The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted product descriptions and external legal research data to generate risk assessments.
Ingestion points: Processes user-provided feature descriptions and external search results from legal tools such as "北大法宝" (Baidu Law) or "威科先行" (Wolters Kluwer).
Boundary markers: Mandates the use of source attribution tags (e.g., [联网检索 — 需复核], [模型知识 — 需验证], [用户提供]) to clearly distinguish the origin and reliability of data.
Capability inventory: Writes generated risk assessment files to specific local storage paths.
Sanitization: Implements a "Prohibit Silent Supplement" policy and requires manual citation verification against authoritative legal databases before finalization.
[DATA_EXFILTRATION]: The skill accesses specific local file paths to maintain context and save results.
File paths: Reads from and writes to ~/.claude/plugins/config/claude-for-legal-zh/product-legal/ and its subdirectories, including matter.md and matter-specific folders.
Context: These operations are functionally necessary for the skill's workspace management and persistent storage of legal work products within its designated environment.

feature-risk-assessment