buildbetter-review-improve-smart-tag

Pass

Audited by Gen Agent Trust Hub on Nov 12, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from evaluation cases to drive prompt refinement, creating a surface for indirect prompt injection.
  • Ingestion points: The skill retrieves evaluation predictions and evidence via the run-smart-tag-evaluation tool as described in Step 5 of the workflow.
  • Boundary markers: Human-in-the-loop guardrails are present; the skill must ask the user to mark judgments and obtain explicit approval before spending credits on improvement iterations.
  • Capability inventory: The skill has the capability to modify prompts via improve-smart-tag and manage revisions via create-smart-tag-revision.
  • Sanitization: No automated sanitization of evidence text is specified; the security posture relies on the user's manual review of the cases before they are incorporated into the revision logic.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 12, 2026, 09:33 AM