prd-v01-user-value-articulation

Pass

Audited by Gen Agent Trust Hub on May 12, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection by ingesting and processing untrusted data from both local and remote sources.\n- Ingestion points: The agent context is populated with data from local project files (e.g., CFD-* entries, PRD.md) and external web content retrieved via research templates targeting sites like Reddit, G2, and Capterra.\n- Boundary markers: The instructions do not provide explicit delimiters or "ignore embedded instructions" warnings for the agent when reading from these untrusted sources.\n- Capability inventory: The skill is configured with powerful capabilities including Read, Write, Edit, WebSearch, and WebFetch, which could be leveraged if an injection occurs.\n- Sanitization: There is no logic for sanitizing or validating the content of the data ingested before it is interpolated into the agent's workflow.
Audit Metadata
Risk Level
SAFE
Analyzed
May 12, 2026, 09:42 PM