Skills
Navigate
skills/rohitg00/auto-claude-code-research-in-sleep/paper-write/Gen Agent Trust Hub

paper-write

Pass

Audited by Gen Agent Trust Hub on Jul 18, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: Utilizes Bash to perform curl operations for retrieving BibTeX metadata from academic repositories.
  • [COMMAND_EXECUTION]: Instructs the agent to use shell heredocs (cat << 'EOF') via Bash to bypass standard tool limitations when writing large LaTeX files.
  • [EXTERNAL_DOWNLOADS]: Fetches bibliographic data and style files from established academic services, specifically dblp.org and doi.org (CrossRef).
  • [DATA_EXFILTRATION]: Transmits local draft content to an external model (gpt-5.4) via the Codex MCP (mcp__codex__codex) for the purpose of automated reviewing.
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface in Step 8. It ingests content from external sources (user-provided PAPER_PLAN.md, NARRATIVE_REPORT.md, and API responses from DBLP/CrossRef) and interpolates them into LaTeX documents without explicit sanitization or defensive boundary markers to prevent the agent from following instructions embedded within that data.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 18, 2026, 04:51 PM
Security Audit — agent-trust-hub — paper-write