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, andEdittools. - Sanitization: The provided code snippets do not include logic for sanitizing or escaping log content before it is processed by the agent.
Audit Metadata