sleuth
Pass
Audited by Gen Agent Trust Hub on Oct 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill implements a robust safety framework that requires redacting secrets and personal data before processing logs or reports.
- [SAFE]: It emphasizes minimal intervention and requires explicit user consent before making any production changes or executing expensive operations.
- [SAFE]: All external references link to official documentation from trusted technology organizations like Google, Python, and OpenTelemetry.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data such as logs and stack traces. This vulnerability surface is mitigated by instructions to redact PII and use a structured notebook to separate raw data from agent reasoning.
- [INDIRECT_PROMPT_INJECTION_EVIDENCE]: (1) Ingestion points:
SKILL.md(logs, traces, database state, error messages). (2) Boundary markers: Instructions to redact secrets and separate facts from inferences. (3) Capability inventory: Running tests, shell commands, and git operations across multiple files. (4) Sanitization: Mandatory redaction of credentials and personal data.
Audit Metadata