perf(ci): cache dependencies and shard web and server tests - #1985

## Problem

After #1985, the pooled **`Test` job is the critical path** at **8:43** (run 41293089108).
Inside it, **`apps/mobile` ran 432 s** (260 files) and finished about 20 s after every other package; the next longest took ~157 s.

## Change

- **New `Test Mobile` job:** mobile tests only
- **Setup:** no Electron, no libsecret
- **`Test` drops `@t3tools/mobile`**
- **`Test` skips** on a scripts or mobile-only change
- **`CI` aggregate** needs `test_mobile`

**Gating:** `changes` gains a fork-only `test_mobile` output, derived in the workflow from `test` and `test_packages`.
The shared `ci-changes.cjs` stays upstream's, and the job skips without taking a runner; `CI` already tolerates a skip whose gate is `"false"`.

`test_mobile` per change, from `ci-changes.cjs` output:

| Change | `test_packages` | `test_mobile` | `Test` |
| --- | --- | --- | --- |
| `apps/web` only | `@t3tools/scripts` | false | skips |
| `apps/mobile` only | mobile, scripts | true | skips |
| `apps/server` only | evals, scripts | false | runs |
| `packages/contracts` | includes mobile | true | runs |
| Workflow file | empty (all) | true | runs |
| Docs only | empty, `test=false` | false | skipped |

## Scope and approval

Fork CI tuning only; follows #1985. No product code.

## Verification

- `actionlint` on the workflow → exit 0
- `vp run check:changed` → exit 0
- `vp run fork:ci` → exit 0

**Wall times** (before: run 41293089108; after: run 41296032295):

| Job | Before | After |
| --- | --- | --- |
| Test | 8:43 | 5:11 |
| Test Mobile | n/a | 3:05 |
| Slowest Test Server | 5:51 | 5:21 |
| End to end | ~9:00 | 7:37 |

- **Critical path:** Test Server 6, 2:03 queued
- **Runner cap:** 5 jobs queued 1:36 to 2:44
- **Test Server 2 red:** unrelated test time bomb

Claude Opus 5.5, Claude Code harness in T3 Code.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: qol
Co-authored-by: donjor <[email protected]>

… changes since the pin (#1991)

Closes #1987

## Close conditions

The three copies above:

- [ ] **Each copy** carries a pin
- [ ] **`fork:scan` fails** a stale pin
- [ ] **Failure names** the file and commits
- [ ] **Tests:** stale, current, missing pin
- [ ] **Fork-sync doc:** adapt or no-change step

## Problem

The `hyprws-*.yml` copies of upstream workflows drift silently: upstream `ci.yml` has 26 unmirrored commits.

## Change

- **Pin:** `# fork-mirror: <path>@<sha>` in each header
- **Stale pin:** `fork:scan --mirror-pins` fails it
- **No pin:** fails too
- **Sync tip only:** flag set when `rebased`
- **Ordinary PRs:** never fail on a stale pin
- **No ledger:** the pin lives in the copy
- **Docs:** `fork-sync.md` *Workflow copies*
- **Skill:** fork-sync tip-work bullet

The guard #1151 deleted stays deleted.

Scan of this head against the current base:

```text
$ node scripts/fork-scan.ts --no-typecheck --mirror-pins --target ae521d1 --since origin/hyprws
Workflow mirrors, 1 finding(s):
  MIRROR  .github/workflows/hyprws-ci.yml: upstream .github/workflows/ci.yml changed in 26 commit(s) since pin 5d14c0e; adapt the copy or record no change, then pin ae521d1
  f6d3700 ci: server test shards survive a throttled task cache (pingdotgg#26332)
  67661cf chore: delete 18k lines of code that nothing reaches (pingdotgg#26238)
  ...
exit 1
```

## Pins

| Copy | Pin | Adapted |
| --- | --- | --- |
| `hyprws-ci.yml` | `5d14c0e9686` | none |
| `hyprws-release.yml` | `ae521d1d796` | `e6ca17001d0` |
| `hyprws-agent-context-budget.yml` | `ae521d1d796` | `da1eab01f71` |

- **CI pin:** the last upstream commit it matched
- **CI drift:** the next sync reviews it
- **`e6ca17001d0`:** prerelease drafts until assets land
- **`da1eab01f71`:** error names the evals rerun

#1985 took upstream cache steps only, so the CI pin stays.

### Release: no change

| Upstream commits | Reason |
| --- | --- |
| `f41518f` | launch check: untested release behaviour |
| `30f4d63` `fe27e8e` `e407f9b` `25159b6` `fd1c338` | test jobs: the gate needs green CI |
| `a9418a1` `6327560` `a43b003` | task cache: no secret, `setup-vp` |
| `1688f96` `783ccf0` `7479625` `bd0b9ed` `e74c668` | deploys: the fork has none |
| `e64766c` `569bd02` `14b47ee` `e9a5974` `c57a04b` | macOS, Windows: Linux only |
| `c13f7d9` | `.deb`, arm64: AppImage only |
| `6b450c7` `91cd91c` | npm: not published |
| `83d6583` | mobile: no fork build |
| `b438198` | Discord: no announce |
| `b5fc7be` | OpenAI notes: generated notes kept |
| `3060cc4` | channel defaults: fork resolves its own |
| `bed15b9` | skip in forks: copy is fork-owned |
| `28c32d2` | telemetry: defaults to `dev` here |
| `eb8f6f4` `c7f23c4` `0754920` `2c54f2f` `2f7616e` | CLI archives: already built |

## Scope and approval

The maintainer filed #1987 and assigned this scope.

## Verification

- [x] **Typecheck** (scripts) → exit 0
- [x] **Tests** (3 files) → exit 0, 27 passed
- [x] **check:changed** → exit 1
- [x] **Its two failures:** web browser timeouts
- [x] **No web code** in this diff
- [x] **Both pass** run alone
- [x] **fork:ci** (bwrap HOME workaround) → exit 0

Claude Opus 5.5, Claude Code in T3 Code.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: core
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

## Problem

PR CI takes **10.4 minutes median** on GitHub-hosted runners; server and web tests share the critical path.

## Change

Reuse upstream's dependency, Chromium, and Electron cache steps without its private task-cache token or Blacksmith assumptions.

```diff
- Test Server: 3 shards
- Test Web: 1 job
+ Test Server: 6 shards
+ Test Web: 2 shards
```

Six server shards finish in **4:48–5:51**; two web shards finish in **4:59 / 5:34**.
Server test steps measure 209–273s (232s median), against the supplied three-shard median of 338s; web steps measure 256s / 265s, against 478s unsharded.
More web shards would not shorten this run: the unchanged pooled `Test` job is now the critical path at **8:43**, including 432s for mobile tests.
The approximate six-minute goal is not reached.
The workflow has 14 jobs including Changes and CI, below the org's 20 concurrent hosted-job limit.
The aggregate still depends on both matrix job IDs; ruleset 21768840 requires only `CI` and `merge-tree`, so no ruleset change is needed.

