FAQ
What people ask before installing, including what it does not do. The kit's handbook has the rest.
Will it write my product for me?
No. The architects research, design and review. Solution and security may edit docs and code when a design calls for it, but building the product is your team's job.
Does it replace an experienced architect's judgment?
No. You get options, trade-offs and evidence, each claim labelled fact, claim or guess. It does not know what you promised your board or which customer is about to leave. A person decides. That is the point.
How good is the advice?
Only as good as the profile and the model behind it. An empty project.yaml gets careful questions, not answers; a filled one gets designs that name your stack, your laws and your standards. Read it as you would a capable new hire's first design.
What does it cost?
The kit is free. The architects run on the model in your Claude Code or Copilot plan, so you pay for that plan as you do today. No account, no server of ours, nothing to renew.
Do I need both Claude Code and Copilot?
No. Either works on its own. The architects are written for Claude Code, and the sync script generates the Copilot copies in .github/.
Will an architect change my code, or commit?
Solution and security may edit docs and code, then run the typecheck and test commands from your profile. Cloud may edit infrastructure files but never changes live resources. Research writes only in its research folder. None of them commits or pushes. You review and commit.
My project has no architecture docs yet. Does it still work?
Yes, and you do not write the first document yourself. The architects write them from the kit's templates: the HLD and the SAD, threat models, change plans, ADRs and research briefs, saved into docs/architecture/ as they go, and they create the CSVLOD folders the first time they write into one. An empty profile is enough to begin, because they ask for the facts they need and write the answers back into it. Diagrams are drawn as code, in your pipeline where the profile names one and as Mermaid in the Markdown where it does not. Nothing is committed until you commit it.
Can I send a design to someone who does not read Markdown?
Yes. An optional exporter, docs/architecture/tools/export.py, turns any architecture document into HTML, PDF or Word with the diagrams drawn. HTML needs only Python; PDF needs weasyprint or Chrome, Word pandoc, diagrams Node. It tells you what is missing. A workflow runs it on every pull request, so a diagram that does not parse fails the check.
Which model does it use?
Any model that can hold a long instruction. The kit is a customization rather than a model: each architect's lane, the templates it writes from and the rules it keeps to are all plain text, so they work with whatever you run. In Claude Code the agents declare model: opus in their front matter as a default. Change that line to suit your plan, or delete it and the agent follows the model the session is set to. The sync script drops the line for Copilot, which uses the model you pick in Copilot Chat.
What happens if an architect finds a live security weakness?
It never goes into the repository: not a threat model, a brief, a design, a diagram or a commit message. It goes to the restricted register you name in the profile, and straight to you.
Do I have to read Kotusev first?
No. The agents carry the method, and the kit's EA playbook is the short version. Read the book to know why the folders and rules are as they are.
Can I remove an architect, or add one?
Yes. Delete or copy an agent file and its command, update the board's list in arch-board.md and sync-copilot.sh, then run the sync script. The handbook lists every step.
How do I update the kit later?
Run the installer again with --update: it refreshes the kit's own files and leaves your project.yaml and documents alone. With the zip, download it again, unzip into an empty folder and compare.