ideate

Files (5)

• SKILL.md
• evals.json · evals/
• product-idea-bad.md · evals/files/
• product-idea-good.md · evals/files/
• targets.json · evals/

SKILL.md (first lines)

# Ideate

Turn an operator's business context and expertise into one sharpened product
direction. The runners-up are preserved so a direction can be reconsidered
later without rerunning the whole pass.

Patterns adapted from `buildgreatproducts/builder-os` (idea-generator), MIT
licensed. Adapted, not copied.

## Not the intake skill

`ideate` generates a net-new direction from nothing. `intake` re-homes an
existing external client project onto your own GitHub and Vercel. No functional
overlap, only the English word "intake" is close. If the operator already has a
project to adopt, use `intake`. If they have a business and no product yet, use
this.

## When to Use

- The operator has a business, expertise, or audience and no decided product.
- Before the design pass, to feed it a validated direction instead of a cold
  start.

Not for: a decided product that needs a spec (go straight to the design pass).
Not for: adopting an existing external project (use `intake`).

## Profile gate

None. This skill is profile-agnostic. It runs before `/arkira-init`, because a
greenfield idea may have no repo yet. It runs whether or not
`.arkira/config.json` exists, and it does not nag the operator to run
`/arkira-init`. It references no app-only or static-web-only standard.

## Artifact contract

Reads operator free text. Writes `docs/product-idea.md`. Consumed by `validate`
and the design pass. See `docs/document-contract.md`.

## Workflow

### Step 1 - Mine

Ask for the operator's business, expertise, unfair advantages, and the customers
they already reach. Summarize that context in a short block. This becomes the
Operator input section.

### Step 2 - Synthesize

Produce three to five distinct candidate directions. Each is one paragraph: the
problem, who has it, and the wedge. Keep them genuinely different, not variations
of one idea.

### Step 3 - Score

Score every candidate on the five-axis card, one (weak) to five (strong), each
with a one-line justification:

- Pain. How acute is the problem for the person who has it.
- Reach. How many people or accounts have it.
- Willingness to pay. How ready the buyer is to pay to solve it.
- Founder edge. How well the operator's expertise, assets, and audience fit.
- Feasibility. How buildable a first version is by a small team.

### Step 4 - Select

Recommend the winner and state why in terms of its scores. The operator confirms
or overrides. Do not invent a market-size or revenue number to justify a choice.
If a number is used, it carries a source.

### Step 5 - Sharpen

Expand the winner into the selected direction: problem, target customer, the
wedge, why now, why this operator.

### Step 6 - Write

Write `docs/product-idea.md` in the structure below.

## Output: docs/product-idea.md

```markdown
# Product Idea: <working title>

## Operator input

<three to six lines summarizing the mined business, expertise, and reachable customers>

## Candidates

<written once, never rewritten by any later skill>

### C1. <name>

<one paragraph: problem, who has it, the wedge>

| Axis | Score | Justification |
|------|-------|---------------|
| Pain | n | ... |
| Reach | n | ... |
| Willingness to pay | n | ... |
| Founder edge | n | ... |
| Feasibility | n | ... |
| Total | n | |

### C2. <name>

<same shape, three to five candidates total>

## Selected direction

<the sharpened winner: problem, target customer, the wedge, why now, why this operator>
```

## Resumability

Re-read `docs/product-idea.md` if it exists. When a `## Candidates` section is
already present, skip synthesis and resume at selection or sharpening. The
`## Candidates` section is written once and never rewritten. This is the resume
contract. No separate state file is used.

From jeanchastel/arkira, MIT.

More from jeanchastel/arkira