ADHD Skill: Multi-Model Brainstorming for Divergent Ideas

Where this fits

You're deep in an architecture review and your go-to model keeps suggesting the same three patterns you've already considered. You want fresh angles, but manually switching between Claude, Codex, or another model—and pasting the same prompt—eats your flow. The adhd skill automates this: it hands your exact prompt to several models in parallel, letting each apply its own cognitive frame and priors. The result is a wider set of ideas without extra context-switching on your part.

Why agents benefit

• Divergence by design: The same brainstorm is executed by different model families (Claude sub-agent, OpenAI Codex, a cheap third via OpenCode), so you get varied perspectives that a single model would never produce.
• Parallel execution: The skill orchestrates multiple agent calls simultaneously, cutting the wall‑clock time for a brainstorm from minutes to seconds.
• Zero manual hand‑off: No need to copy prompts between tools or remember which model you already asked—the skill handles the routing and returns consolidated results.
• Fits open‑ended queries: It's tuned for fuzzy problems (design questions, naming, schema choices) where width of ideas matters more than a single canonical answer.
• Leverages existing agent stack: Works with your current Bash, Read, Grep, and Agent tools—no new infrastructure required.

Practical scenarios

Naming a new API endpoint

You need a verb that conveys idempotency and partial updates. Run multi adhd "What should we call a PATCH endpoint that only updates provided fields?" and get suggestions from models with different linguistic biases—some lean technical, others product‑friendly.

Debugging a fuzzy concurrency bug

A race condition appears only under load, and logs are ambiguous. Use the skill to brainstorm root‑cause hypotheses across models. Each model's prior on concurrency (e.g., locking vs. transactional) surfaces distinct leads you can then test.

Designing a database schema for an evolving product

Your schema needs to support future multi‑tenancy, but you're unsure about the trade‑offs. The skill generates divergent schema ideas—from normalized to document‑style—so you can compare before committing.

Add it to your agent workflow

Install the skill with:

npx skills add szarkans/multi --skill adhd

Then invoke it in any agent session:

multi adhd "How should we structure our event-driven microservices to handle retries?"

The skill will fan out your prompt to multiple models and return a consolidated, wide‑angle set of answers.

Compare before adopting

Before you rely on it, check: Quality signals—the repo has only 4 stars, so examine the code and test it on your own problems. Maintenance freshness—look at commit history and issue responses; a stale skill may break with agent updates. Alternatives—you could manually run the same prompt in separate model windows, but that loses the orchestration. Workflow fit—the skill shines for open design questions; skip it for lookups or syntax where a single model is faster.

Why it is worth tracking

Despite low stars, the adhd skill is based on a known pattern (UditAkhourii's ADHD) and addresses a real pain: model monoculture in brainstorming. As multi‑model orchestration becomes standard, this skill's utility grows. Watch for updates, community adoption, and whether the author improves the probe scripts. Evaluate it when you need broader ideation without manual model hopping—and see how it fits your existing agent toolkit. For more context, check the skill's listing at OpenAgentSkill.