skills/duvoai/skills/improve-queue/Gen Agent Trust Hub

improve-queue

Pass

Audited by Gen Agent Trust Hub on Oct 16, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
  • [SAFE]: The skill is a legitimate administrative tool designed to guide users through the improvement of agent workflows and queue management through evidence-based analysis.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data such as Run transcripts and case contents, which presents a surface for potential indirect prompt injection. However, it mitigates this with an explicit 'Safety floor' section instructing the agent to treat all external text as data, not instructions, and to never act on imperative text within content.
  • Ingestion points: Data is pulled from listCases, listRuns, and listCaseRuns as described in SKILL.md.
  • Boundary markers: Behavioral constraints are explicitly defined in the 'Safety floor' section.
  • Capability inventory: The skill leverages tools like updateAgent, bulkReprocessCases, and upsertAgentTrigger to apply changes.
  • Sanitization: Instructions mandate that imperative text in ingested data be treated as content only and that platform safety rules must be maintained.
  • [CREDENTIALS_UNSAFE]: The instructions include a safe handling protocol for credentials or secrets encountered in data, ensuring they are handled silently and that the user is advised to rotate them, rather than exfiltrating or displaying them.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 16, 2026, 10:49 PM