Foundation

Establish the shared business context (voice, philosophy, audience, constraints) that every other workbench draws from.

What it does

Foundation captures who you are as a business through a guided conversation, then stores that identity as a set of context files your other workbenches read automatically. Once Foundation is complete, every tool in your workshop knows your voice, your audience, and the operating constraints you care about. None of that has to be re-explained for each task.

There is no form to fill out and no configuration file to hand-edit. Your CLI agent runs the setup as a dialogue – asking one question at a time, following up on short answers, and picking up on your natural voice throughout.

Foundation has two states:

UNCONFIGURED – setup has not been completed. Other workbenches lack shared context.
READY – all six context files are in place. Other workbenches draw from them automatically.

When to use it

Foundation is the first thing to set up in a new workshop. Other workbenches read from it. If Foundation is not complete, they have no shared context to draw on and will produce generic output.

Come back to it whenever your positioning, voice, or audience has shifted. The context files are living documents – edit the relevant one and your workshop reflects the change on the next use. You do not need to re-run the full setup conversation for a minor correction.

What you get

The setup conversation walks six sections in order. Each one produces a context file that other workbenches reference.

Section	What it captures
1 – Soul	Mission, vision, purpose, and promise
2 – Philosophy	Core values, problem-solving approach, and what you stand for
3 – Audience	Who you serve, their pain points, and how they engage with you
4 – Voice	Tone, formality level, preferred terminology, and style
5 – Business Context	Services, pricing structure, sales process, and differentiators
6 – Constraints	Technical limits, scope boundaries, and quality standards

A seventh optional context file holds the name you choose for your CLI agent within this workshop. It ships blank by default. The setup offers the option but never requires it.

Other workbenches load whichever files they need for their task. Content-facing workbenches draw primarily on voice and audience. Visual identity workbenches draw on the branding context. All workbenches use soul and constraints as a baseline reference.

How to use it

Starting setup

Say setup to your CLI agent. The conversation begins immediately – no flags, no prior steps. Plan for 30 to 60 minutes. You can pause at any section boundary and pick up later.

What the conversation looks like

The agent opens with a warm-up phase to calibrate to your pace, then works through Sections 1 through 6 in order. The shape stays consistent throughout:

One question at a time, never a checklist
Follow-up questions when an answer is brief or leaves something implicit
Passive voice observation throughout (the agent notes how you write naturally. It asks you to confirm in Section 4 rather than asking “describe your tone” cold)
A section summary at the end of each section, with a chance to add or correct before moving on
A pause option at every section boundary

After Section 6, the agent revisits anything you skipped and offers low-confidence inferences for your review. You can accept, correct, or leave items blank.

Review mode and confirmation

Before writing any files, the agent displays everything it captured across all six sections and asks for your explicit confirmation. Nothing is saved until you say it looks right. If something is off, say so – the agent edits and loops back until you confirm. Only after confirmation does it write the context files and mark Foundation as READY.

Checking status and resuming

What to say	What happens
`setup`	Start (or restart) the full setup conversation
`status`	Show whether Foundation is UNCONFIGURED or READY
`resume` or `continue`	Pick up a paused setup at the last section boundary

Naming your CLI agent (optional)

During the warm-up phase, the setup offers to give your CLI agent a name for this workshop. This is entirely optional – the default is no name. A name is an alias: how you address the agent in conversation. It is not a persona or a separate product. The agent always remains an AI and never implies otherwise.

Set a name once in Foundation and it is honored in every session that follows, across every workbench in the workshop. Change it here and it changes everywhere – it is the single source for that name.

The name you set is yours. It is not provided by or associated with the platform. It is simply how your own CLI agent is addressed in your workshop.

Common pitfalls

Rushing through sections with short answers. The conversation follows up on stubs, but if you push past the follow-ups and confirm a thin capture, downstream workbenches will produce thin output. The quality of what comes out reflects the quality of what goes in. Take the time. The sections are structured to draw out what you actually mean, not just the first thing you say.

Confirming Review Mode before reading it carefully. Foundation does not write files until you confirm. That confirmation is the gate into READY. Skim-confirming means thin captures propagate everywhere until you fix them manually. Read through the full summary, not just the headings.

Reaching for other workbenches before Foundation is READY. If Foundation is UNCONFIGURED, other workbenches have no shared context to read. Run status first. If the result is UNCONFIGURED, complete setup before moving to the next tool.

Treating context files as static. They are living documents. If your positioning shifts, your audience narrows, or your constraints change, edit the relevant context file. The change takes effect the next time any workbench reads it. You do not need to re-run the full conversation for a targeted update.

Thinking the assistant name carries outside your workshop. The name you set lives in Foundation’s context for your workshop only. It is not a platform-wide identity. Other workshops are named (or unnamed) independently.

Leaving the branding context file empty and expecting visual consistency. The branding context is not produced by the six-section conversation – it exists alongside the other context files and is read by visual identity workbenches, but you fill it in separately. If your brand work is producing generic output, check whether the branding context has been populated.

Troubleshooting

Symptom	Likely cause	Fix
Other workbenches produce generic output, ignoring your voice	Foundation is UNCONFIGURED or key sections were confirmed with thin content	Run `status`. If UNCONFIGURED, run `setup` or `resume` to complete it
Setup was interrupted before you finished	Session ended before reaching a section boundary	Say `resume` or `continue`. The agent restarts from the last completed section
Status shows READY but output still feels off	A section was confirmed with brief or inaccurate content	Edit the relevant context file directly, adding the detail that was missing
You want to update voice or audience without re-running full setup	Business has evolved since initial setup	Open the relevant context file and edit it. The change takes effect immediately on next use
Visual identity workbenches produce no brand output	Branding context file was not filled in	Populate the branding context file with your colors, typography, and logo references
The assistant name you set is not being used	Name was not captured during setup, or the context file is blank	Open the assistant context file and add the name you want. It is honored at the next session start
You confirmed Review Mode but want to change something	Content was persisted after confirmation	Edit the relevant context file directly. The files are plain text and are meant to be edited as the business evolves