## Scope and approval

The maintainer explicitly approved this CI-only implementation and draft PR.
No `test:cached`, flaky-test fixes, ruleset changes, or upstream writes.

## Verification

- [x] Workflow passes actionlint
- [x] Browser/unit shard probe: 18 tests passed
- [x] Each probe shard runs one browser file
- [x] Each probe shard runs one unit file
- [x] No overlap across the four probe files
- [ ] `vp run fork:ci`: exit 1, Scripts failed

The committed-head preflight passed delta, stale-delete, fixup-fold, merge-tree, scan, knip, check, typecheck, and fork-tests (220 files, 1215 tests).
Scripts passed 596 tests and failed one unchanged Discord release workflow test at `scripts/notify-discord-release.test.ts:365`.
A subprocess trace shows the host shell fails before the fixture runs: `/home/user/.bashrc: line 255: path_prepend: command not found`, exit 127.
Full local web attempts also hit browser module/mock errors; the focused shard probe passed with one worker and Playwright's Ubuntu 24.04 x64 platform override on this Ubuntu 26.04 host.
### Hosted CI measurements

[Run 41293089108](https://github.com/RSI-Software/t3code-hyprws/actions/runs/41293089108) is **green on attempt 2**, with 16 checks passed and one unrelated preview check skipped.
Attempt 1 failed only the unchanged Mermaid SVG assertion; its full file passed 10/10 locally, and a governed job rerun passed all 276 shard files and 3677 tests without source changes.
The no-token dependency action succeeded; logs show its private task-cache step skipped, Chromium restored, and the ldd check took under one second.

The green run took **14:51 end-to-end including the failed attempt and rerun**; the rerun attempt itself took 5:08, reusing all other job results.
Attempt 1 took 9:01 but was red, so it is not a green-run speed claim.
The supplied historical median is 10:24; a concrete earlier green run, [41278753510](https://github.com/RSI-Software/t3code-hyprws/actions/runs/41278753510), took 11:50.
These are observations, not a controlled before/after benchmark or a new median.

Job wall times from the Actions jobs API:

| Job | Earlier green | This green run |
| --- | --- | --- |
| Changes | 0:05 | 0:07 |
| Check | 4:37 | 3:54 |
| Test | 6:05 | 8:43 |
| Test Scripts | 2:16 | 2:10 |
| Test Web / Web 1 | 7:18 | 4:59 |
| Test Web 2 | n/a | 5:34 |
| Test Server 1 | 10:17 | 4:49 |
| Test Server 2 | 7:58 | 4:51 |
| Test Server 3 | 6:54 | 5:06 |
| Test Server 4 | n/a | 5:11 |
| Test Server 5 | n/a | 4:48 |
| Test Server 6 | n/a | 5:51 |
| merge-tree | 0:20 | 0:18 |
| CI | 0:03 | 0:03 |

GPT Sol via Pi.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: qol
Co-authored-by: donjor <[email protected]>

## Problem

After #1985, the pooled **`Test` job is the critical path** at **8:43** (run 41293089108).
Inside it, **`apps/mobile` ran 432 s** (260 files) and finished about 20 s after every other package; the next longest took ~157 s.

## Change

- **New `Test Mobile` job:** mobile tests only
- **Setup:** no Electron, no libsecret
- **`Test` drops `@t3tools/mobile`**
- **`Test` skips** on a scripts or mobile-only change
- **`CI` aggregate** needs `test_mobile`

**Gating:** `changes` gains a fork-only `test_mobile` output, derived in the workflow from `test` and `test_packages`.
The shared `ci-changes.cjs` stays upstream's, and the job skips without taking a runner; `CI` already tolerates a skip whose gate is `"false"`.

`test_mobile` per change, from `ci-changes.cjs` output:

| Change | `test_packages` | `test_mobile` | `Test` |
| --- | --- | --- | --- |
| `apps/web` only | `@t3tools/scripts` | false | skips |
| `apps/mobile` only | mobile, scripts | true | skips |
| `apps/server` only | evals, scripts | false | runs |
| `packages/contracts` | includes mobile | true | runs |
| Workflow file | empty (all) | true | runs |
| Docs only | empty, `test=false` | false | skipped |

## Scope and approval

Fork CI tuning only; follows #1985. No product code.

## Verification

- `actionlint` on the workflow → exit 0
- `vp run check:changed` → exit 0
- `vp run fork:ci` → exit 0

**Wall times** (before: run 41293089108; after: run 41296032295):

| Job | Before | After |
| --- | --- | --- |
| Test | 8:43 | 5:11 |
| Test Mobile | n/a | 3:05 |
| Slowest Test Server | 5:51 | 5:21 |
| End to end | ~9:00 | 7:37 |

- **Critical path:** Test Server 6, 2:03 queued
- **Runner cap:** 5 jobs queued 1:36 to 2:44
- **Test Server 2 red:** unrelated test time bomb

Claude Opus 5.5, Claude Code harness in T3 Code.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: qol
Co-authored-by: donjor <[email protected]>

… changes since the pin (#1991)

Closes #1987

## Close conditions

The three copies above:

- [ ] **Each copy** carries a pin
- [ ] **`fork:scan` fails** a stale pin
- [ ] **Failure names** the file and commits
- [ ] **Tests:** stale, current, missing pin
- [ ] **Fork-sync doc:** adapt or no-change step

## Problem

The `hyprws-*.yml` copies of upstream workflows drift silently: upstream `ci.yml` has 26 unmirrored commits.

## Change

- **Pin:** `# fork-mirror: <path>@<sha>` in each header
- **Stale pin:** `fork:scan --mirror-pins` fails it
- **No pin:** fails too
- **Sync tip only:** flag set when `rebased`
- **Ordinary PRs:** never fail on a stale pin
- **No ledger:** the pin lives in the copy
- **Docs:** `fork-sync.md` *Workflow copies*
- **Skill:** fork-sync tip-work bullet

The guard #1151 deleted stays deleted.

Scan of this head against the current base:

```text
$ node scripts/fork-scan.ts --no-typecheck --mirror-pins --target ae521d1 --since origin/hyprws
Workflow mirrors, 1 finding(s):
  MIRROR  .github/workflows/hyprws-ci.yml: upstream .github/workflows/ci.yml changed in 26 commit(s) since pin 5d14c0e; adapt the copy or record no change, then pin ae521d1
  f6d3700 ci: server test shards survive a throttled task cache (pingdotgg#26332)
  67661cf chore: delete 18k lines of code that nothing reaches (pingdotgg#26238)
  ...
exit 1
```

## Pins

| Copy | Pin | Adapted |
| --- | --- | --- |
| `hyprws-ci.yml` | `5d14c0e9686` | none |
| `hyprws-release.yml` | `ae521d1d796` | `e6ca17001d0` |
| `hyprws-agent-context-budget.yml` | `ae521d1d796` | `da1eab01f71` |

- **CI pin:** the last upstream commit it matched
- **CI drift:** the next sync reviews it
- **`e6ca17001d0`:** prerelease drafts until assets land
- **`da1eab01f71`:** error names the evals rerun

#1985 took upstream cache steps only, so the CI pin stays.

