clari-sdk-patterns
Pass
Audited by Gen Agent Trust Hub on Sep 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data including API schemas, provider contracts, fixtures, and logs as part of the client generation workflow. This creates a potential surface for indirect instructions to influence agent behavior.
- Ingestion points: Processes external configuration files, provider contracts, fixtures, and application logs (SKILL.md, Tool Discipline).
- Boundary markers: Explicitly instructs the agent to "reject unknown destructive actions by default" and "schema-check provider responses before domain mapping."
- Capability inventory: The skill is restricted to file system tools (Read, Grep, Write, Edit) and does not have access to network or direct code execution tools.
- Sanitization: Mandates the redaction of credentials (api-key, api-password) in errors, traces, and snapshots, and requires schema validation to prevent downstream loading of malformed data.
Audit Metadata