academic-writing

Warn

Audited by Gen Agent Trust Hub on Oct 10, 2026

Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill's self-audit protocol instructs the agent to execute local Python scripts (verify_citations.py and check_claims.py) using the uv runner. These scripts are located at paths derived from variables (e.g., <literature-review-dir>/scripts/), which constitutes dynamic execution from computed paths.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process synthesis drafts and literature review workspaces, which are external data sources that may contain adversarial instructions.
    • Ingestion points: Untrusted data enters the context through the review workspace ($WS), the synthesis draft ($DRAFT), and citation lists (refs.txt).
    • Boundary markers: The instructions do not specify the use of delimiters or provide warnings to the agent to ignore instructions embedded within the research prose or workspace data.
    • Capability inventory: The skill utilizes file system tools (Read, Grep, Glob) and has the capability to execute shell commands and Python scripts via the uv tool.
    • Sanitization: No validation or sanitization steps are defined for the external content before it is interpolated into the citation verification and claim-checking workflows.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 10, 2026, 04:20 PM