### Release: no change

| Upstream commits | Reason |
| --- | --- |
| `f41518f` | launch check: untested release behaviour |
| `30f4d63` `fe27e8e` `e407f9b` `25159b6` `fd1c338` | test jobs: the gate needs green CI |
| `a9418a1` `6327560` `a43b003` | task cache: no secret, `setup-vp` |
| `1688f96` `783ccf0` `7479625` `bd0b9ed` `e74c668` | deploys: the fork has none |
| `e64766c` `569bd02` `14b47ee` `e9a5974` `c57a04b` | macOS, Windows: Linux only |
| `c13f7d9` | `.deb`, arm64: AppImage only |
| `6b450c7` `91cd91c` | npm: not published |
| `83d6583` | mobile: no fork build |
| `b438198` | Discord: no announce |
| `b5fc7be` | OpenAI notes: generated notes kept |
| `3060cc4` | channel defaults: fork resolves its own |
| `bed15b9` | skip in forks: copy is fork-owned |
| `28c32d2` | telemetry: defaults to `dev` here |
| `eb8f6f4` `c7f23c4` `0754920` `2c54f2f` `2f7616e` | CLI archives: already built |

## Scope and approval

The maintainer filed #1987 and assigned this scope.

## Verification

- [x] **Typecheck** (scripts) → exit 0
- [x] **Tests** (3 files) → exit 0, 27 passed
- [x] **check:changed** → exit 1
- [x] **Its two failures:** web browser timeouts
- [x] **No web code** in this diff
- [x] **Both pass** run alone
- [x] **fork:ci** (bwrap HOME workaround) → exit 0

Claude Opus 5.5, Claude Code in T3 Code.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: core
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

## Problem

PR CI takes **10.4 minutes median** on GitHub-hosted runners; server and web tests share the critical path.

## Change

Reuse upstream's dependency, Chromium, and Electron cache steps without its private task-cache token or Blacksmith assumptions.

```diff
- Test Server: 3 shards
- Test Web: 1 job
+ Test Server: 6 shards
+ Test Web: 2 shards
```

Six server shards finish in **4:48–5:51**; two web shards finish in **4:59 / 5:34**.
Server test steps measure 209–273s (232s median), against the supplied three-shard median of 338s; web steps measure 256s / 265s, against 478s unsharded.
More web shards would not shorten this run: the unchanged pooled `Test` job is now the critical path at **8:43**, including 432s for mobile tests.
The approximate six-minute goal is not reached.
The workflow has 14 jobs including Changes and CI, below the org's 20 concurrent hosted-job limit.
The aggregate still depends on both matrix job IDs; ruleset 21768840 requires only `CI` and `merge-tree`, so no ruleset change is needed.

## Scope and approval

The maintainer explicitly approved this CI-only implementation and draft PR.
No `test:cached`, flaky-test fixes, ruleset changes, or upstream writes.

## Verification

- [x] Workflow passes actionlint
- [x] Browser/unit shard probe: 18 tests passed
- [x] Each probe shard runs one browser file
- [x] Each probe shard runs one unit file
- [x] No overlap across the four probe files
- [ ] `vp run fork:ci`: exit 1, Scripts failed

The committed-head preflight passed delta, stale-delete, fixup-fold, merge-tree, scan, knip, check, typecheck, and fork-tests (220 files, 1215 tests).
Scripts passed 596 tests and failed one unchanged Discord release workflow test at `scripts/notify-discord-release.test.ts:365`.
A subprocess trace shows the host shell fails before the fixture runs: `/home/user/.bashrc: line 255: path_prepend: command not found`, exit 127.
Full local web attempts also hit browser module/mock errors; the focused shard probe passed with one worker and Playwright's Ubuntu 24.04 x64 platform override on this Ubuntu 26.04 host.
### Hosted CI measurements

[Run 41293089108](https://github.com/RSI-Software/t3code-hyprws/actions/runs/41293089108) is **green on attempt 2**, with 16 checks passed and one unrelated preview check skipped.
Attempt 1 failed only the unchanged Mermaid SVG assertion; its full file passed 10/10 locally, and a governed job rerun passed all 276 shard files and 3677 tests without source changes.
The no-token dependency action succeeded; logs show its private task-cache step skipped, Chromium restored, and the ldd check took under one second.

The green run took **14:51 end-to-end including the failed attempt and rerun**; the rerun attempt itself took 5:08, reusing all other job results.
Attempt 1 took 9:01 but was red, so it is not a green-run speed claim.
The supplied historical median is 10:24; a concrete earlier green run, [41278753510](https://github.com/RSI-Software/t3code-hyprws/actions/runs/41278753510), took 11:50.
These are observations, not a controlled before/after benchmark or a new median.

Job wall times from the Actions jobs API:

| Job | Earlier green | This green run |
| --- | --- | --- |
| Changes | 0:05 | 0:07 |
| Check | 4:37 | 3:54 |
| Test | 6:05 | 8:43 |
| Test Scripts | 2:16 | 2:10 |
| Test Web / Web 1 | 7:18 | 4:59 |
| Test Web 2 | n/a | 5:34 |
| Test Server 1 | 10:17 | 4:49 |
| Test Server 2 | 7:58 | 4:51 |
| Test Server 3 | 6:54 | 5:06 |
| Test Server 4 | n/a | 5:11 |
| Test Server 5 | n/a | 4:48 |
| Test Server 6 | n/a | 5:51 |
| merge-tree | 0:20 | 0:18 |
| CI | 0:03 | 0:03 |

GPT Sol via Pi.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: qol
Co-authored-by: donjor <[email protected]>

## Problem

After #1985, the pooled **`Test` job is the critical path** at **8:43** (run 41293089108).
Inside it, **`apps/mobile` ran 432 s** (260 files) and finished about 20 s after every other package; the next longest took ~157 s.

## Change

- **New `Test Mobile` job:** mobile tests only
- **Setup:** no Electron, no libsecret
- **`Test` drops `@t3tools/mobile`**
- **`Test` skips** on a scripts or mobile-only change
- **`CI` aggregate** needs `test_mobile`

**Gating:** `changes` gains a fork-only `test_mobile` output, derived in the workflow from `test` and `test_packages`.
The shared `ci-changes.cjs` stays upstream's, and the job skips without taking a runner; `CI` already tolerates a skip whose gate is `"false"`.

