write-span-summary-templates
Pass
Audited by Gen Agent Trust Hub on Aug 6, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill provides documentation and patterns for using Liquid templates within the Prefactor ecosystem. No malicious patterns, obfuscation, or unauthorized access attempts were detected.
- [DATA_EXFILTRATION]: The skill instructions advise on how to summarize trace data for display. It explicitly cautions against dumping full prompts or outputs and recommends using filters like 'truncate' to limit the size of displayed data, which helps prevent accidental exposure of sensitive information in monitoring interfaces.
- [PROMPT_INJECTION]: The skill defines a surface for processing external span data. 1. Ingestion points: Span payloads (inputs and outputs) processed as described in SKILL.md. 2. Boundary markers: Liquid template tags (e.g., {{ model }}). 3. Capability inventory: Display of summaries in the Prefactor UI; the templates are rendered on the backend and not by the SDK locally. 4. Sanitization: The skill recommends using Liquid filters like 'truncate' and 'default' to limit output size and handle missing data, reducing the risk of indirect prompt injection affecting downstream agent perception.
Audit Metadata