prompt-retro

name	prompt-retro
description	Review the day's prompts for effectiveness using the 6-criteria rubric. Scores work-initiating prompts, surfaces patterns, and suggests rewrites. Use when asked for prompt review, prompt retro, or prompt quality check.
argument-hint	["YYYY-MM-DD"]
allowed-tools	Read, Glob, Write, Edit, Bash(date:*), Bash(wc:*), Bash(ls:*)

Prompt Retro

Review today's prompts for effectiveness. Score work-initiating prompts against the 6-criteria rubric. Surface patterns and suggest rewrites for the weakest prompts.

Context

Arguments: $ARGUMENTS

If arguments contain a date (YYYY-MM-DD), review that day's log. If empty, review today's log.

Rubric reference: ~/.me/prompting-guide.md

Definitions

Work-initiating prompt: A prompt that starts a new task, requests a change, or asks for analysis. These are what the rubric applies to.

Conversational prompt: A short reply, confirmation, clarification, or follow-up. Examples: "yes", "do it", "looks good", "the second one", "what about X?". These are excluded from scoring — they're valid as-is.

Threshold: A prompt under 15 words that doesn't contain a file path, function name, or technical reference is likely conversational. Don't score it, but do count it.

The 6 Criteria

Score each work-initiating prompt 0 or 1 on:

1. Context — Does it reference specific files, functions, or systems?
2. Intent — Is the why clear, not just the what?
3. Scope — Is it bounded? Could the agent know when it's done?
4. Acceptance Criteria — Would you know if the result was wrong?
5. Behavioral Instructions — Does it say how to approach the work?
6. Minimalism — Right altitude? High signal-to-noise?

A score of 5-6 is strong, 3-4 is decent, 0-2 needs a rewrite.

Workflow

Phase 1: Load Data

1. Get the target date: parse $ARGUMENTS or use date +"%Y-%m-%d".
2. Read the prompt log: ~/.me/prompts/log/YYYY-MM-DD.md.
3. Read the rubric: ~/.me/prompting-guide.md (for reference, don't quote it back).
4. If the log doesn't exist, say so and stop.

Phase 2: Classify Prompts

Read through all logged prompts. Classify each as work-initiating or conversational.

Phase 3: Score Work-Initiating Prompts

For each work-initiating prompt, score it against the 6 criteria. Be honest — the point is to surface weaknesses, not validate.

Phase 4: Write the Report

Write to: ~/.me/prompts/retros/YYYY-MM-DD.md

Create directories with mkdir -p as needed.

Structure:

# Prompt Retro — YYYY-MM-DD

**Total prompts:** N | **Work-initiating:** N | **Conversational:** N
**Average score:** X.X / 6

## Summary

[2-3 sentences: overall quality, dominant pattern, one thing to change tomorrow]

## Weakest Prompts

[For each prompt scoring 0-3, show:]

### HH:MM:SS — Score: N/6

> [The original prompt, truncated to ~100 chars if long]

| Criterion | Score | Note |
|-----------|-------|------|
| Context | 0/1 | [Why it missed] |
| Intent | 0/1 | ... |
| ... | | |

**Suggested rewrite:**

> [A concrete rewrite that would score 5-6]

## Strongest Prompts

[List 1-3 prompts scoring 5-6 with a one-line note on what made them good.
If none scored that high, say so.]

## Patterns

[Recurring weaknesses across prompts. Examples:]
- "You rarely specify acceptance criteria — 80% of prompts scored 0 on this."
- "Scope is your strongest criterion — you naturally bound your requests."

## Trend

[If previous retros exist in ~/.me/prompts/retros/, compare today's average
to the last 3 retros. Show direction: improving, flat, or declining.
If no history, skip this section.]

---

*Generated by `/prompt-retro`. Rubric: ~/.me/prompting-guide.md*

Phase 5: Print Summary to Chat

After writing the file, print a short summary:

Prompt retro saved to ~/.me/prompts/retros/YYYY-MM-DD.md
Total: N | Scored: N | Avg: X.X/6 | Weakest: N/6 | Strongest: N/6
Top pattern: [one-line dominant weakness]

Tone

Be direct and constructive. Don't soften findings. The user asked for this feedback — deliver it clearly.