`test_mobile` per change, from `ci-changes.cjs` output:

| Change | `test_packages` | `test_mobile` | `Test` |
| --- | --- | --- | --- |
| `apps/web` only | `@t3tools/scripts` | false | skips |
| `apps/mobile` only | mobile, scripts | true | skips |
| `apps/server` only | evals, scripts | false | runs |
| `packages/contracts` | includes mobile | true | runs |
| Workflow file | empty (all) | true | runs |
| Docs only | empty, `test=false` | false | skipped |

## Scope and approval

Fork CI tuning only; follows #1985. No product code.

## Verification

- `actionlint` on the workflow → exit 0
- `vp run check:changed` → exit 0
- `vp run fork:ci` → exit 0

**Wall times** (before: run 41293089108; after: run 41296032295):

| Job | Before | After |
| --- | --- | --- |
| Test | 8:43 | 5:11 |
| Test Mobile | n/a | 3:05 |
| Slowest Test Server | 5:51 | 5:21 |
| End to end | ~9:00 | 7:37 |

- **Critical path:** Test Server 6, 2:03 queued
- **Runner cap:** 5 jobs queued 1:36 to 2:44
- **Test Server 2 red:** unrelated test time bomb

Claude Opus 5.5, Claude Code harness in T3 Code.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: qol
Co-authored-by: donjor <[email protected]>

… changes since the pin (#1991)

Closes #1987

## Close conditions

The three copies above:

- [ ] **Each copy** carries a pin
- [ ] **`fork:scan` fails** a stale pin
- [ ] **Failure names** the file and commits
- [ ] **Tests:** stale, current, missing pin
- [ ] **Fork-sync doc:** adapt or no-change step

## Problem

The `hyprws-*.yml` copies of upstream workflows drift silently: upstream `ci.yml` has 26 unmirrored commits.

## Change

- **Pin:** `# fork-mirror: <path>@<sha>` in each header
- **Stale pin:** `fork:scan --mirror-pins` fails it
- **No pin:** fails too
- **Sync tip only:** flag set when `rebased`
- **Ordinary PRs:** never fail on a stale pin
- **No ledger:** the pin lives in the copy
- **Docs:** `fork-sync.md` *Workflow copies*
- **Skill:** fork-sync tip-work bullet

The guard #1151 deleted stays deleted.

Scan of this head against the current base:

```text
$ node scripts/fork-scan.ts --no-typecheck --mirror-pins --target ae521d1 --since origin/hyprws
Workflow mirrors, 1 finding(s):
  MIRROR  .github/workflows/hyprws-ci.yml: upstream .github/workflows/ci.yml changed in 26 commit(s) since pin 5d14c0e; adapt the copy or record no change, then pin ae521d1
  f6d3700 ci: server test shards survive a throttled task cache (pingdotgg#26332)
  67661cf chore: delete 18k lines of code that nothing reaches (pingdotgg#26238)
  ...
exit 1
```

## Pins

| Copy | Pin | Adapted |
| --- | --- | --- |
| `hyprws-ci.yml` | `5d14c0e9686` | none |
| `hyprws-release.yml` | `ae521d1d796` | `e6ca17001d0` |
| `hyprws-agent-context-budget.yml` | `ae521d1d796` | `da1eab01f71` |

- **CI pin:** the last upstream commit it matched
- **CI drift:** the next sync reviews it
- **`e6ca17001d0`:** prerelease drafts until assets land
- **`da1eab01f71`:** error names the evals rerun

#1985 took upstream cache steps only, so the CI pin stays.

### Release: no change

| Upstream commits | Reason |
| --- | --- |
| `f41518f` | launch check: untested release behaviour |
| `30f4d63` `fe27e8e` `e407f9b` `25159b6` `fd1c338` | test jobs: the gate needs green CI |
| `a9418a1` `6327560` `a43b003` | task cache: no secret, `setup-vp` |
| `1688f96` `783ccf0` `7479625` `bd0b9ed` `e74c668` | deploys: the fork has none |
| `e64766c` `569bd02` `14b47ee` `e9a5974` `c57a04b` | macOS, Windows: Linux only |
| `c13f7d9` | `.deb`, arm64: AppImage only |
| `6b450c7` `91cd91c` | npm: not published |
| `83d6583` | mobile: no fork build |
| `b438198` | Discord: no announce |
| `b5fc7be` | OpenAI notes: generated notes kept |
| `3060cc4` | channel defaults: fork resolves its own |
| `bed15b9` | skip in forks: copy is fork-owned |
| `28c32d2` | telemetry: defaults to `dev` here |
| `eb8f6f4` `c7f23c4` `0754920` `2c54f2f` `2f7616e` | CLI archives: already built |

## Scope and approval

The maintainer filed #1987 and assigned this scope.

## Verification

- [x] **Typecheck** (scripts) → exit 0
- [x] **Tests** (3 files) → exit 0, 27 passed
- [x] **check:changed** → exit 1
- [x] **Its two failures:** web browser timeouts
- [x] **No web code** in this diff
- [x] **Both pass** run alone
- [x] **fork:ci** (bwrap HOME workaround) → exit 0

Claude Opus 5.5, Claude Code in T3 Code.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: core
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

## Problem

PR CI takes **10.4 minutes median** on GitHub-hosted runners; server and web tests share the critical path.

## Change

Reuse upstream's dependency, Chromium, and Electron cache steps without its private task-cache token or Blacksmith assumptions.

```diff
- Test Server: 3 shards
- Test Web: 1 job
+ Test Server: 6 shards
+ Test Web: 2 shards
```

Six server shards finish in **4:48–5:51**; two web shards finish in **4:59 / 5:34**.
Server test steps measure 209–273s (232s median), against the supplied three-shard median of 338s; web steps measure 256s / 265s, against 478s unsharded.
More web shards would not shorten this run: the unchanged pooled `Test` job is now the critical path at **8:43**, including 432s for mobile tests.
The approximate six-minute goal is not reached.
The workflow has 14 jobs including Changes and CI, below the org's 20 concurrent hosted-job limit.
The aggregate still depends on both matrix job IDs; ruleset 21768840 requires only `CI` and `merge-tree`, so no ruleset change is needed.

## Scope and approval

The maintainer explicitly approved this CI-only implementation and draft PR.
No `test:cached`, flaky-test fixes, ruleset changes, or upstream writes.

## Verification

- [x] Workflow passes actionlint
- [x] Browser/unit shard probe: 18 tests passed
- [x] Each probe shard runs one browser file
- [x] Each probe shard runs one unit file
- [x] No overlap across the four probe files
- [ ] `vp run fork:ci`: exit 1, Scripts failed

The committed-head preflight passed delta, stale-delete, fixup-fold, merge-tree, scan, knip, check, typecheck, and fork-tests (220 files, 1215 tests).
Scripts passed 596 tests and failed one unchanged Discord release workflow test at `scripts/notify-discord-release.test.ts:365`.
A subprocess trace shows the host shell fails before the fixture runs: `/home/user/.bashrc: line 255: path_prepend: command not found`, exit 127.
Full local web attempts also hit browser module/mock errors; the focused shard probe passed with one worker and Playwright's Ubuntu 24.04 x64 platform override on this Ubuntu 26.04 host.
### Hosted CI measurements

[Run 41293089108](https://github.com/RSI-Software/t3code-hyprws/actions/runs/41293089108) is **green on attempt 2**, with 16 checks passed and one unrelated preview check skipped.
Attempt 1 failed only the unchanged Mermaid SVG assertion; its full file passed 10/10 locally, and a governed job rerun passed all 276 shard files and 3677 tests without source changes.
The no-token dependency action succeeded; logs show its private task-cache step skipped, Chromium restored, and the ldd check took under one second.

