The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests legal texts and enterprise data from the Yuan Dian API, which represents an indirect prompt injection surface. This is mitigated by explicit instructions in SKILL.md and references/08-selected-sources-delivery.md requiring the agent to perform a item-by-item review and verify the relevance and authenticity of each source before inclusion in reports. Ingestion points occur in yd_search.py API responses; boundary markers include the mandatory selected-sources.json verification gate; capabilities include network requests and file system writes for reports; sanitization includes filename slugification in yd_search.py.- [DATA_EXFILTRATION]: The skill performs network operations to open.chineselaw.com and ydzk.chineselaw.com to fetch legal data. These operations are the primary purpose of the skill. SKILL.md and README.md include comprehensive warnings regarding the transmission of potentially sensitive case information to the external platform and recommend data de-identification.- [COMMAND_EXECUTION]: The skill executes a local Python script scripts/yd_search.py via the scripts/yd-run wrapper. The wrapper implements security hardening by using env -i to clear the process environment of potentially malicious variables, which is a recommended safety practice.- [SAFE]: Credential management is handled through a .env file for the YD_API_KEY, which is documented as a safe practice. The skill author has also proactively removed older features like automatic updaters and dynamic code execution logic in recent versions to minimize the attack surface.

yuandian-law-search