EXPOUNDER

"I barely know her."

A Claude Code skill that turns a thin, napkin-thin project brief into a deep, layered, build-ready spec — before anyone writes a line of code (or prose, or a business plan). You bring six words; you leave with a plan you actually recognize as your idea, only sharper.

you:        "build me a game like Street Fighter"
EXPOUNDER:  → 4 sharp questions → a full spec in EXPOUNDED-SPEC.md → "want me to build it?"

Why it exists

A thin brief is a seed, not a plan. If you start building from the seed, you build the wrong thing twice. Most "expand my idea" tools balloon the scope and bury what you actually wanted. EXPOUNDER's single rule is go deeper, never sideways — every layer must be your idea expounded, not the model's idea inserted.

What it does

1. Offers first — on a thin brief it asks "Want me to EXPOUNDER this first?" and only runs if you say yes.
2. Anchors — asks 3–5 sharp questions where a wrong guess would poison everything (the north star, what's out of scope, platform, "like X but unlike X how?").
3. Expands quietly — does the heavy thinking into a file (EXPOUNDED-SPEC.md), not as a wall of text in chat. Small forks it just decides, logging each as a flippable Assumption.
4. Surfaces only real blockers — when two paths make genuinely different projects, it stops and asks. Otherwise it keeps going.
5. Hands off — ends with a Build Handoff block (a paste-ready kickoff prompt + first tasks) so the plan flows straight into building. Then offers to build.

The spec comes in layers — Anchor → Core → Systems → Depth → Build path → Optional — plus an Assumptions log, any Open Decisions, and the Build Handoff.

Features

• Anti-drift by design — explicit Out-of-Scope fence, [OPTIONAL] tags, invented names flagged as [PLACEHOLDER NAME], and a final anti-drift checklist.
• Any domain — software, games, writing, business, content. The structure adapts.
• Iterative deepening — already have a spec? Say "go deeper on the combat system" or "add a layer" and it expands in place without re-starting.
• Knows when to stay quiet — it does not fire on scoped work (a named file, a bugfix, a refactor, an already-detailed brief). If there's already a concrete thing to act on, it gets out of your way.

Install

As a plugin (recommended):

/plugin marketplace add danbengal/expounder
/plugin install expounder

Or drop the skill in manually — copy skills/expounder/ into ~/.claude/skills/.

Then just describe a thin project. On a thin brief, Claude offers to EXPOUNDER it — or invoke it any time with /expounder or by saying "expound this."

Heads up: it spends tokens

EXPOUNDER deliberately does a lot of background thinking — that's the trade: more tokens up front, a much sharper project on the other side. It caps itself at ~3k words by default and asks before going bigger.

Repo layout

expounder/                      (this repo = a one-plugin marketplace)
├── .claude-plugin/
│   ├── plugin.json             # plugin manifest
│   └── marketplace.json        # marketplace listing
└── skills/
    └── expounder/
        ├── SKILL.md            # the skill
        ├── README.md           # skill-level readme
        └── references/
            └── spec-template.md

License

MIT — do whatever you like with it. Built with Claude.