research-refine

Pass

Audited by Gen Agent Trust Hub on Jul 18, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses Bash with a quoted heredoc (cat << 'EOF') to write large files to the local file system. This is a standard workaround for platform-specific file size limitations and follows safe practices by quoting the delimiter to prevent command substitution.
  • [EXTERNAL_DOWNLOADS]: Utilizes WebSearch and WebFetch to retrieve information from the internet for grounding research proposals. This is a core functionality for research-oriented skills.
  • [PROMPT_INJECTION]: Exhibits an indirect prompt injection surface as it ingests untrusted data from local paper directories and web searches to generate research proposals. 1. Ingestion points: Files in papers/ and literature/ directories, and content retrieved via the WebFetch tool. 2. Boundary markers: Absent. External content is not wrapped in specific delimiters or preceded by instructions to the model to ignore embedded commands. 3. Capability inventory: The skill has broad capabilities including file system modification (Write, Edit, Bash) and network access (WebSearch, WebFetch). 4. Sanitization: Absent. No explicit sanitization or validation of the ingested external content is performed before it is incorporated into the prompt context.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 18, 2026, 04:52 PM
Security Audit — agent-trust-hub — research-refine