feedback-processor

  APPROVAL:
  - "LGTM", "Looks good", "Approved", "Ship it", "Great work"
  - Action: Transition to Done

  REQUESTED_CHANGES:
  - Lists specific issues or improvements
  - Contains "but", "however", "needs", "should", "please fix"
  - Action: Extract issues, iterate

  CLARIFICATION_NEEDED:
  - Questions, confusion, requests for explanation
  - Contains "?", "what", "why", "how", "don't understand"
  - Action: Block task, request clarification
</feedback_types>

<iteration_limits>
  **Enforce iteration limits:**

  - Max feedback rounds: 5 (from AUTOPILOT_MAX_FEEDBACK_ROUNDS)
  - If limit reached: Escalate to manual intervention
</iteration_limits>

</critical_constraints>

<phase number="2" name="Classify Feedback">
  <steps>
    <step>
      Analyze comment for classification signals:

      APPROVAL signals:
      - Positive sentiment words
      - Short affirmative responses
      - Explicit approval language

      REQUESTED_CHANGES signals:
      - Numbered lists
      - "Should", "needs to", "please"
      - Specific component/file references
      - Before/after comparisons

      CLARIFICATION signals:
      - Questions
      - Confusion language
      - Request for explanation
    </step>
    <step>
      Determine confidence in classification:
      - High: Clear signals, unambiguous
      - Medium: Mixed signals
      - Low: Unclear, ask for confirmation
    </step>
  </steps>
</phase>

<phase number="3a" name="Handle Approval" condition="APPROVAL">
  <steps>
    <step>
      Transition issue to Done:
      ```bash
      bun run ${CLAUDE_PLUGIN_ROOT}/scripts/linear-client.ts transition \
        --id "$ISSUE_ID" \
        --state "Done"
      ```
    </step>
    <step>
      Add completion comment:
      ```bash
      bun run ${CLAUDE_PLUGIN_ROOT}/scripts/linear-client.ts add-comment \
        --id "$ISSUE_ID" \
        --body "Task completed and approved. Thank you!"
      ```
    </step>
    <step>Return: "APPROVED - Task marked as Done"</step>
  </steps>
</phase>

<phase number="3b" name="Handle Changes" condition="REQUESTED_CHANGES">
  <steps>
    <step>
      Check iteration count:
      - If >= MAX_FEEDBACK_ROUNDS: Escalate
      - Else: Continue
    </step>
    <step>
      Extract structured issues from comment:
      ```
      Issue 1:
        Component: {component}
        Problem: {description}
        Expected: {expected}
        Severity: {MEDIUM/HIGH}

      Issue 2:
        ...
      ```
    </step>
    <step>
      Write feedback context to SESSION_PATH/feedback/round-{n}.md:
      ```markdown
      # Feedback Round {n}

      **User Comment**: {original_comment}

      **Extracted Issues**:
      1. {issue_1}
      2. {issue_2}

      **Previous Iteration**:
      - Confidence: {prev_confidence}%
      - Issues fixed: {count}
      ```
    </step>
    <step>
      Transition back to In Progress:
      ```bash
      bun run ${CLAUDE_PLUGIN_ROOT}/scripts/linear-client.ts transition \
        --id "$ISSUE_ID" \
        --state "In Progress"
      ```
    </step>
    <step>
      Trigger re-execution:
      Return: "ITERATE - {count} issues to fix"
    </step>
  </steps>
</phase>