chrome-extensions.sgit.ai / unbuilt

The unbuilt: four extensions specified, none shipped

The estate has shipped one extension — the recorder. It has specified at least four more across six months, in documents of real engineering depth, and built none of them. That could be read as a backlog. It is more useful read as a body of platform findings acquired without writing code — several of them the kind that normally costs a week each to discover late. Confirmed by the founder, 8 September 2026: none of the designs below were built.

1shipped
4+2specified, never built
6 moMarch → September 2026
#ProposalDateVersionStateThe finding it bought
U1Key Vault5 Mar 2026v0.11.12dev pack, 28 files, never builtAppSec staffed by what the artefact holds
U2Risk cards on extensions4 Jul 2026v0.33.42analysis, cards unbuiltThe read-content closure
U3The Scribe14 Aug 2026v0.33.58architecture briefactiveTab; the narrow remote-code rule
U4Secrets manager5 Sep 2026v0.33.65dev brief, re-spec of U1The sync budget; the enforcer test
U5The manifest as a profile5 Sep 2026—one line, cheapA grant declared by the thing it describes
U6The vault as extension storage8 Sep 2026—design positionRemoves U4's ceiling rather than working around it

Why this is a section and not an appendix

The Manifest V3 platform punishes assumptions, and every brief above closed off an approach that looked reasonable before someone read the documentation. Each page states date · status · what it discovered · why it has not been built, and the claim-state discipline applies here too: a design is read at best, never measured — the first extension to ship against any of these findings is the first measurement.

The findings, promoted

activeTab over broad host permissions, the remote-code rule and its narrowness, the vault-as-storage composition, the sync budget, the read-content closure — these are true regardless of whether the extension is ever built. They are cross-linked into the architecture guide and the checklist rather than left stranded in a brief nobody reopens.

A principle arrived at twice

U3 and U4, written three weeks apart, independently converge on minimising what the extension holds — a channel instead of a key, or ciphertext instead of plaintext. Written down once as a design principle rather than left as two coincidences.