aminer-deep-search
Pass
Audited by Gen Agent Trust Hub on Nov 15, 2026
Risk Level: SAFE
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill requires an
AMINER_API_KEYto function but includes explicit instructions to never print or log the key. The providedaminer_api.pyscript implements a_redactfunction that automatically removes the API token from error messages and diagnostic output to prevent accidental exposure. - [EXTERNAL_DOWNLOADS]: Network operations are restricted to official AMiner domains (
datacenter.aminer.cnandwww.aminer.cn) for API interaction and paper metadata retrieval. These are verified vendor-owned resources directly related to the skill's purpose. - [UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The bundled scripts (
aminer_api.pyandpaper_set.py) use only Python standard library modules. The skill does not perform any dynamic package installation or remote script execution, maintaining a minimal and verifiable attack surface. - [INDIRECT_PROMPT_INJECTION]: The skill processes academic data retrieved from external API calls. Potential risks are mitigated by a multi-layered approach:
- Ingestion points: Academic paper metadata (titles, abstracts) ingested via
scripts/aminer_api.py. - Boundary markers: The agent is instructed to perform manual semantic relevance checks and follow a strict round-based protocol.
- Capability inventory: The agent's capabilities are limited to executing local Python scripts for network search and local JSON state management.
- Sanitization: The
scripts/paper_set.pyutility mechanically enforces structural constraints (year ranges and required fields) to ensure processed data remains within the user's defined scope.
Audit Metadata