The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Python scripts (orchestrator.py) to manage data flow and image processing. It specifically instructs the agent to use files.begin_write, append, and finish_write to generate HTML reports manually, which is a standard method for report generation in this environment.
[EXTERNAL_DOWNLOADS]: The skill fetches external resources such as patent images and information from Patsnap services and web.search. These are essential for its stated purpose of patent litigation analysis. It also references chart.js via CDN for report visualizations, which is a well-known and trusted service for charting.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from web.search (news and litigation details) and patent metadata. There is a surface for indirect prompt injection if malicious instructions were embedded in patent titles or news snippets. However, the skill primarily uses this data for structural report generation and specific data field extraction (e.g., patent numbers, dates), which limits the impact.
Ingestion points: web.search results and patent.fetch metadata.
Boundary markers: The instructions mandate strict structural output (JSON/CSV/HTML) but do not explicitly detail LLM-level delimiters for data interpolation.
Capability inventory: File writing (files.write), network search (web.search), and patent database access (patent.fetch).
Sanitization: The skill involves manual string assembly for HTML, which requires careful handling of character escaping, though not explicitly detailed in the prompt logic.
[NO_CODE]: The skill provides Python scripts for orchestration and configuration, but the primary logic for data gathering and reporting is driven by the agent's instructions following a defined workflow.

litigation-risk-monitor