build-new
Skill1 starsChanged 3 months agoMITView on GitHub
Build something new that does not exist yet, with tests. Use when the user asks for code that is not there yet — "write a X", "build me a X", "create a script that Y", "I need something that Y", "implement X from scratch", "start a new X", "we need a X". Use it for any request to write something new, rather than to change something that already works.
In this repository6 other files it gives its agents.
- debug-failureSkill
- explain-codeSkill
- plan-approachSkill
- review-diffSkill
- update-featureSkill
- write-testsSkill
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 7 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈7
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 13
Look for what already exists first.
- Distinctive: line 17
If the user asked outright for a replacement or a separate version, say what exists, build the new one beside it, and leave removing the …
- Distinctive: line 19
Read neighbouring code and tests before writing yours.
- Distinctive: line 24
Write tests alongside the code — the ordinary path, the boundaries, and the failure the code is meant to handle.
- Distinctive: line 28
Run the tests. Run the thing itself only where that is safe and it exits on its own: against a fixture, a temporary directory or a dry-run …
- Distinctive: line 34
Use what the project already depends on.
- Distinctive: line 38
If part of the request is ambiguous and the readings differ only in detail, build the one a careful colleague would pick, state the …
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Build something new
Build this: $ARGUMENTS. If that is empty, build what the user's last message asks for.
How to work
- Look for what already exists first. If a version of this is already here, this is an
update — read it and its tests, and work from them instead of writing a second one.
The same holds for existing code your new code is wired into: its tests stay green
and are not edited to pass, as
/update-featurerequires. If the user asked outright for a replacement or a separate version, say what exists, build the new one beside it, and leave removing the old one to them. - Read neighbouring code and tests before writing yours. New code that ignores the surrounding naming, structure, and idiom is a cost paid on every later read. In an empty project, pick the plainest layout and test setup for the language and say so.
- Build the smallest thing that satisfies the request end to end, and nothing beyond it. Do not add configuration, abstraction layers, or options nobody asked for.
- Write tests alongside the code — the ordinary path, the boundaries, and the failure the code is meant to handle. If the project has a test convention, follow that one. For something that cannot be unit-tested (a config file, a document, a one-off query), say so and show how you checked it instead.
- Run the tests. Run the thing itself only where that is safe and it exits on its own: against a fixture, a temporary directory or a dry-run flag. If it deletes, sends, deploys, migrates or writes outside the project, do not run it for real — give the user the command.
Rules
- Use what the project already depends on. A new dependency for something small needs a reason stated out loud.
- Handle the errors you can actually foresee. Do not wrap everything in a catch-all that swallows the failure.
- If part of the request is ambiguous and the readings differ only in detail, build the one a careful colleague would pick, state the assumption, and keep going. If they differ in what gets deleted, sent, exposed or charged, ask before building that part.
Report back
State what you built, where it lives, how to run it, and the exact test command with its summary line (passed, failed, skipped). If you could not run the tests, say so and why; do not call them passing. Name any dependency you added and anything you deliberately left out.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

