The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided patent and literature records in CSV and XLSX formats, which creates a surface for indirect prompt injection. The risk is mitigated by explicit process separation and mandatory authorization requirements before external tool use. Ingestion points: Source Data sheets and CSV files processed in SKILL.md (Step 2) and scripts/inspect_labeling_input.mjs. Boundary markers: Instructions mandate separating extraction, judgment, and evidence, and require explicit confirmed gates. Capability inventory: Workbook generation via scripts/create_labeling_workbook.mjs and external technical retrieval via Patsnap MCP tools. Sanitization: Implements mandatory authorization gates and payload minimization logic as detailed in references/zhihuiya-mcp-orchestration.md.- [EXTERNAL_DOWNLOADS]: The skill documentation and orchestration guides reference Patsnap MCP services for technical evidence enrichment. Evidence: References to official Patsnap domains such as open.patsnap.com in README.md and references/zhihuiya-mcp-orchestration.md for keyword expansion and patent mining.

classify-technical-evidence-ip