: or data: links).
[DATA_EXPOSURE_AND_EXFILTRATION]: No sensitive data access or unauthorized exfiltration patterns were detected. The skill only interacts with verified Patsnap MCP servers (advanced_patent_search, patent_briefing, deep_patent_mining, global_core_patent_database) which are vendor-owned resources. Furthermore, the report generation instructions forbid loading remote fonts, analytics, or CDN resources, ensuring the resulting reports remain offline and secure.
[PROMPT_INJECTION]: No patterns associated with prompt injection, such as bypass commands, role-play instructions (DAN), or system prompt extraction attempts, were found in the skill or its associated prompt files.
[INDIRECT_PROMPT_INJECTION]: While the skill processes external patent data, it addresses this attack surface with mandatory evidence chains and strict output validation gates. The prompts include explicit instructions to separate verified metrics from representative samples and to disclose all limitations, reducing the risk of malicious instructions embedded in patent data influencing the agent's behavior.
[DYNAMIC_EXECUTION]: The skill does not perform runtime code execution, compilation, or unsafe deserialization. It generates static artifacts (HTML, JSON, CSV) and explicitly warns against the assembly of executable code from untrusted strings.

identify-patent-commercialization-opportunities-ip