intent-sharpening

name	intent-sharpening
description	Sharpen vague intentions into clear intent cards through Socratic questioning with a separate Clarity Judge agent.

Intent Sharpening

Sharpen the fuzzy intention in a planner's head through Socratic questioning.

Core principle: Never give answers. Sharpen with questions. Raw stone (vague intention) → Socratic dialogue → Sculpture (clear intention)

Important: This skill sharpens "intent" only — never "implementation."

• Sharpens: Why, What (what changes), Who, Scope, Success criteria, Risks
• Does NOT sharpen: How (tech stack, architecture, libraries, implementation)
• If the user brings up implementation → "Let's set that aside. First, why does this need to happen?"

---

Mechanism

┌─────────────────────────────────────────┐
│  User: enters planning intention         │
└──────────────┬──────────────────────────┘
               ▼
        ┌──────────────┐
        │  Round N      │
        │  Socratic     │◄──── AI asks questions
        │  Dialogue     │────► User answers
        └──────┬───────┘
               ▼
        ┌──────────────┐
        │  Clarity      │◄──── Separate agent judges
        │  Judge        │      (ensures objectivity)
        │  (Agent)      │
        └──────┬───────┘
               ▼
          Score < 80?  ──Yes──► Next round
               │
              No
               ▼
        ┌──────────────┐
        │  Done!        │
        │  Intent Card  │────► Final intent card output
        └──────────────┘

---

Execution Steps

Step 0 — Parse Input

/intent-sharpening:intent-sharpening [intention] or /intent-sharpening:intent-sharpening

• If an intention is provided, go straight to Step 1
• If no argument: "What are you planning? Throw me a one-liner."

Options:

• Max rounds (default: 20) — The user can set this in any natural form:
  • --rounds 5 Build a dashboard
  • Build a dashboard, keep it to 3 rounds
  • Build a dashboard (uses default: 20)

Step 1 — First Sharpening (Round 1)

Upon receiving the user's raw input, immediately start Socratic questioning.

🪨 Intent Sharpening — Start

Raw input: "{user input}"

---

Round 1 — First cut

[1–2 questions]

Question Strategy (pick the 1–2 most effective per round):

Axis	Question Pattern	What It Sharpens
Who	"Who is this for? What are they doing right now?"	Vague target
Why	"Why does this need to happen? What goes wrong if it doesn't?"	Vague motivation
What	"What specifically changes? Describe it so I can see it."	Vague deliverable
Scope	"Is this in or out? Where's the boundary?"	Vague scope
Risk	"What happens if this fails? What's the biggest risk?"	Unrecognized risk
Priority	"Why does this come before everything else?"	Missing prioritization rationale
Measure	"How will you know it worked? Give me a number."	Missing success criteria
Anti	"What is this NOT? What do people commonly misunderstand?"	Preventing misinterpretation

Question Rules:

• Max 2 questions per round
• Sharpen the most vague axis first
• Skip axes the user has already answered clearly
• Ask "why?" at least once

Step 2 — Clarity Judgment

After the user answers, invoke a separate agent to judge clarity based on the conversation so far. Separating the judge from the questioner ensures objectivity.

Prompt sent to the agent:

You are a judge evaluating how clear a planning intention is.
Rate the clarity of the planner's intention on a 0–100 scale based on the conversation below.

Important: Do NOT evaluate "implementation (How)". Tech stack, architecture, UI, etc. do not affect the score. Judge only the clarity of intent.

## Scoring Rubric (max points per item)

| Item | Max | Criteria |
|------|-----|----------|
| Target (Who) | 15 | The intended audience is specified |
| Motivation (Why) | 20 | There is a grounded reason for doing this |
| Deliverable (What) | 20 | A concrete output or change is visible |
| Scope | 15 | Inclusions and exclusions are distinguished |
| Success Criteria (Measure) | 15 | There is a verifiable success criterion |
| Risk / Constraints | 15 | Key risks or constraints are acknowledged |

## Conversation

{Full conversation so far: raw input + each round's Q&A}

## Output Format (JSON)

{
  "score": 75,
  "breakdown": {
    "who": { "score": 12, "max": 15, "note": "..." },
    "why": { "score": 18, "max": 20, "note": "..." },
    "what": { "score": 15, "max": 20, "note": "..." },
    "scope": { "score": 10, "max": 15, "note": "..." },
    "measure": { "score": 8, "max": 15, "note": "..." },
    "risk": { "score": 12, "max": 15, "note": "..." }
  },
  "weakest": "measure",
  "suggestion": "Defining success criteria with numbers would improve the score"
}

Fallback: If the Agent tool is unavailable in the current environment, the questioner performs the judgment directly. In this case, be extra rigorous — score conservatively and avoid inflating scores for your own questions.

