Explore until it's ready

Exploring turns a one-liner into something another person, or an agent with no context, could build from. You write durably on the idea itself as you learn, and only write a spec once the idea is ready.

What an explored idea looks like

Every idea has three sections. Exploring fills them in:

flowchart TB
    subgraph idea["IDEA-042 · Weekly digest should group by project"]
        direction TB
        S["<b>Summary</b><br/>the problem, who has it, what 'better' looks like,<br/>the rough shape of a solution, what's out of scope"]
        Q["<b>Open questions</b><br/>OPEN · RESOLVED — each one a decision<br/>to make before building"]
        L["<b>Log</b><br/>dated findings: what you tried, read,<br/>measured, ruled out"]
    end
    S --- Q --- L

Section	Command	Write here
Summary	nw idea summary IDEA-042 --set "…"	The current best understanding, rewritten as it improves. This is the gate for writing a spec
Open questions	nw idea questions IDEA-042 --add "…", --resolve 1	Decisions still to be made. Resolve them by recording the answer in the Summary or the Log
Log	nw idea log IDEA-042 --note "…"	Dated, append-only findings: code you read, experiments, numbers, dead ends

nw idea show IDEA-042      # read all three
nw idea mark-explored IDEA-042   # record that an exploration pass happened

The exploration loop

flowchart LR
    A["Read the idea<br/><code>nw idea show</code>"] --> B["Investigate<br/>code · data · prior ideas"]
    B --> C["Record findings<br/><code>nw idea log</code>"]
    C --> D["Update the Summary<br/>resolve / add questions"]
    D --> E{"Ready for a spec?"}
    E -- no --> B
    E -- yes --> F(["Write the spec"])

In an agent session, /nw-explore IDEA-042 runs this loop: it investigates, logs what it found, and sharpens the Summary and questions. It never promotes the idea; that's a deliberate, separate step.

Is it ready?

Write the spec only when you can tick these:

• [ ] The problem is stated, with who has it and why it matters, in the Summary.
• [ ] "Done" is concrete enough to test. You could list 3–8 checkable acceptance criteria.
• [ ] No blocking open questions: everything that changes the design is resolved.
• [ ] The scope fits one campaign. If it's several independent pieces, it's an epic (see Write a spec).
• [ ] Out of scope is written down, so the loop won't build what you didn't ask for.
• [ ] Related work is linked: other ideas, specs, files in the code.

If exploring shows the idea isn't worth building, close it with a reason, so the lesson stays findable:

nw idea close IDEA-042 --as researched --reason "solved by existing --by-project flag"
nw idea close IDEA-043 --as drop --reason "duplicate" --folded-into IDEA-012

Next: Write a spec the loop can build.