detection-engineering-coverage-evaluation
Pass
Audited by Gen Agent Trust Hub on Aug 10, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- Indirect Prompt Injection Surface: The skill is designed to process external content from URLs and blogs, which represents an entry point for untrusted data into the agent's context. The skill incorporates a proactive mitigation strategy by instructing the agent to scan for specific injection patterns (e.g., 'ignore previous instructions') before processing content.
- Ingestion points:
SKILL.mdStep 1 (URL content extraction and raw text input). - Boundary markers: Step 1.3 defines specific patterns to detect and provides instructions to halt execution if they are found.
- Capability inventory: The skill has the ability to generate new security logic via
generate_rulesand deploy it to a production environment viacreate_rule, though this requires an explicit user approval step. - Sanitization: Step 1.1 and 1.4 include procedures to strip HTML elements (script, style) and UI boilerplate to reduce the risk of malicious payload inclusion.
Audit Metadata