analyze-plan-feedback（分析计划反馈）

Feedback Analysis & Prioritization

Step 1: Parse arguments

$ARGUMENTS

Parse $ARGUMENTS for:

• Plan path (optional): The first argument that looks like a file path. Resolve to absolute and read it.
• Reviewer count (optional): A number (1-10). Defaults to 3 if not provided.

If no plan path is provided, auto-detect candidates and present an interactive selection:

1. Detect candidates (run these in parallel):

- Check if a plan file was recently created, read, or discussed in this session - Check git diff --name-only and git log --oneline -5 --diff-filter=AM -- 'docs/plans/**/*.md' for recently added/modified plans - Look for plan files/directories matching docs/plans/**/*.md

1. ALWAYS present an AskUserQuestion multiple-choice UI with up to 4 detected candidates. Never just ask a text question — always use the interactive chip UI. Example: AskUserQuestion(questions: [{question: "Which plan should I collect feedback for?", header: "Plan", options: [{label: "phase-15-transcript-import/", description: "Modified 2 mins ago — 8 files"}, {label: "phase-14-streaming.md", description: "Modified 3 days ago"}], multiSelect: false}]) The user can always select "Other" (built into AskUserQuestion) to type a custom path.
2. If zero candidates are found, still use AskUserQuestion with helpful options: options: [{label: "docs/plans/", description: "Browse the plans directory"}, {label: "Browse recent files", description: "Show recently modified.md files"}]

Examples:

• /analyze-plan-feedback docs/plans/my-plan.md → 3 reviewers
• /analyze-plan-feedback docs/plans/my-plan.md 2 → 2 reviewers
• /analyze-plan-feedback 2 → auto-detect plan, 2 reviewers
• /analyze-plan-feedback → auto-detect plan, 3 reviewers

Step 2: Collect feedback interactively

Ask the user to paste feedback one reviewer at a time, up to the reviewer count. After each paste, confirm receipt and ask for the next.

For each reviewer (1 through N):

• Ask: "Paste reviewer N feedback:"
• The user may paste text directly or provide a file path — if it looks like a path, read the file.

Do NOT proceed to analysis until all N reviewers' feedback has been collected.

Step 3: Analyze

Review and analyze all collected feedback against the plan:

1. Technical Assessment

For each piece of feedback from each reviewer:

• Is the technical point valid and accurate?
• Does it apply to our specific stack and codebase?
• Is it based on current best practices or outdated patterns?
• Cross-reference against codebase patterns where relevant

2. Priority Classification

Categorize every distinct feedback point:

3. Conflict Resolution

Where reviewers disagree:

• State each position clearly
• Evaluate which is more applicable to our context
• Make a recommendation with reasoning

4. Action Items

Create a numbered list of specific, actionable tasks:

• Each task references the feedback source(s)
• Each task has a clear definition of done
• Group by the plan section they affect

5. Implementation Timeline

For each action item, estimate:

• Effort: 1 (trivial) to 5 (significant rework)
• Risk of skipping: What happens if we don't do this?
• Dependencies: Does this block or depend on other items?

Output Format

## Feedback Summary

### Critical (must address)
- [ ] Item with effort estimate and source

### High Priority
- [ ] Item with effort estimate and source

### Medium Priority
- [ ] Item with effort estimate and source

### Low Priority / Deferred
- [ ] Item with effort estimate and source

## Recommended Next Steps (in order)
1. First thing to do
2. Second thing to do
...

## Reviewer Agreement Matrix
| Topic | Reviewer 1 | ... | Reviewer N | Verdict |

After presenting the analysis, ask if the user wants to apply any of the feedback directly to the plan.