paper-submission

Pass

Audited by Gen Agent Trust Hub on Jun 22, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the processing of untrusted academic papers.
  • Ingestion points: The agent reads user-provided PDF or LaTeX files in SKILL.md (Phase 1) using the Read tool.
  • Boundary markers: The workflow lacks clear delimiters or instructions to the model to ignore potential commands embedded within the paper text.
  • Capability inventory: The agent possesses powerful capabilities including WebSearch, WebFetch, Bash, and Write tools.
  • Sanitization: There is no evidence of filtering or escaping extracted text before it is interpolated into search queries or passed to the report generation scripts.
  • [COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute local Python scripts (extract_journals.py and generate_report.py). While these scripts are bundled with the skill, this pattern allows the model to interact with the underlying operating system and process local data assets (like the ABS journal list and system fonts) via shell commands.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 22, 2026, 03:11 AM
Security Audit — agent-trust-hub — paper-submission