Joseki
In Go, a joseki is a sequence that has been studied until it is known to be sound — you play it because the work of checking has already been done. That is what we build for engineering work an agent does: the studied part, so the agent's speed is worth having.
Why toolkits
A model that writes infrastructure, or a migration, or a release, is genuinely fast — and it is exactly as fluent when it is wrong. The output looks the same either way. That is the whole problem: the failure mode is not gibberish you would catch, it is something plausible you would not.
Prompting harder does not fix that, because the thing missing is not instruction. What is missing is the part a careful engineer supplies without being asked — knowing which move is safe here, refusing the one that is not, and stopping to show you the diff before it lands.
So we build that part: rules an agent reads before it writes, tools that make the real change deterministic and reviewable, and guards that refuse the commands nobody should run by accident. Each tool states what it does not cover, because a safety claim you cannot keep is worse than none.
What we build
Clawform is on sale. ClaudeBrew is not, and its card says which parts exist rather than implying the rest are close.
Both are early versions of something wider. The pattern — rules read before writing, tools that make the real change reviewable, guards that refuse the irreversible — does not depend on AWS, on CloudFormation, or on Claude Code. Other clouds, other infrastructure languages and other agents are where this goes. No dates: a schedule from a one-person shop is a guess wearing a calendar. Buy what works today.
available
Safe CloudFormation workflows for Claude Code
Claude writes CloudFormation fast, and just as confidently when it is wrong. Clawform sits between the agent and the account: rules it reads before it writes, a CLI that runs every AWS call through a changeset you confirm, and a hook that vetoes dangerous commands before they execute. It is a deterrent at the point of action, not a sandbox, and it says so.
$49 one-time, per seat — 3 device activations
in development
A full development lifecycle, as Claude Code skills
One skill per stage — brainstorm, requirement, design, code, test, ship — each handing a reviewed artifact to the next and stopping at a hard gate rather than cascading. Nothing is assumed: uncertainty is surfaced, claims are cited, and no stage begins until you approve the one before it. Early, and honest about which stages exist.
2 of 7 stages built — installable now
Contact
No form, no list, no sequence of emails you did not ask for. If you want to know when something ships, or whether it would fit what you run, write and say so.
