Files
zeekay 079c63f6cb Add the iOS lane: hanzo-templates had zero of them
The catalogue shipped 71 web templates and one EMPTY `mobile` repo, so
"build an app" had no starting point on the platform people actually
ship to. This is the iOS lane — Expo/React Native and Flutter for
cross-platform, SwiftUI for native.

The constraint that shaped every file here: an iOS archive needs macOS +
Xcode + a signing identity, and our arc fleet has none. hanzoai has 519
registered runners and every one reports `os: linux`; luxfi has 151, same
story; zooai has none. The only macOS builds in the org today
(hanzoai/world build-desktop.yml) run on GitHub-HOSTED macos-14, which is
exactly what the house CI rule says not to do.

So `hanzo.yml` declares only what a Linux arc runner can genuinely
execute — typecheck + web bundle for Expo, analyze + test + web for
Flutter, `swiftc -parse` for SwiftUI — and the macOS archive lives in a
separate `workflow_dispatch`-only `ios.yml`. A job pinned to a label no
runner advertises would sit queued forever if it fired on push; parking
it is honest, a permanently-yellow check is not. The README of every
template states the gap and the one command that closes it.

Same reasoning on the demos. Expo and Flutter both have a real web
target, so `<slug>.hanzo.app` serves the SAME source the simulator runs.
SwiftUI has no web target, so those two templates ship with no demo link
rather than a screenshot pretending to be one.

Upstreams are permissive only and recorded per template: expo/expo
templates (MIT), flutter/samples (BSD-3-Clause), XcodeGen (MIT).

Second measured gap, found while wiring this up: `hanzo-templates` is not in
arcd's `orgs:` list either (hanzoai, luxfi, zooai, parsdao, zenlm), so the
Linux jobs would queue with no runner too. Both gaps are one installer
re-run, no code change; both are spelled out in every README rather than
left to be discovered on the first red build.

Third gap, found by probing rather than assuming, and FIXED here: every
run in this org died as `startup_failure` with no logs and no annotation.
A bare `runs-on: [self-hosted, linux]` probe queued fine while the same
job importing hanzoai/ci failed identically at @v1 and at a pinned SHA —
so it was never the tag. The org defaulted GITHUB_TOKEN to `read`, and
the reusable declares `permissions: packages: write`; a caller cannot
exceed the org maximum, so GitHub kills the run before creating a job.
hanzo-templates now defaults to `write` like hanzoai/luxfi/zooai, and
runs reach `queued` — the honest state, waiting on gap 2.
2026-07-27 23:25:30 -07:00

22 lines
1.0 KiB
Plaintext

MIT License
Copyright (c) Hanzo AI
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.