BUILD 0.1.0 / A workflow skill for Codex
From a good idea
to a real thing.
CONCEIVE designs.
BRAID makes and checks.
Five perspectives work through the design. Three responsibilities carry it into the world. Your purpose, constraints and good reasons travel with it.
Complete ZIP · v0.1.0 · Host tools and model allowance required
The handoff matters
A finished thing should still know why it was made. BUILD carries the original intent through design, implementation and independent checking.
01 / The method
Give thought room.
Give work responsibility.
Separate contexts can challenge a design. Separate responsibilities can make it, judge it and keep it moving.
CONCEIVE
Design through five distinct contexts
Invent and reframe.
Integrate the whole.
Bring the world in.
Test the inference.
Follow the consequences.
One substantive handoff
Purpose, design, constraints, decisions, uncertainty and what success looks like.
BRAID
Make, check and supervise
MAKER makes.
Owns the deliverable and the changes needed to make it useful.
CHECKER checks.
Independently judges the actual result against the intended outcome.
SUPERVISOR supervises.
Directs the effort, resolves drift and keeps the next action moving.
If making exposes a material design problem, return the question and evidence to design. Preserve the work that still holds.
02 / Choose your starting point
Use the part
your work needs.
BUILD includes complete CONCEIVE and BRAID workflows. Both can be requested directly.
For a tool, lesson, publication, event or other concrete outcome. The host determines what can actually be executed.
03 / Get started
Keep the package together.
Start with something small.
Download & extract
Get the release ZIP. Confirm it contains
Download v0.1.0build/SKILL.md,INSTALL.mdandREADME.md. Keep the completebuildfolder.Install in Codex
Give Codex access to the extracted folder. Ask it to install the complete skill, preserve any existing version and report the installed location.
Full installation stepsVerify & try
Have Codex read the installed entrypoint and find all three workflows. Then try the first-look prompt below before a full launch.
Try a first look
Using another chat? It must be able to read the complete extracted resources. That is session use, not permanent installation or a verified native integration. Read the session-use limits.
Begin without launching collaborators
Take a first look
Use BUILD to explain how you would approach a small offline vocabulary practice tool. It should take an editable word list, work with a keyboard, and allow the learner to stop at any time. Do not create chats, schedules or files yet.Expect the understood goal, available collaboration mode and a proposed next step.
The full journey
Launch a BUILD
This command requests five communicating design chats, then three execution chats when ready, plus one recurring SUPERVISOR check normally every ten minutes until completion. Your narrower instructions and host capacity still govern. Add “Do not schedule unattended supervision” to exclude the recurring check.
Create a BUILD called Vocabulary Lab: design and make a small offline vocabulary practice tool. It should accept an editable word list, work with a keyboard, and let the learner stop at any time. Keep the first version small. Do not publish it or use paid services.Expect real participant references and observed state. At the end: a usable result, its independent verdict and material remaining limits.
Copy a CONCEIVE design-only request
Requests five communicating design chats and stops at the design.
Create a CONCEIVE named Workshop Kit to design a 45-minute neighborhood repair workshop in five communicating chats. Assume twelve adults, basic hand tools and no internet at the venue. Deliver the design, its reasons, unresolved conditions and next action. Stop at design; do not implement or publish.Copy a BRAID execution request
Supply the approved design with this request. It creates three roles; this example excludes unattended supervision.
Create a BRAID to do Workshop Kit: turn this approved workshop design into a printable facilitator guide and participant checklist. Preserve the stated time, audience and tool constraints. Check whether a facilitator can follow the sequence and tell when the workshop is complete. Do not publish it or use paid services. Do not schedule unattended supervision.04 / Stay in control
More context.
Real limits.
Set the boundary before launch.
Multiple contexts and recurring supervision use your host's time and model allowance. BUILD includes no model quota and promises no fixed cost. State a budget, choose only the workflow you need, and exclude unattended supervision when it does not fit.
Bring the right context.
Provide the outcome, users, constraints and existing decisions. Relevant material passes to collaborators. Keep secrets out of briefs; your host and connected tools govern data handling.
Finish means the thing is usable.
Prepared briefs, created chats and active collaboration are different states. Completion means the intended result exists, its delivered revision matches the applicable independent verdict, and remaining limits are explicit.
Guide scope, cost and completion05 / The user guide
A way in.
A way back.
A clear evidence boundary
Package checks and three simulated discovery requests support the release's structure and workflow routing. They do not establish a live native eight-chat launch, uninterrupted supervision or universal host support. The examples illustrate the method.
Need a hand?
Report a reproducible problem with your version, host, workflow and observed state. Remove private material and secrets.
BUILD / 0.1.0