personalization-pipeline
Pass
Audited by Gen Agent Trust Hub on Nov 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill describes a workflow that ingests external, untrusted data (job postings and raw quotes from an ATS aggregator) to be processed by a language model.
- Ingestion points: External data is ingested in the 'signal' and 'text' steps defined in the pipeline description in SKILL.md.
- Boundary markers: The skill recommends using a 'fixed record' input format and provides a 'hard constraint' in the prompt to restrict the model's output to the provided quote, which acts as a logical boundary.
- Capability inventory: The skill is purely instructional and contains no scripts, subprocess calls, or network operations.
- Sanitization: The 'Gate' section details a verification step that checks for 'template tells' and length constraints to validate the model's output before it proceeds through the pipeline.
Audit Metadata