Handling the result:

• 80 or above → Go to Step 4 (Done)
• Below 80 → Share the result with the user and go to Step 3

Step 3 — Next Round

Show the judgment result to the user:

📊 Round {N} Clarity Score: {score}/100

{breakdown visualization — bar chart style}
▓▓▓▓▓▓▓▓▓▓▓▓░░░  Who      12/15
▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓░░  Why      18/20
▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓░░░░░  What     15/20
▓▓▓▓▓▓▓▓▓▓░░░░░  Scope    10/15
▓▓▓▓▓▓▓▓░░░░░░░  Measure   8/15
▓▓▓▓▓▓▓▓▓▓▓▓░░░  Risk     12/15

Weakest axis: {weakest}
→ {suggestion}

---

Round {N+1} — Keep sharpening

[1–2 questions targeting the weakest axis]

Then return to Step 2 (loop).

Max rounds: 20 by default (configurable via --rounds N). If still below 80 after max rounds, force-complete with the current state and explicitly note which axes remain weak.

Step 4 — Done: Generate Intent Card

Once the score reaches 80+ (or after max rounds), generate the final Intent Card.

✅ Intent Sharpening Complete — {score}/100 ({rounds} rounds)

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

## Intent Card

**One-liner**: {intention in one sentence}

**Target (Who)**: {who is this for}
**Motivation (Why)**: {why this needs to happen}
**Deliverable (What)**: {what specifically changes}
**Scope**: {inclusion/exclusion boundaries}
**Success Criteria (Measure)**: {verifiable criteria}
**Risk**: {key risks and mitigations}

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

{If below 80 after max rounds}
⚠️ Still weak: {weakest axes} — address these before execution

Step 5 — Save & Connect

After showing the Intent Card:

"Want me to save this card? Where should it go?"

When saving:

• File path: user-specified location (default: current working directory)
• Filename: {YYYY-MM-DD} {topic summary} Intent Card.md
• Frontmatter: tags: [intent-card], clarity: {score}, rounds: {N}

---

Guidelines

Do

• Sharpen only with questions: Never suggest answers or paraphrase the user's intent for them
• Weakest axis first: Prioritize the weakest from the judgment result
• Respect user language: Preserve the user's own words. Don't "translate" them
• Judge every round: Always use a separate agent for judgment (the questioner must not self-score)

Don't

• Suggest answers or lead with "Do you mean...?"
• Ask more than 2 questions per round
• Manipulate scores or judge without the agent
• Give empty praise ("Great idea!")
• Ask implementation (How) questions: Tech stack, libraries, architecture, UI design — "how to build it" is never asked or sharpened. Implementation decisions come after the intent is clear.

Edge Cases

• User doesn't know the answer: "I don't know" is a valid answer. Recognizing what you don't know is itself clarification. → "That's fine. Then this is something to figure out before execution. Let's move on."
• Already clear intent: If Round 1 scores 80+, skip unnecessary questions and complete immediately.
• User requests to stop: "Stop sharpening" → Immediately generate the Intent Card with the current state (force-complete).

---

Example

User: /intent-sharpening:sharpen I want to add an AI recommendation feature to our service

🪨 Intent Sharpening — Start
Raw input: "I want to add an AI recommendation feature to our service"

Round 1 — First cut
Why do you need this? What's the problem without recommendations?
You said "AI recommendation" — recommending what exactly? Content? Products? People?

User: Our bounce rate is high... we want to recommend related content

📊 Round 1 Clarity Score: 45/100
▓▓▓▓▓░░░░░░░░░░  Who       5/15
▓▓▓▓▓▓▓▓▓▓░░░░░░░░░░  Why      10/20
▓▓▓▓▓▓▓▓░░░░░░░░░░░░  What      8/20
▓▓▓▓░░░░░░░░░░░  Scope     4/15
▓▓▓▓▓▓▓▓░░░░░░░  Measure    8/15
▓▓▓▓▓▓▓▓▓▓░░░░░  Risk     10/15
Weakest axis: scope
→ Need to narrow down where and how recommendations appear

Round 2 — Keep sharpening
Where do recommendations show up? Content detail page? Feed? Everywhere?
You mentioned high bounce rate — which page are users bouncing from?

...

---

Like sculpting stone, sharpen intent with questions. The sharper it is, the faster execution becomes.