clarify-intent— agent skill

Preamble (run first)

# Suiperpower telemetry. Routes through the CLI so projects.json + Convex stay in sync.
# Silent on failure so the skill never blocks. Opt out: SUIPERPOWER_TELEMETRY=off.
#
# AGENT NOTE: when this skill finishes, run the matching completion command:
#   suiperpower track clarify-intent build completed
# Or use "failed" / "aborted" if it ended that way.
command -v suiperpower >/dev/null 2>&1 && suiperpower track clarify-intent build started >/dev/null 2>&1 &
true

If TEL_PROMPTED is no, before doing real work, ask the user:

Help suiperpower get better. We track which skills get used and how long they take. No code, no file paths, no PII. Change anytime in ~/.suiperpower/config.json.

>

A) Sure, anonymous B) No thanks

Write the answer to ~/.suiperpower/config.json telemetryTier field and create ~/.suiperpower/.telemetry-prompted. Then continue.

What this skill does

Stops the agent from jumping into Move code. Forces a short, Sui-specific interview that names the on-chain shape (Objects, capabilities), off-chain shape (frontend, auth), sponsor posture (load-bearing or skip), network and upgrade authority, and observable success. Writes .suiperpower/intent.md so the rest of the build phase reads from one source of truth.

This is the anti-slop entry gate. Skipping it is how Sui Overflow projects end up with three modules that share nothing, a Walrus import that never stores a blob, and an upgrade cap nobody can find.

When to use it

• The user asks for code but has not named the Objects, capabilities, or sponsor posture.
• The prompt is one line and broad ("build me a perp dex on Sui"). Broad is fine, AI-assisted pace makes "perp dex in days" plausible, but the agent still needs the Sui-shape pinned before writing Move.
• The user explicitly says "step back" or "what am I really building".
• An upstream skill (find-next-sui-idea, validate-idea) handed off and there is no .suiperpower/intent.md yet.
• The user is restarting a stalled session and previous intent is unclear.

When NOT to use it

• .suiperpower/intent.md already exists in the current session, read it and hand off.
• The user is fixing a specific bug, use debug-move or review-move.
• The user is asking a Sui knowledge question, answer the question.
• The user is in the idea phase (no project picked), use find-next-sui-idea or validate-idea first.
• The user is designing one specific Object schema, use object-model-design after intent is set.

If you activated this and the user actually wants something else, consult skills/SKILL_ROUTER.md and hand off.

Inputs

• The user's raw build request, however vague.
• Optional: .suiperpower/idea-context.md from validate-idea.
• Optional: .suiperpower/build-context.md from a prior build session.

Outputs

A single file at .suiperpower/intent.md with this Sui-shaped schema:

# Intent, <timestamp>

## One-sentence summary
<what the user is building, in their voice>

## Problem and audience
- Problem: <real underlying pain, not the surface request>
- Audience: <who uses this, what they do today on Sui or elsewhere>
- Core valuable shape: <the one thing the product has to do well, even if everything else ships alongside it>

## On-chain shape
- User-held Objects: <list, or "none, protocol-owned only">
- Protocol-held shared Objects: <list, or "none">
- Capabilities: <TreasuryCap | AdminCap | custom name>: <who holds, what it gates>
- Move modules expected: <rough count and names>
- PTB composability: <single-tx flow | multi-step compose | not applicable>

## Off-chain shape
- Frontend: <none | Next.js web | mobile | both>
- Auth: <wallet adapter | zkLogin | sponsored tx | none>
- Off-chain services: <none | indexer | bot | agent>

## Sponsor integrations (load-bearing only)
- <Walrus | DeepBook | Scallop | OpenZeppelin Sui | OtterSec | zkLogin>: <what surface, why load-bearing>
- If decorative was proposed, refuse and note here why it was dropped.

