Skip to main content

Feedback

Feedback lets your team turn real user reactions into quality improvements.

How to get there: Agent → Feedback (when enabled)

Why Feedback matters​

Feedback helps you:

  • identify response quality gaps,
  • separate noise from true defects,
  • feed high-quality signal into memory/improvement loops,
  • close the iteration loop after deployment.

Feedback controls​

Filter bar​

FilterOptions
StatusNeeds Review / Approved / Rejected
TypeThumbs Up / Thumbs Down
RefreshReload list

feedback-filter placeholder

Feedback table​

ColumnWhat it shows
FeedbackPositive/negative reaction
Written FeedbackFree-text comment
UserUser identifier
StatusReview state
DateTimestamp
ThreadConversation context link

feedback-table placeholder

Review workflow​

  1. Filter to Needs Review
  2. Open thread context
  3. Judge whether signal is valid
  4. Approve or reject
  5. Track recurring categories for roadmap fixes

Batch operations​

ActionEffect
Approve SelectedMarks valid feedback for improvement loop
Reject SelectedExcludes feedback from improvement loop
Delete SelectedPermanently removes selected entries
note

Batch actions can be temporarily limited while memory processing is running.

Context-first decision making​

Always inspect conversation context before approving negative feedback.

A thumbs-down may be due to:

  • missing Drive context,
  • wrong Subagent routing,
  • poor user prompt clarity,
  • expected refusal under Guardrails.

feedback-context-drawer placeholder

Improvement loop​

A practical loop:

  1. Collect feedback
  2. Review and curate
  3. Convert valid signal into prompt/tool/Drive/Guardrail improvements
  4. Redeploy
  5. Re-measure feedback trend

feedback-procedure-memory placeholder

Best practices​

  • Review Feedback after each deployment
  • Prioritize repeated negative themes over one-off outliers
  • Maintain internal taxonomy for recurring issues
  • Pair feedback review with Threads review for full context