Ralph Feedback - Parse Raw Feedback

Name: Ralph Feedback
Author: harakiro

Convert unstructured feedback into a structured format for the REVIEW phase.

Arguments

The user should provide:

feature: Feature slug being reviewed (required)
input: Raw feedback text or file path (ask if not provided)

Process

Identify the Feature
- Use the provided feature argument
- If not provided, ask: "What feature is this feedback for?"
Get Raw Feedback
- If file path provided, read the file
- If text provided, use directly
- If neither, ask user to paste feedback
Extract Issues Parse and identify distinct problems:
- Error messages → BUG
- "It should..." statements → BUG or CHANGE
- "It would be nice if..." → ENHANCEMENT
- "This doesn't work..." → BUG
- Performance complaints → BUG or ENHANCEMENT
Classify Each Issue

Type:
- BUG: Code doesn't work as specified
- CHANGE: Works but needs modification
- ENHANCEMENT: Nice-to-have, not in original spec
Severity:
- CRITICAL: Blocks usage, data loss, security
- HIGH: Major feature broken
- MEDIUM: Noticeable issues
- LOW: Minor cosmetic/UX

Create Structured File

Create .ralph/feedback/<feature>_review_<N>.md:

# Feedback: [Feature] — Review Cycle N

> **Feature:** feature-slug
> **Date:** YYYY-MM-DD
> **Source:** [QA/user/self-review]

## Summary
- Total Issues: X
- Bugs: Y
- Changes: Z
- Enhancements: W

## Issues

### Issue 1: [Title]
**Type:** BUG
**Severity:** HIGH

**Description:**
[What's wrong]

**Steps to Reproduce:**
1. Step 1
2. Step 2

**Expected:** [What should happen]
**Actual:** [What actually happens]

---

### Issue 2: [Title]
...

Determine Cycle Number
- Check existing files in .ralph/feedback/
- Increment from last cycle

Output

FEEDBACK PARSED — [feature] Review Cycle N

Issues extracted: X
- CRITICAL: count
- HIGH: count
- MEDIUM: count
- LOW: count

Output: .ralph/feedback/<feature>_review_N.md

Next: Run /ralph-review to create fix tasks

Then output <promise>COMPLETE</promise>.