Scope Hammer

Scope Hammer

The brainstorm produced 30 ideas. The scope hammer asks: which 5 do you actually need to ship?

Companion to wide-open-brainstorm. Where brainstorm goes wide, scope-hammer goes ruthless.

When to Use

• After a brainstorm session
• When a feature list has grown unmanageable
• When the team is debating too many options and needs a forcing function
• When "MVP" has quietly expanded to "everything"
• When someone asks "what's the minimum we need to launch?"

When NOT to Use

• Before brainstorming (you need ideas before you can kill them)
• When the goal is exhaustive requirements gathering
• When stakeholders need to see all options (use brainstorm output; this skill produces the answer)

---

Phase 1: Load the Input

Accept one of:

• The output of a wide-open-brainstorm session (pasted or from context)
• A feature list, backlog, or spec
• A verbal description ("we're thinking of building X with features A, B, C, D, E...")

If the input is vague, ask: "Give me the list of features or ideas you want me to scope."

Extract the core goal first: "What does this product need to do to be worth using on day one?"

Phase 2: Classify Each Item

For every idea or feature in the input, apply exactly one classification:

Classifications

DELETE — This will never be in the product. Reasons:

• Adds complexity without adding core value
• Solves a problem users don't have yet
• Duplicates something else in the list
• Is actually a distraction from the core loop
• Would require significant infrastructure not justified by usage

MOCK — The user thinks they need this, but a fake version will do for launch. Examples:

• Manual process behind a button (instead of automation)
• Static content instead of dynamic CMS
• CSV import instead of API integration
• Human review instead of AI (at first)
• Hardcoded list instead of user-configured settings

ALREADY EXISTS — A standard library, tool, or service already does this. Don't build it. Examples:

• Auth → use an auth provider
• Payments → use Stripe
• Email → use Resend/Postmark
• Analytics → use Plausible/PostHog
• Search → use Algolia (or Postgres full-text for small scale)

SHIP — This is core. Users cannot get value without it. Ship this first.

Classification Rules

• Everything starts as DELETE. Earn its way to SHIP.
• MOCK is underused. Push more items to MOCK.
• If you can't describe why a SHIP item directly serves day-one user value, reconsider.
• Maximum 5 SHIP items for an MVP. If you have 8, go back and mock or delete 3.

Phase 3: Parallel Ruthlessness Check

Launch a second analysis as a subagent with the mandate: "Here is the proposed MVP scope (SHIP items only). Challenge it. Find: (1) SHIP items that could be MOCKed for launch without users noticing, (2) SHIP items that are actually implicit in other SHIP items, (3) anything that will block day-one launch but isn't in the SHIP list."

Incorporate findings into the final output.

Phase 4: MVP Specification

Produce:

The Core Loop

One paragraph: what does the user do from first launch to first value? This is the path the MVP must make possible. Everything else is optional.

SHIP Items (numbered, max 5)

For each: what it does, why it's required (not just nice-to-have), estimated complexity (S/M/L/XL).

MOCK Items (numbered)

For each: what we'd fake, how we'd fake it, when we'd replace it with real implementation.

ALREADY EXISTS Items (numbered)

For each: which tool/service handles it, link if obvious.

DELETE Items (brief list)

Just names — no need to justify each one in detail. "These don't make the cut for v1."

Launch Sequence Estimate

Rough order of implementation: what must be built first (blocking), what can be built in parallel, what can be built post-launch.

The Honest Summary

One paragraph: "To launch this, you need to build X, Y, and Z. Everything else is a future version. The first version can be live in [rough estimate] if you start with [first thing]."

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: carlkibler
• Source: carlkibler/agent-skills
• License: MIT

Install and usage instructions live in the source repository linked above.