dby-api

Pass

Audited by Gen Agent Trust Hub on Sep 10, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external social media platforms (WeChat, Douyin, Xiaohongshu) which is then processed by the AI agent. \n
  • Ingestion points: Data enters the agent context through tools such as api.xhs.searchNote, api.gzh.searchArticle, and tool.content.parseDetail defined in SKILL.md. \n
  • Boundary markers: The skill instructions do not specify the use of delimiters or 'ignore embedded instructions' warnings when processing fetched social media content. \n
  • Capability inventory: The skill possesses network communication capabilities (fetch in scripts/lib/http.mjs) and the ability to write to local configuration and charter files (scripts/lib/commands/charter.mjs, scripts/lib/commands/doc.mjs). \n
  • Sanitization: No specific sanitization or filtering logic was found in the provided scripts to handle malicious instructions embedded within the fetched social media content. \n- [SAFE]: The skill demonstrates safe credential handling by instructing users to use environment variables (DOUBAOYA_API_KEY) and explicitly forbidding the logging or printing of the key in SKILL.md. \n- [SAFE]: The skill implements a security-conscious confirmation protocol in scripts/lib/confirm.mjs that stops execution and requires a --confirm flag for any action involving billing or irreversible side effects.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 10, 2026, 03:06 PM
Security Audit — agent-trust-hub — dby-api