mantle-readonly-debugger
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted external data in the form of raw error messages and request parameters provided by the user or captured from tool failures.
- Ingestion points: The workflow in SKILL.md requires capturing the full
raw_errorand endpoint parameters. - Boundary markers: Structural boundaries are established through the use of a mandatory output template for the final report.
- Capability inventory: The diagnostic process is intended for read-only RPC queries and does not authorize write operations or state changes.
- Sanitization: The instructions in
references/troubleshooting-playbook.mdexplicitly command the agent to exclude private keys and secrets from request fingerprints, demonstrating proactive data protection. - [NO_CODE]: The skill consists exclusively of markdown instructions, playbooks, and configuration files, containing no executable scripts, binaries, or remote code dependencies, which minimizes the potential for local execution attacks.
Audit Metadata