ljg-paper-flow
Pass
Audited by Gen Agent Trust Hub on Apr 11, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill instructs the agent to operate in a 'NATIVE' mode, explicitly bypassing standard cognitive cycles such as OBSERVE, THINK, PLAN, and VERIFY. This overrides the agent's default operational guidelines to speed up execution.
- [PROMPT_INJECTION]: The skill exposes an indirect prompt injection surface by processing external content from paper URLs and PDF files without explicit sanitization or boundary markers. * Ingestion points: User-provided arxiv URLs, paper URLs, and PDF file paths extracted from the conversation. * Boundary markers: None identified in the workflow to delimit or ignore instructions within the ingested paper content. * Capability inventory: Spawns subagents and executes tool calls (ljg-paper, ljg-card) to process the ingested data. * Sanitization: No evidence of sanitization or validation of the ingested content before it is processed by subsequent tools.
Audit Metadata