The green run took **14:51 end-to-end including the failed attempt and rerun**; the rerun attempt itself took 5:08, reusing all other job results.
Attempt 1 took 9:01 but was red, so it is not a green-run speed claim.
The supplied historical median is 10:24; a concrete earlier green run, [41278753510](https://github.com/RSI-Software/t3code-hyprws/actions/runs/41278753510), took 11:50.
These are observations, not a controlled before/after benchmark or a new median.

Job wall times from the Actions jobs API:

| Job | Earlier green | This green run |
| --- | --- | --- |
| Changes | 0:05 | 0:07 |
| Check | 4:37 | 3:54 |
| Test | 6:05 | 8:43 |
| Test Scripts | 2:16 | 2:10 |
| Test Web / Web 1 | 7:18 | 4:59 |
| Test Web 2 | n/a | 5:34 |
| Test Server 1 | 10:17 | 4:49 |
| Test Server 2 | 7:58 | 4:51 |
| Test Server 3 | 6:54 | 5:06 |
| Test Server 4 | n/a | 5:11 |
| Test Server 5 | n/a | 4:48 |
| Test Server 6 | n/a | 5:51 |
| merge-tree | 0:20 | 0:18 |
| CI | 0:03 | 0:03 |

GPT Sol via Pi.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: qol
Co-authored-by: donjor <[email protected]>

## Problem

After #1985, the pooled **`Test` job is the critical path** at **8:43** (run 41293089108).
Inside it, **`apps/mobile` ran 432 s** (260 files) and finished about 20 s after every other package; the next longest took ~157 s.

## Change

- **New `Test Mobile` job:** mobile tests only
- **Setup:** no Electron, no libsecret
- **`Test` drops `@t3tools/mobile`**
- **`Test` skips** on a scripts or mobile-only change
- **`CI` aggregate** needs `test_mobile`

**Gating:** `changes` gains a fork-only `test_mobile` output, derived in the workflow from `test` and `test_packages`.
The shared `ci-changes.cjs` stays upstream's, and the job skips without taking a runner; `CI` already tolerates a skip whose gate is `"false"`.

`test_mobile` per change, from `ci-changes.cjs` output:

| Change | `test_packages` | `test_mobile` | `Test` |
| --- | --- | --- | --- |
| `apps/web` only | `@t3tools/scripts` | false | skips |
| `apps/mobile` only | mobile, scripts | true | skips |
| `apps/server` only | evals, scripts | false | runs |
| `packages/contracts` | includes mobile | true | runs |
| Workflow file | empty (all) | true | runs |
| Docs only | empty, `test=false` | false | skipped |

## Scope and approval

Fork CI tuning only; follows #1985. No product code.

## Verification

- `actionlint` on the workflow → exit 0
- `vp run check:changed` → exit 0
- `vp run fork:ci` → exit 0

**Wall times** (before: run 41293089108; after: run 41296032295):

| Job | Before | After |
| --- | --- | --- |
| Test | 8:43 | 5:11 |
| Test Mobile | n/a | 3:05 |
| Slowest Test Server | 5:51 | 5:21 |
| End to end | ~9:00 | 7:37 |

- **Critical path:** Test Server 6, 2:03 queued
- **Runner cap:** 5 jobs queued 1:36 to 2:44
- **Test Server 2 red:** unrelated test time bomb

Claude Opus 5.5, Claude Code harness in T3 Code.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: qol
Co-authored-by: donjor <[email protected]>

… changes since the pin (#1991)

Closes #1987

## Close conditions

The three copies above:

- [ ] **Each copy** carries a pin
- [ ] **`fork:scan` fails** a stale pin
- [ ] **Failure names** the file and commits
- [ ] **Tests:** stale, current, missing pin
- [ ] **Fork-sync doc:** adapt or no-change step

## Problem

The `hyprws-*.yml` copies of upstream workflows drift silently: upstream `ci.yml` has 26 unmirrored commits.

## Change

- **Pin:** `# fork-mirror: <path>@<sha>` in each header
- **Stale pin:** `fork:scan --mirror-pins` fails it
- **No pin:** fails too
- **Sync tip only:** flag set when `rebased`
- **Ordinary PRs:** never fail on a stale pin
- **No ledger:** the pin lives in the copy
- **Docs:** `fork-sync.md` *Workflow copies*
- **Skill:** fork-sync tip-work bullet

The guard #1151 deleted stays deleted.

Scan of this head against the current base:

```text
$ node scripts/fork-scan.ts --no-typecheck --mirror-pins --target ae521d1 --since origin/hyprws
Workflow mirrors, 1 finding(s):
  MIRROR  .github/workflows/hyprws-ci.yml: upstream .github/workflows/ci.yml changed in 26 commit(s) since pin 5d14c0e; adapt the copy or record no change, then pin ae521d1
  f6d3700 ci: server test shards survive a throttled task cache (pingdotgg#26332)
  67661cf chore: delete 18k lines of code that nothing reaches (pingdotgg#26238)
  ...
exit 1
```

## Pins

| Copy | Pin | Adapted |
| --- | --- | --- |
| `hyprws-ci.yml` | `5d14c0e9686` | none |
| `hyprws-release.yml` | `ae521d1d796` | `e6ca17001d0` |
| `hyprws-agent-context-budget.yml` | `ae521d1d796` | `da1eab01f71` |

- **CI pin:** the last upstream commit it matched
- **CI drift:** the next sync reviews it
- **`e6ca17001d0`:** prerelease drafts until assets land
- **`da1eab01f71`:** error names the evals rerun

#1985 took upstream cache steps only, so the CI pin stays.

### Release: no change

| Upstream commits | Reason |
| --- | --- |
| `f41518f` | launch check: untested release behaviour |
| `30f4d63` `fe27e8e` `e407f9b` `25159b6` `fd1c338` | test jobs: the gate needs green CI |
| `a9418a1` `6327560` `a43b003` | task cache: no secret, `setup-vp` |
| `1688f96` `783ccf0` `7479625` `bd0b9ed` `e74c668` | deploys: the fork has none |
| `e64766c` `569bd02` `14b47ee` `e9a5974` `c57a04b` | macOS, Windows: Linux only |
| `c13f7d9` | `.deb`, arm64: AppImage only |
| `6b450c7` `91cd91c` | npm: not published |
| `83d6583` | mobile: no fork build |
| `b438198` | Discord: no announce |
| `b5fc7be` | OpenAI notes: generated notes kept |
| `3060cc4` | channel defaults: fork resolves its own |
| `bed15b9` | skip in forks: copy is fork-owned |
| `28c32d2` | telemetry: defaults to `dev` here |
| `eb8f6f4` `c7f23c4` `0754920` `2c54f2f` `2f7616e` | CLI archives: already built |

## Scope and approval

The maintainer filed #1987 and assigned this scope.

