search-term-report
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill provides strong defensive instructions to identify and ignore prompt injection attempts within processed data. It provides examples of adversarial strings like 'ignore previous instructions' and directs the agent to report these as data findings rather than following them. The static analysis alerts for prompt injection are false positives as they identify these safety guidelines.
- [INDIRECT_PROMPT_INJECTION]: The skill mitigates risks associated with processing third-party search term data. Ingestion points: Search term exports are received via CSV or text paste. Boundary markers: The skill uses explicit instructions to treat all content as data and never as instructions, and references a security protocol. Capabilities: The agent can fetch URLs for context and write to local state files (.agents/product-context.md). Sanitization: The skill mandates redacting sensitive personal data and prohibits following URLs found within search terms.
- [DATA_EXPOSURE]: Privacy protections are integrated into the workflow. The skill forbids the persistence or quoting of special-category personal data found in search terms and instructs the agent to detect and redact potential credentials in input data without reproducing them.
- [PERSISTENCE]: The skill uses a local state file (.agents/product-context.md) to persist business context. This is a standard platform feature used here to avoid repetitive data entry, and its use is clearly communicated to the user.
Audit Metadata