self-review

Pass

Audited by Gen Agent Trust Hub on May 5, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute specific Python scripts located in the local workspace or home directory (verify_refs.py and check_xref.py) to perform automated audits of bibliography and cross-references.
  • [PROMPT_INJECTION]: The skill has a surface for indirect prompt injection because it ingests untrusted external manuscripts and has high-privilege capabilities such as file editing and command execution.
  • Ingestion points: SKILL.md (Phase 1, Step 1) describes taking manuscript input via PDF, Word, or pasted text.
  • Boundary markers: None identified; instructions do not require the agent to wrap manuscript content in delimiters or ignore instructions within that content.
  • Capability inventory: Access to tools Read, Write, Edit, Grep, and Glob, along with the ability to execute shell commands for auditing scripts.
  • Sanitization: No validation or sanitization of the manuscript content is performed before processing or applying automatic fixes.
Audit Metadata
Risk Level
SAFE
Analyzed
May 5, 2026, 12:27 PM
Security Audit — agent-trust-hub — self-review