meta-abstract-screener
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted research paper titles and abstracts which are then processed by the LLM and used in subsequent tool calls.\n
- Ingestion points: Paper titles and abstracts provided by the user (SKILL.md Analysis Phase).\n
- Boundary markers: None explicitly defined in the provided instruction file to separate untrusted text from instructions.\n
- Capability inventory: Execution of the local validation script
scripts/screen_paper.pyvia shell command (SKILL.md Workflow).\n - Sanitization: No documented sanitization of the abstract content before processing or script execution.\n- [DYNAMIC_EXECUTION]: The skill documentation describes a workflow that involves interpolating agent-generated content into a shell command for validation.\n
- Evidence:
python scripts/screen_paper.py '<json_output>'in SKILL.md.
Audit Metadata