research-refine
Pass
Audited by Gen Agent Trust Hub on May 13, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface. It ingests untrusted data from external sources and local files, which is then passed to the agent and external reviewer models without sanitization.
- Ingestion points:
$ARGUMENTS,papers/,literature/,WebSearch,WebFetch. - Boundary markers: Absent (no explicit delimiters or instructions to ignore embedded content).
- Capability inventory:
Bash,Write,Edit,Agent,mcp__codex__codex. - Sanitization: Absent (external content is interpolated directly into refinement/review prompts).
Audit Metadata