atomic-decomposition

Pass

Audited by Gen Agent Trust Hub on Feb 20, 2026

Risk Level: SAFEPROMPT_INJECTIONNO_CODE
Full Analysis
  • Indirect Prompt Injection (LOW): The skill is designed to ingest and process data from external, untrusted sources such as academic papers and GitHub repositories during the research decomposition phase.
  • Ingestion points: User-provided research ideas ({idea_description}), searched paper content (LaTeX), and external code repository snippets.
  • Boundary markers: Absent. The prompts do not include instructions to treat external data as untrusted or use specific delimiters to separate data from instructions.
  • Capability inventory: The skill itself does not possess command execution or file-write capabilities; however, it encourages the extraction of code that might be executed by the user or other tools in the agent's environment.
  • Sanitization: No sanitization or verification logic is present to handle malicious instructions hidden in external research materials.
  • No Code (SAFE): The provided files consist entirely of Markdown documentation and JSON templates. No shell scripts, Python files, or other executable modules are included in the skill definition.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 20, 2026, 05:22 AM