Standards that outlive one repository

Why it is written down

Two projects sharing conventions by memory is a coincidence waiting to drift. I wrote the conventions down as a cross-project engineering standard — stack defaults, the performance budget, testing priorities, and a ranked source-of-truth hierarchy that ends: when a document and the code disagree, the code is right and the document is a bug.

What is deliberately absent

It includes a section listing what is deliberately absent, because unrecorded omissions get re-litigated every quarter by whoever arrives next. Adding one requires a written operating need. “It would be convenient” is explicitly not a need.

No CMS

Content lives in typed accessors.

No client-state library

Server components hold the state.

No component library

The design system is the tokens.

How AI is allowed to touch the work

The same standard defines how I use AI on delivery work: five specialist roles under a separation of powers written into the role contracts, reporting findings on a P0/P1/P2 severity taxonomy for a human to decide on.

Builder

May write files.

Reviewer

Report-only.

Tester

Report-only.

Auditor

Report-only.

Content strategist

Report-only.

Where the restriction actually lives

The restriction lives in each role’s instructions, not in tool permissions. No role declares `allowed-tools`, so nothing in the tooling enforces it — worth saying precisely, because the difference is exactly the kind of thing this page is about.

The tester’s protocol

The tester’s protocol puts it plainly: the pull request description is a claim, the diff is ground truth, and disagreement between them is itself a finding.