guidewire-observability

Pass

Audited by Gen Agent Trust Hub on Mar 13, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill's design involves ingesting and analyzing untrusted application logs, creating a surface for indirect prompt injection.
  • Ingestion points: Log data is ingested through SQL analysis queries and Gosu-based context extraction in SKILL.md.
  • Boundary markers: No delimiters or instructions are present to prevent the agent from obeying instructions potentially embedded within log messages.
  • Capability inventory: The agent is granted powerful capabilities including Bash(curl:*), Write, and Edit tools.
  • Sanitization: The provided code snippets do not include logic for sanitizing or escaping log content before it is processed by the agent.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 13, 2026, 11:20 AM