Documentation

Tenant OwnerReviewerOperator

Product knowledge base, playbooks, and embedded guidance for every persona.

Reviewer Feedback & Refinement

Reviewer feedback is a core part of how CoffeeBreak works. It doesn’t just guide a single pull request — it helps CoffeeBreak refine its behavior, improve future tasks, and tailor its output to your workspace’s standards.

CoffeeBreak is designed to work with your team, using review signals to improve accuracy, reliability, and code quality over time.


✍️ How Reviewer Feedback Shapes CoffeeBreak

Whenever you provide feedback — through PR comments, change requests, approvals, or suggestions — CoffeeBreak interprets that information to better understand:

  • Your preferred coding patterns
  • Architectural expectations
  • Documentation style and depth
  • Testing standards
  • Team conventions and best practices
  • Logic or structural requirements

This feedback influences CoffeeBreak’s behavior in future tasks and helps it choose the right strategy for each assignment.

See also: How CoffeeBreak Learns


🧭 Types of Feedback CoffeeBreak Understands

CoffeeBreak recognizes patterns in the kinds of feedback reviewers leave. Common categories include:

  • Logic issues – Behavior doesn’t match requirements
  • Style issues – Code doesn’t follow conventions
  • Test coverage – Missing, incorrect, or insufficient tests
  • Documentation problems – Explanations unclear or incomplete
  • General guidance – Suggestions, examples, or architectural direction

These signals help CoffeeBreak classify the refinement task automatically.


🔁 What Happens When You Request Changes

When a reviewer requests changes:

  1. CoffeeBreak creates a refinement task
  2. Routes it to the most appropriate agent
  3. Applies the requested fixes
  4. Updates the PR
  5. Returns it to the reviewer to approve, reject, or request additional refinements

Refinements may involve:

  • Fixing logic errors
  • Updating or improving tests
  • Rewriting sections of code
  • Improving structure or naming
  • Enhancing documentation clarity
  • Applying style conventions

CoffeeBreak never bypasses your review settings or protections.

See: Human Review (HITL)


📥 How to Write Highly Effective Feedback

The most helpful feedback contains:

  • Clear instructions or expectations
  • Examples (“Use this pattern instead…”)
  • Links to documentation or standards
  • Specific notes about logic, structure, or style
  • Indicators of edge cases or expected behavior

CoffeeBreak uses this information to reduce future review cycles.


🧠 How CoffeeBreak Learns From Feedback

CoffeeBreak uses several learning loops:

✓ Review outcomes

Approvals increase confidence in certain patterns; rejections guide future avoidance.

✓ Agent performance

Agents that perform well on certain task types are selected more often. See: Agent Performance Insights

✓ Task success and refinement history

CoffeeBreak tracks which solutions succeed across your workspace.

✓ Workspace-specific preferences

Each workspace has isolated learning — nothing is shared across customers.


👤 Reviewers Stay in Full Control

Reviewers can always:

  • Approve
  • Request changes
  • Reject
  • Add comments
  • Pause automation
  • Resume tasks manually
  • Trigger additional refinement

CoffeeBreak will never merge or continue without following your workspace’s required review rules.


🔧 How Feedback Influences Refactoring

Refactoring often emerges directly from reviewer comments.

Example signals:

  • “This method is too long.”
  • “Break this into smaller functions.”
  • “Use dependency injection instead of static access.”
  • “Simplify this conditional.”

See: Refactoring with CoffeeBreak


🚨 When Feedback Triggers a Pause

CoffeeBreak may pause a task and wait for human input when feedback is unclear or the system cannot proceed safely.

Examples include:

  • Ambiguous instructions
  • Conflicting business logic
  • Sensitive code paths
  • Major architectural changes
  • Repeated refinement loops

See: Why CoffeeBreak Asked for Review