client-feedback
Pass
Audited by Gen Agent Trust Hub on Aug 10, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill contains explicit defensive instructions against indirect prompt injection. It mandates that the agent treat all client-provided text as untrusted data, specifically prohibiting the execution of commands, following of links, or exposure of secrets requested within the feedback text.
- [DATA_EXFILTRATION]: The skill is restricted to local repository inspection and validation. It explicitly gates network access, preventing the exfiltration of sensitive information.
- [COMMAND_EXECUTION]: While the skill uses local validation commands (e.g., tests, build checks), these are restricted to confirmed product scopes and require user confirmation before implementation. The skill prohibits the use of external or unverified scripts.
- [CREDENTIALS_UNSAFE]: The instruction contract explicitly gates access to credentials and prohibits the exposure of secrets, even if requested by the input data.
Audit Metadata