## Verification

- [x] **Typecheck** (scripts) → exit 0
- [x] **Tests** (3 files) → exit 0, 27 passed
- [x] **check:changed** → exit 1
- [x] **Its two failures:** web browser timeouts
- [x] **No web code** in this diff
- [x] **Both pass** run alone
- [x] **fork:ci** (bwrap HOME workaround) → exit 0

Claude Opus 5.5, Claude Code in T3 Code.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: core
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

## Problem

PR CI takes **10.4 minutes median** on GitHub-hosted runners; server and web tests share the critical path.

## Change

Reuse upstream's dependency, Chromium, and Electron cache steps without its private task-cache token or Blacksmith assumptions.

```diff
- Test Server: 3 shards
- Test Web: 1 job
+ Test Server: 6 shards
+ Test Web: 2 shards
```

Six server shards finish in **4:48–5:51**; two web shards finish in **4:59 / 5:34**.
Server test steps measure 209–273s (232s median), against the supplied three-shard median of 338s; web steps measure 256s / 265s, against 478s unsharded.
More web shards would not shorten this run: the unchanged pooled `Test` job is now the critical path at **8:43**, including 432s for mobile tests.
The approximate six-minute goal is not reached.
The workflow has 14 jobs including Changes and CI, below the org's 20 concurrent hosted-job limit.
The aggregate still depends on both matrix job IDs; ruleset 21768840 requires only `CI` and `merge-tree`, so no ruleset change is needed.

## Scope and approval

The maintainer explicitly approved this CI-only implementation and draft PR.
No `test:cached`, flaky-test fixes, ruleset changes, or upstream writes.

## Verification

- [x] Workflow passes actionlint
- [x] Browser/unit shard probe: 18 tests passed
- [x] Each probe shard runs one browser file
- [x] Each probe shard runs one unit file
- [x] No overlap across the four probe files
- [ ] `vp run fork:ci`: exit 1, Scripts failed

The committed-head preflight passed delta, stale-delete, fixup-fold, merge-tree, scan, knip, check, typecheck, and fork-tests (220 files, 1215 tests).
Scripts passed 596 tests and failed one unchanged Discord release workflow test at `scripts/notify-discord-release.test.ts:365`.
A subprocess trace shows the host shell fails before the fixture runs: `/home/user/.bashrc: line 255: path_prepend: command not found`, exit 127.
Full local web attempts also hit browser module/mock errors; the focused shard probe passed with one worker and Playwright's Ubuntu 24.04 x64 platform override on this Ubuntu 26.04 host.
### Hosted CI measurements

[Run 41293089108](https://github.com/RSI-Software/t3code-hyprws/actions/runs/41293089108) is **green on attempt 2**, with 16 checks passed and one unrelated preview check skipped.
Attempt 1 failed only the unchanged Mermaid SVG assertion; its full file passed 10/10 locally, and a governed job rerun passed all 276 shard files and 3677 tests without source changes.
The no-token dependency action succeeded; logs show its private task-cache step skipped, Chromium restored, and the ldd check took under one second.

The green run took **14:51 end-to-end including the failed attempt and rerun**; the rerun attempt itself took 5:08, reusing all other job results.
Attempt 1 took 9:01 but was red, so it is not a green-run speed claim.
The supplied historical median is 10:24; a concrete earlier green run, [41278753510](https://github.com/RSI-Software/t3code-hyprws/actions/runs/41278753510), took 11:50.
These are observations, not a controlled before/after benchmark or a new median.

Job wall times from the Actions jobs API:

| Job | Earlier green | This green run |
| --- | --- | --- |
| Changes | 0:05 | 0:07 |
| Check | 4:37 | 3:54 |
| Test | 6:05 | 8:43 |
| Test Scripts | 2:16 | 2:10 |
| Test Web / Web 1 | 7:18 | 4:59 |
| Test Web 2 | n/a | 5:34 |
| Test Server 1 | 10:17 | 4:49 |
| Test Server 2 | 7:58 | 4:51 |
| Test Server 3 | 6:54 | 5:06 |
| Test Server 4 | n/a | 5:11 |
| Test Server 5 | n/a | 4:48 |
| Test Server 6 | n/a | 5:51 |
| merge-tree | 0:20 | 0:18 |
| CI | 0:03 | 0:03 |

GPT Sol via Pi.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: qol
Co-authored-by: donjor <[email protected]>

## Problem

After #1985, the pooled **`Test` job is the critical path** at **8:43** (run 41293089108).
Inside it, **`apps/mobile` ran 432 s** (260 files) and finished about 20 s after every other package; the next longest took ~157 s.

## Change

- **New `Test Mobile` job:** mobile tests only
- **Setup:** no Electron, no libsecret
- **`Test` drops `@t3tools/mobile`**
- **`Test` skips** on a scripts or mobile-only change
- **`CI` aggregate** needs `test_mobile`

**Gating:** `changes` gains a fork-only `test_mobile` output, derived in the workflow from `test` and `test_packages`.
The shared `ci-changes.cjs` stays upstream's, and the job skips without taking a runner; `CI` already tolerates a skip whose gate is `"false"`.

`test_mobile` per change, from `ci-changes.cjs` output:

| Change | `test_packages` | `test_mobile` | `Test` |
| --- | --- | --- | --- |
| `apps/web` only | `@t3tools/scripts` | false | skips |
| `apps/mobile` only | mobile, scripts | true | skips |
| `apps/server` only | evals, scripts | false | runs |
| `packages/contracts` | includes mobile | true | runs |
| Workflow file | empty (all) | true | runs |
| Docs only | empty, `test=false` | false | skipped |

## Scope and approval

Fork CI tuning only; follows #1985. No product code.

## Verification

- `actionlint` on the workflow → exit 0
- `vp run check:changed` → exit 0
- `vp run fork:ci` → exit 0

**Wall times** (before: run 41293089108; after: run 41296032295):

| Job | Before | After |
| --- | --- | --- |
| Test | 8:43 | 5:11 |
| Test Mobile | n/a | 3:05 |
| Slowest Test Server | 5:51 | 5:21 |
| End to end | ~9:00 | 7:37 |

- **Critical path:** Test Server 6, 2:03 queued
- **Runner cap:** 5 jobs queued 1:36 to 2:44
- **Test Server 2 red:** unrelated test time bomb

Claude Opus 5.5, Claude Code harness in T3 Code.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: qol
Co-authored-by: donjor <[email protected]>

… changes since the pin (#1991)

Closes #1987

## Close conditions

The three copies above:

- [ ] **Each copy** carries a pin
- [ ] **`fork:scan` fails** a stale pin
- [ ] **Failure names** the file and commits
- [ ] **Tests:** stale, current, missing pin
- [ ] **Fork-sync doc:** adapt or no-change step

## Problem

The `hyprws-*.yml` copies of upstream workflows drift silently: upstream `ci.yml` has 26 unmirrored commits.

