Agent Skills: intent

Turn rough ideas into clear work orders before planning or building. Use when request is vague like "add a button", "make it better", "fix the thing". Triggers on ambiguous or underspecified requests.

UncategorizedID: saadshahd/moo.md/intent

Install this agent skill to your local

pnpm dlx add-skill https://github.com/saadshahd/moo.md/tree/HEAD/hope/skills/intent

Skill Files

Browse the full folder contents for intent.

Download Skill

Loading file tree…

hope/skills/intent/SKILL.md

Skill Metadata

Name
intent
Description
Turn a rough ask into a clear work order. Use when the ask is rough, unspoken, or double-readable — before planning or building.

Extract

From the conversation/context/user:

  • The ask as given — the words on the table.
  • What blocks it — everything that would have to be invented to start the work now.

Gate

Proceed only when: starting the work now would mean inventing what the user means.

Anything else, say which in one plain line — never work the ask over anyway:

  • The ask is already one confirmed thing → proceed with the work.
  • The WHAT is confirmed and the open question is the technical path — architecture, technology, design → use shape skill.

Route

Before routing anything: retrieve what the record, the repo, or the world can answer. Only what the user alone holds routes to a skill.

Route each blocker by what it is, invoking the named skill with the Skill tool:

  • The words admit two readings that would build different things → use clarify skill.
  • The user holds what they want but can't yet say it → use elicit skill.
  • The user is missing something already settled — what exists, what a thing means — and can't ask past it → use explain skill.
  • The ask arrives already committed — a locked proposal with undefended decisions inside → use interrogate skill.

What each skill gives back re-enters the routing: a clarify pick can surface something the user can't express; an elicit record can leave words with two readings; an explain can let the user finally say the ask; an interrogate answer can amend the ask. Route again until nothing blocks the ask.

Where the input leaves it open, ask the negative once — what should this NOT do — before closing.

Output

The work order is the result: the ask restated in one line, confirmed by the user — plus what it must NOT do, where that surfaced. The routing ends only when starting the work would invent nothing.

Then check what still stands open:

  • Nothing — or one way to build it is already plain → proceed with the work.
  • The WHAT is confirmed but more than one way to build it remains → use shape skill.

The routing is spent once the user confirms the one-line ask — anything further runs under the branch picked here, never as another round of working the ask over.