literature-review

Pass

Audited by Gen Agent Trust Hub on Nov 2, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill instructions include a fallback setup method using curl | sh to install the uv developer tool from its official domain astral.sh. While piped shell execution is a high-risk pattern, the source is the official domain of a recognized and reputable technology provider.
  • [COMMAND_EXECUTION]: Python scripts within the skill utilize subprocess.run to execute git commands for maintaining research provenance and uv run for internal tool invocation. These operations use controlled, static argument lists and are not exposed to unvalidated user input injection into shell commands.
  • [EXTERNAL_DOWNLOADS]: The skill performs network operations to retrieve scholarly works and bibliographic metadata from established and well-known academic services such as OpenAlex, arXiv, Europe PMC, Crossref, and DataCite.
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface through read_paper.py and import_library.py, which process untrusted academic publications. It manages this risk through an extensive automated verification framework: it cryptographically hashes all research artifacts, performs verbatim quote matching via check_claims.py, and cross-references identifiers against external academic registries. This defense-in-depth system ensures that all synthesized claims in the final output are traceable and grounded in verified source text, preventing the agent from being influenced by instructions potentially embedded in external papers.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 2, 2026, 08:38 AM