## Change

- **Pin:** `# fork-mirror: <path>@<sha>` in each header
- **Stale pin:** `fork:scan --mirror-pins` fails it
- **No pin:** fails too
- **Sync tip only:** flag set when `rebased`
- **Ordinary PRs:** never fail on a stale pin
- **No ledger:** the pin lives in the copy
- **Docs:** `fork-sync.md` *Workflow copies*
- **Skill:** fork-sync tip-work bullet

The guard #1151 deleted stays deleted.

Scan of this head against the current base:

```text
$ node scripts/fork-scan.ts --no-typecheck --mirror-pins --target ae521d1 --since origin/hyprws
Workflow mirrors, 1 finding(s):
  MIRROR  .github/workflows/hyprws-ci.yml: upstream .github/workflows/ci.yml changed in 26 commit(s) since pin 5d14c0e; adapt the copy or record no change, then pin ae521d1
  f6d3700 ci: server test shards survive a throttled task cache (pingdotgg#26332)
  67661cf chore: delete 18k lines of code that nothing reaches (pingdotgg#26238)
  ...
exit 1
```

## Pins

| Copy | Pin | Adapted |
| --- | --- | --- |
| `hyprws-ci.yml` | `5d14c0e9686` | none |
| `hyprws-release.yml` | `ae521d1d796` | `e6ca17001d0` |
| `hyprws-agent-context-budget.yml` | `ae521d1d796` | `da1eab01f71` |

- **CI pin:** the last upstream commit it matched
- **CI drift:** the next sync reviews it
- **`e6ca17001d0`:** prerelease drafts until assets land
- **`da1eab01f71`:** error names the evals rerun

#1985 took upstream cache steps only, so the CI pin stays.

### Release: no change

| Upstream commits | Reason |
| --- | --- |
| `f41518f` | launch check: untested release behaviour |
| `30f4d63` `fe27e8e` `e407f9b` `25159b6` `fd1c338` | test jobs: the gate needs green CI |
| `a9418a1` `6327560` `a43b003` | task cache: no secret, `setup-vp` |
| `1688f96` `783ccf0` `7479625` `bd0b9ed` `e74c668` | deploys: the fork has none |
| `e64766c` `569bd02` `14b47ee` `e9a5974` `c57a04b` | macOS, Windows: Linux only |
| `c13f7d9` | `.deb`, arm64: AppImage only |
| `6b450c7` `91cd91c` | npm: not published |
| `83d6583` | mobile: no fork build |
| `b438198` | Discord: no announce |
| `b5fc7be` | OpenAI notes: generated notes kept |
| `3060cc4` | channel defaults: fork resolves its own |
| `bed15b9` | skip in forks: copy is fork-owned |
| `28c32d2` | telemetry: defaults to `dev` here |
| `eb8f6f4` `c7f23c4` `0754920` `2c54f2f` `2f7616e` | CLI archives: already built |

## Scope and approval

The maintainer filed #1987 and assigned this scope.

## Verification

- [x] **Typecheck** (scripts) → exit 0
- [x] **Tests** (3 files) → exit 0, 27 passed
- [x] **check:changed** → exit 1
- [x] **Its two failures:** web browser timeouts
- [x] **No web code** in this diff
- [x] **Both pass** run alone
- [x] **fork:ci** (bwrap HOME workaround) → exit 0

Claude Opus 5.5, Claude Code in T3 Code.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: core
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

## Problem

PR CI takes **10.4 minutes median** on GitHub-hosted runners; server and web tests share the critical path.

## Change

Reuse upstream's dependency, Chromium, and Electron cache steps without its private task-cache token or Blacksmith assumptions.

```diff
- Test Server: 3 shards
- Test Web: 1 job
+ Test Server: 6 shards
+ Test Web: 2 shards
```

Six server shards finish in **4:48–5:51**; two web shards finish in **4:59 / 5:34**.
Server test steps measure 209–273s (232s median), against the supplied three-shard median of 338s; web steps measure 256s / 265s, against 478s unsharded.
More web shards would not shorten this run: the unchanged pooled `Test` job is now the critical path at **8:43**, including 432s for mobile tests.
The approximate six-minute goal is not reached.
The workflow has 14 jobs including Changes and CI, below the org's 20 concurrent hosted-job limit.
The aggregate still depends on both matrix job IDs; ruleset 21768840 requires only `CI` and `merge-tree`, so no ruleset change is needed.

## Scope and approval

The maintainer explicitly approved this CI-only implementation and draft PR.
No `test:cached`, flaky-test fixes, ruleset changes, or upstream writes.

## Verification

- [x] Workflow passes actionlint
- [x] Browser/unit shard probe: 18 tests passed
- [x] Each probe shard runs one browser file
- [x] Each probe shard runs one unit file
- [x] No overlap across the four probe files
- [ ] `vp run fork:ci`: exit 1, Scripts failed

The committed-head preflight passed delta, stale-delete, fixup-fold, merge-tree, scan, knip, check, typecheck, and fork-tests (220 files, 1215 tests).
Scripts passed 596 tests and failed one unchanged Discord release workflow test at `scripts/notify-discord-release.test.ts:365`.
A subprocess trace shows the host shell fails before the fixture runs: `/home/user/.bashrc: line 255: path_prepend: command not found`, exit 127.
Full local web attempts also hit browser module/mock errors; the focused shard probe passed with one worker and Playwright's Ubuntu 24.04 x64 platform override on this Ubuntu 26.04 host.
### Hosted CI measurements

[Run 41293089108](https://github.com/RSI-Software/t3code-hyprws/actions/runs/41293089108) is **green on attempt 2**, with 16 checks passed and one unrelated preview check skipped.
Attempt 1 failed only the unchanged Mermaid SVG assertion; its full file passed 10/10 locally, and a governed job rerun passed all 276 shard files and 3677 tests without source changes.
The no-token dependency action succeeded; logs show its private task-cache step skipped, Chromium restored, and the ldd check took under one second.

The green run took **14:51 end-to-end including the failed attempt and rerun**; the rerun attempt itself took 5:08, reusing all other job results.
Attempt 1 took 9:01 but was red, so it is not a green-run speed claim.
The supplied historical median is 10:24; a concrete earlier green run, [41278753510](https://github.com/RSI-Software/t3code-hyprws/actions/runs/41278753510), took 11:50.
These are observations, not a controlled before/after benchmark or a new median.