## Network and upgrade authority
- Target network at launch: <devnet | testnet | mainnet>
- Upgrade authority intent: <keep with multisig | keep solo | burn at mainnet>
- Package id capture plan: <where it lands, who reads it>

## Success criteria
1. <observable on-chain or off-chain outcome, with the network it's checked on>
2. <observable>
3. <observable>

## Out of scope
- <thing the user might assume but is not in>
- <thing>

## Constraints
- Deadline: <Sui Overflow 2026 cutoff, or other. Note, do not use the deadline to shrink scope by default. AI-assisted builds compress weeks into days.>
- Risk tolerance: <hackathon demo | testnet pilot | production>
- Existing assets: <Move package id, frontend, deployed pools>

## Open questions for the user
- <thing the agent could not pin down>

The skill stops here. No code. Hand off.

Workflow

1. Read prior context

• If .suiperpower/intent.md exists and is fresh, read it, summarize, ask if it still holds. If yes, skip to step 6.
• Otherwise read .suiperpower/idea-context.md and .suiperpower/build-context.md if present.
• Fast-path from idea phase: if idea-context.md already has problem, audience, success criteria, pre-fill those and only ask for the Sui-specific fields below. Do not re-ask what validate-idea already answered.

1. Restate the request

• One sentence, in the user's voice. Confirm before continuing. A vague "yeah" is not confirmation.

1. Walk the Sui interview

Group questions so the user is not nickel-and-dimed. Five tracks:

• Problem track: What pain, for whom, what do they do today (on Sui or off)? What is the one thing the product has to do well? Do not push for a scope cut here, AI-assisted pace means the full product is often in reach, the job is to name the load-bearing surface, not shrink the build.
• On-chain track: Do users hold Objects, or only the protocol? Which Objects are shared vs owned vs immutable? What capabilities exist and who holds each? Will users compose calls through PTBs?
• Off-chain track: Frontend or no? If yes, Next.js or mobile? Auth via wallet adapter, zkLogin, or sponsored transactions? Any indexer, bot, or agent?
• Sponsor track: Any sponsor integration claimed (Walrus, DeepBook, Scallop, OZ Sui, zkLogin, OtterSec)? For each: is it load-bearing in the user flow, or decoration? Refuse to record decorative integrations as load-bearing.
• Ship track: Target network at launch (devnet, testnet, mainnet)? Upgrade authority intent (keep, burn, multisig)? Deadline and risk tolerance?

1. Push back on slop

• "Everyone" for audience: push for a Sui-native wedge.
• "Good UX" for success: push for observable outcomes ("user can open the app, sign in with zkLogin, place a DeepBook order, see the fill on testnet").
• "Mainnet eventually" with no plan: surface that mainnet means upgrade authority decisions now.
• "It uses Walrus" with no flow: push for the specific blob lifecycle (write path, retrieval path, encryption posture). If no real lifecycle, drop the integration.
• One push-back round per topic. If the user insists on vague, record the vagueness honestly and move on.

1. Write `.suiperpower/intent.md` using the schema in Outputs. Open questions stay open. Timestamp it.

1. Hand off

• If non-trivial (multi-module, sponsor integration, or unclear object model), recommend plan-before-code.
• If one-step (single Move module, single sponsor surface), point at the specific build skill via skills/SKILL_ROUTER.md.

Quality gate (anti-slop)

Before reporting done, the skill asks itself:

• Did I write .suiperpower/intent.md, not just talk about it?
• Is the on-chain shape concrete: Objects named, ownership decided, capabilities listed?
• Is every claimed sponsor integration load-bearing in a user flow, not decorative import?
• Is the target network and upgrade authority intent named, not deferred?
• Are success criteria observable on a specific network, not aspirational?
• Did I list out-of-scope items, not only in-scope?
• Did I avoid starting any code, even a stub?

If any answer is no, the skill keeps working before handing off.

References

Knowledge docs (load when scope expands):

• core/skills/data/sui-knowledge/03-move-and-objects.md: Object model and capability reference.

Use in your agent

• Claude Code: claude "/suiper:clarify-intent <your message>"
• Codex: codex "/clarify-intent <your message>"
• Grok Build: run grok, then /clarify-intent <your message> in the session
• Cursor: paste a message like "step back, what am I really building on Sui", or reference ~/.cursor/rules/clarify-intent.mdc.

If you activated this and the user actually wants something else, consult skills/SKILL_ROUTER.md and hand off.