How do I install the Do Something New skill?

Run npx skillmds@latest add imbue-ai/do-something-new in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Do Something New safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Do Something New?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Do Something New free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Do Something New?

imbue-ai (@imbue-ai) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Do Something New by imbue-ai — Agent Skill

Name: Do Something New
Author: imbue-ai

Do Something New

Use immediately when the user asks you to do something net-new -- a task you haven't done before, no existing skill or app applies, and getting it right will require nontrivial research, exploration, or experimentation. Routes the request to the right interactive flow. When doing this, give a very short confirmation message to the user's request, then load this immediately before responding further. Your confirmation message shouldn't mention loading the skill. Skip when an applicable skill or app already exists or for pure dev/code-writing work.

imbue-ai cd2f67f 3.1 KB Updated Sep 17, 2026

File contents

Doing something new

A net-new task that needs research or experimentation has landed. This skill is a thin router: it scans for an existing skill, then sends you to the right interactive flow. The actual flow lives in the specialization you route to (or, for the rare task that fits neither, in the shared principles).

The overriding goal of any net-new interactive task is the same: deliver a result the user cares about fast, confirm the basic shape cheaply, and defer the expensive thorough work to the background. That shape is documented once in .agents/shared/references/interactive-delivery.md -- read it; every route below is a specialization of it.

Step 0: Existing-skill scan

Match the user's ask against the descriptions of the existing skills in .agents/skills/. If a single skill covers the whole ask, bail out:

"There's already a <name> skill for this -- using it instead."

Then invoke that skill and stop. For partial matches (a skill covers a subset of the ask), continue routing but reuse the matching skill for the subset it handles.

Step 1: Route

Pick the flow that fits the ask:

Fetch / process / show data -- the ask is "go get this data (from an external service, API, or third-party source), do something to it, and show me the result." Route to fetch-process-show.
Build a web view -- the ask is "build me a page / dashboard / app I can look at." Route to build-app. It owns the interactive mock-confirmation flow for web work.
A hybrid: a web view over fetched data -- start with fetch-process-show to confirm the data sample, then it hands the surface to build-app (which runs its own UI-mock confirmation on top of the confirmed data). Begin with fetch-process-show.
Neither cleanly applies -- the task is genuinely net-new and not a data fetch or a web view. Apply the interactive-delivery skeleton directly from .agents/shared/references/interactive-delivery.md: clarify only what blocks, a fast feasibility pass, a small plan, validate the risky dependency first, a cheap throwaway creation looped to explicit confirmation, then hardening in the background. Fill in the skeleton's phases with the specifics of the task at hand.

Load the routed skill and proceed; do not re-implement its flow here.

When NOT to use this skill

An applicable single skill already exists -- use it.
Pure code-writing or dev work (bug fixes, refactors, adding a button). Those aren't research/exploration tasks.

---name: do-something-newdescription: Use immediately when the user asks you to do something net-new -- a task you haven't done before, no existing skill or app applies, and getting it right will require nontrivial research, exploration, or experimentation. Routes the request to the right interactive flow. When doing this, give a very short confirmation message to the user's request, then load this immediately before responding further. Your confirmation message shouldn't mention loading the skill. Skip when an applicable skill or app already exists or for pure dev/code-writing work.---# Doing something newA net-new task that needs research or experimentation has landed. This skill is athin **router**: it scans for an existing skill, then sends you to the rightinteractive flow. The actual flow lives in the specialization you route to (or,for the rare task that fits neither, in the shared principles).The overriding goal of any net-new interactive task is the same: deliver a resultthe user cares about *fast*, confirm the basic shape cheaply, and defer theexpensive thorough work to the background. That shape is documented once in`.agents/shared/references/interactive-delivery.md` -- **read it**; every routebelow is a specialization of it.## Step 0: Existing-skill scanMatch the user's ask against the descriptions of the existing skills in`.agents/skills/`. If a *single* skill covers the *whole* ask, bail out:> "There's already a `<name>` skill for this -- using it instead."Then invoke that skill and stop. For *partial* matches (a skill covers a subsetof the ask), continue routing but reuse the matching skill for the subset ithandles.## Step 1: RoutePick the flow that fits the ask:- **Fetch / process / show data** -- the ask is "go get this data (from an external service, API, or third-party source), do something to it, and show me the result." Route to **`fetch-process-show`**.- **Build a web view** -- the ask is "build me a page / dashboard / app I can look at." Route to **`build-app`**. It owns the interactive mock-confirmation flow for web work.- **A hybrid: a web view over fetched data** -- start with `fetch-process-show` to confirm the data sample, then it hands the surface to `build-app` (which runs its own UI-mock confirmation on top of the confirmed data). Begin with `fetch-process-show`.- **Neither cleanly applies** -- the task is genuinely net-new and not a data fetch or a web view. Apply the interactive-delivery skeleton directly from `.agents/shared/references/interactive-delivery.md`: clarify only what blocks, a fast feasibility pass, a small plan, validate the risky dependency first, a cheap throwaway creation looped to explicit confirmation, then hardening in the background. Fill in the skeleton's phases with the specifics of the task at hand.Load the routed skill and proceed; do not re-implement its flow here.## When NOT to use this skill- An applicable single skill already exists -- use it.- Pure code-writing or dev work (bug fixes, refactors, adding a button). Those aren't research/exploration tasks.

Frequently asked questions