eva-learn
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content and maintains the capability to write data to the local file system. \n
- Ingestion points: As described in
SKILL.md, the agent reads user-provided files, pasted third-party content, screenshots, and tables during the learning process.\n - Capability inventory: The skill performs file system write operations to archive learning history (e.g.,
00-学习进度.md,07-学习问答原稿.md) in~/Documents/eva-learn/or specific user project directories. It also creates organizational structures includingsources/INDEX.md.\n - Boundary markers: The skill includes logic to conditionally load a safety protocol (
../eva-shared/references/shared/06_external-material-safety_外部材料安全边界.md) specifically when processing external materials to mitigate risks from embedded instructions.\n - Sanitization: The skill's instructions mandate adherence to the external material safety boundary when handling user-provided data, providing a layer of defense against injection attacks.
Audit Metadata