auto-paper-improvement-loop

Pass

Audited by Gen Agent Trust Hub on May 18, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses shell commands to manage the paper improvement process. It concatenates LaTeX source files, runs latexmk for compilation, and executes an inline Python script to verify consistency between main-body theorems and appendix restatements. It also reads configuration from ~/.claude/feishu.json to send notifications.
  • [PROMPT_INJECTION]: The skill processes untrusted LaTeX source code which is interpolated into prompts for the reviewer LLM, creating a surface for indirect prompt injection.
  • Ingestion points: LaTeX files from the paper/sections/ directory are read and passed to the mcp__codex__codex tool during the review phases (Steps 2 and 5).
  • Boundary markers: The prompt structure uses clear markdown headers (e.g., ## Paper Files:) to separate the paper content from the senior reviewer instructions.
  • Capability inventory: The skill possesses Bash, Write, and Edit tools, allowing it to modify the filesystem and recompile documents based on LLM-generated fixes.
  • Sanitization: No explicit sanitization or escaping of the LaTeX source content is performed before it is included in the LLM prompt.
Audit Metadata
Risk Level
SAFE
Analyzed
May 18, 2026, 06:06 PM
Security Audit — agent-trust-hub — auto-paper-improvement-loop