sciatlas-literature-review

Pass

Audited by Gen Agent Trust Hub on Jun 23, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill downloads and installs the sciatlas package from the author's GitHub repository: github.com/zjunlp/SciAtlas.git. This is a vendor-owned resource used for the skill's primary functionality.
  • [COMMAND_EXECUTION]: Executes shell commands to install dependencies (pip install) and verify the installation environment (Get-Command or command -v).
  • [COMMAND_EXECUTION]: Permanently configures system environment variables on Windows using setx and session-based variables on Unix-like systems using export to store the API base URL and key.
  • [CREDENTIALS_UNSAFE]: Requests the user to provide an API token (sciatlas_xxx) and store it in environment variables for authenticating requests to the SciAtlas service at sciatlas.openkg.cn.
  • [DATA_EXFILTRATION]: Ingests and processes data from local files (report.md, response.json, summary.txt, request.json) generated by the search tool to construct the final literature review.
  • Ingestion points: Reads tool-generated artifacts including report.md and response.json (documented in SKILL.md).
  • Boundary markers: None present to distinguish untrusted paper metadata from instructions.
  • Capability inventory: Executes the sciatlas search-papers command and manages environment variables.
  • Sanitization: No explicit sanitization of the paper abstracts or summaries before interpolation into the final report.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 23, 2026, 12:19 PM
Security Audit — agent-trust-hub — sciatlas-literature-review