paper-orchestra

Pass

Audited by Gen Agent Trust Hub on May 7, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The script "scripts/check_tex_packages.py" executes the local LaTeX compiler ("pdflatex") using "subprocess.run" to verify the presence of required formatting packages. This is a standard environment check to ensure compilation will succeed later in the pipeline. \n- [COMMAND_EXECUTION]: The orchestrator instructions in "SKILL.md" direct the host agent to execute command-line tools such as "latexmk" for final document compilation and various provided validation scripts. \n- [PROMPT_INJECTION]: The skill processes untrusted research materials which could contain malicious instructions designed to influence the paper generation process. \n
  • Ingestion points: Reads "idea.md", "experimental_log.md", "template.tex", and "conference_guidelines.md" from the workspace as defined in "SKILL.md" and "references/io-contract.md". \n
  • Boundary markers: Implements an "Anti-Leakage Prompt" (Appendix D.4) found in "references/anti-leakage-prompt.md" to restrict the model to only provided inputs and prevent data leakage. \n
  • Capability inventory: Includes subprocess execution for LaTeX environment checks ("scripts/check_tex_packages.py") and instructions for shell-based compilation ("SKILL.md"). \n
  • Sanitization: "scripts/validate_inputs.py" validates structural formatting such as headers and file existence, but does not perform semantic sanitization of text content for indirect injection.
Audit Metadata
Risk Level
SAFE
Analyzed
May 7, 2026, 08:28 AM