model-tiers
Pass
Audited by Gen Agent Trust Hub on Nov 19, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: No malicious patterns, obfuscation, or unauthorized network operations were detected. The skill follows best practices by ensuring subagents are spawned with fresh contexts to prevent instruction leakage.
- [INDIRECT_PROMPT_INJECTION]: The skill identifies a potential surface for indirect prompt injection as it ingests data from local configuration files and user-supplied model names.
- Ingestion points: Local configuration (~/.agents/model-tiers.md), defaults.md, and tool schemas (rejection messages).
- Boundary markers: Absent (the skill parses text entries directly).
- Capability inventory: Reads and writes local files, spawns subagents via built-in tools or external orchestrators (Orca).
- Sanitization: Absent (relies on string matching against tool-provided schemas). The risk is assessed as safe because the configuration is user-controlled and the logic is limited to model resolution.
Audit Metadata