refine-product-100

Pass

Audited by Gen Agent Trust Hub on Nov 11, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill implements a sophisticated 'refine' loop that uses a local state file located at .git/info/refine-product-state.md to track progress across sessions. This is a legitimate use of the file system for task persistence within a project context.
  • [SAFE]: It contains strong built-in safety instructions, explicitly defining 'safety' as operating within the requested scope, avoiding the use of secrets or personal data, and requiring local verification before any changes are finalized.
  • [SAFE]: The 'Release Intent' logic provides a security guard against accidental publishing, requiring explicit and unambiguous commands (e.g., release, deploy) to activate the distribution workflow. It specifically instructs the agent to treat ambiguous terms as non-authoritative for execution.
  • [SAFE]: Command execution capabilities (such as npm install, git push, or build tasks) are directly aligned with the primary purpose of the skill (product development and release) and are subject to the safety classification gates defined in the prompt.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 11, 2026, 04:17 AM