guidewire-core-workflow-b

Pass

Audited by Gen Agent Trust Hub on Mar 12, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill metadata in SKILL.md requests access to Bash(curl:*) and Bash(npm:*) tools, which allows the agent to execute shell commands for network operations and package management. This is consistent with the skill's enterprise integration purpose.
  • [EXTERNAL_DOWNLOADS]: The skill references documentation from docs.guidewire.com in SKILL.md. Guidewire is a well-known technology service, and these references are documented neutrally.
  • [PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection. 1. Ingestion points: Functions like createFNOL, addExposure, and createPayment in SKILL.md ingest untrusted data from external sources. 2. Boundary markers: The provided code snippets lack delimiters or 'ignore embedded instruction' warnings to isolate user data from the logic. 3. Capability inventory: The skill has access to file system modification (Write, Edit) and shell execution (Bash). 4. Sanitization: No input validation or escaping routines are implemented in the logic provided.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 12, 2026, 01:08 AM