Job wall times from the Actions jobs API:

| Job | Earlier green | This green run |
| --- | --- | --- |
| Changes | 0:05 | 0:07 |
| Check | 4:37 | 3:54 |
| Test | 6:05 | 8:43 |
| Test Scripts | 2:16 | 2:10 |
| Test Web / Web 1 | 7:18 | 4:59 |
| Test Web 2 | n/a | 5:34 |
| Test Server 1 | 10:17 | 4:49 |
| Test Server 2 | 7:58 | 4:51 |
| Test Server 3 | 6:54 | 5:06 |
| Test Server 4 | n/a | 5:11 |
| Test Server 5 | n/a | 4:48 |
| Test Server 6 | n/a | 5:51 |
| merge-tree | 0:20 | 0:18 |
| CI | 0:03 | 0:03 |

GPT Sol via Pi.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: qol
Co-authored-by: donjor <[email protected]>

## Problem

After #1985, the pooled **`Test` job is the critical path** at **8:43** (run 41293089108).
Inside it, **`apps/mobile` ran 432 s** (260 files) and finished about 20 s after every other package; the next longest took ~157 s.

## Change

- **New `Test Mobile` job:** mobile tests only
- **Setup:** no Electron, no libsecret
- **`Test` drops `@t3tools/mobile`**
- **`Test` skips** on a scripts or mobile-only change
- **`CI` aggregate** needs `test_mobile`

**Gating:** `changes` gains a fork-only `test_mobile` output, derived in the workflow from `test` and `test_packages`.
The shared `ci-changes.cjs` stays upstream's, and the job skips without taking a runner; `CI` already tolerates a skip whose gate is `"false"`.

`test_mobile` per change, from `ci-changes.cjs` output:

| Change | `test_packages` | `test_mobile` | `Test` |
| --- | --- | --- | --- |
| `apps/web` only | `@t3tools/scripts` | false | skips |
| `apps/mobile` only | mobile, scripts | true | skips |
| `apps/server` only | evals, scripts | false | runs |
| `packages/contracts` | includes mobile | true | runs |
| Workflow file | empty (all) | true | runs |
| Docs only | empty, `test=false` | false | skipped |

## Scope and approval

Fork CI tuning only; follows #1985. No product code.

## Verification

- `actionlint` on the workflow → exit 0
- `vp run check:changed` → exit 0
- `vp run fork:ci` → exit 0

**Wall times** (before: run 41293089108; after: run 41296032295):

| Job | Before | After |
| --- | --- | --- |
| Test | 8:43 | 5:11 |
| Test Mobile | n/a | 3:05 |
| Slowest Test Server | 5:51 | 5:21 |
| End to end | ~9:00 | 7:37 |

- **Critical path:** Test Server 6, 2:03 queued
- **Runner cap:** 5 jobs queued 1:36 to 2:44
- **Test Server 2 red:** unrelated test time bomb

Claude Opus 5.5, Claude Code harness in T3 Code.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: qol
Co-authored-by: donjor <[email protected]>

… changes since the pin (#1991)

Closes #1987

## Close conditions

The three copies above:

- [ ] **Each copy** carries a pin
- [ ] **`fork:scan` fails** a stale pin
- [ ] **Failure names** the file and commits
- [ ] **Tests:** stale, current, missing pin
- [ ] **Fork-sync doc:** adapt or no-change step

## Problem

The `hyprws-*.yml` copies of upstream workflows drift silently: upstream `ci.yml` has 26 unmirrored commits.

## Change

- **Pin:** `# fork-mirror: <path>@<sha>` in each header
- **Stale pin:** `fork:scan --mirror-pins` fails it
- **No pin:** fails too
- **Sync tip only:** flag set when `rebased`
- **Ordinary PRs:** never fail on a stale pin
- **No ledger:** the pin lives in the copy
- **Docs:** `fork-sync.md` *Workflow copies*
- **Skill:** fork-sync tip-work bullet

The guard #1151 deleted stays deleted.

Scan of this head against the current base:

```text
$ node scripts/fork-scan.ts --no-typecheck --mirror-pins --target ae521d1 --since origin/hyprws
Workflow mirrors, 1 finding(s):
  MIRROR  .github/workflows/hyprws-ci.yml: upstream .github/workflows/ci.yml changed in 26 commit(s) since pin 5d14c0e; adapt the copy or record no change, then pin ae521d1
  f6d3700 ci: server test shards survive a throttled task cache (pingdotgg#26332)
  67661cf chore: delete 18k lines of code that nothing reaches (pingdotgg#26238)
  ...
exit 1
```

## Pins

| Copy | Pin | Adapted |
| --- | --- | --- |
| `hyprws-ci.yml` | `5d14c0e9686` | none |
| `hyprws-release.yml` | `ae521d1d796` | `e6ca17001d0` |
| `hyprws-agent-context-budget.yml` | `ae521d1d796` | `da1eab01f71` |

- **CI pin:** the last upstream commit it matched
- **CI drift:** the next sync reviews it
- **`e6ca17001d0`:** prerelease drafts until assets land
- **`da1eab01f71`:** error names the evals rerun

#1985 took upstream cache steps only, so the CI pin stays.

### Release: no change

| Upstream commits | Reason |
| --- | --- |
| `f41518f` | launch check: untested release behaviour |
| `30f4d63` `fe27e8e` `e407f9b` `25159b6` `fd1c338` | test jobs: the gate needs green CI |
| `a9418a1` `6327560` `a43b003` | task cache: no secret, `setup-vp` |
| `1688f96` `783ccf0` `7479625` `bd0b9ed` `e74c668` | deploys: the fork has none |
| `e64766c` `569bd02` `14b47ee` `e9a5974` `c57a04b` | macOS, Windows: Linux only |
| `c13f7d9` | `.deb`, arm64: AppImage only |
| `6b450c7` `91cd91c` | npm: not published |
| `83d6583` | mobile: no fork build |
| `b438198` | Discord: no announce |
| `b5fc7be` | OpenAI notes: generated notes kept |
| `3060cc4` | channel defaults: fork resolves its own |
| `bed15b9` | skip in forks: copy is fork-owned |
| `28c32d2` | telemetry: defaults to `dev` here |
| `eb8f6f4` `c7f23c4` `0754920` `2c54f2f` `2f7616e` | CLI archives: already built |

## Scope and approval

The maintainer filed #1987 and assigned this scope.

## Verification

- [x] **Typecheck** (scripts) → exit 0
- [x] **Tests** (3 files) → exit 0, 27 passed
- [x] **check:changed** → exit 1
- [x] **Its two failures:** web browser timeouts
- [x] **No web code** in this diff
- [x] **Both pass** run alone
- [x] **fork:ci** (bwrap HOME workaround) → exit 0

Claude Opus 5.5, Claude Code in T3 Code.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: core
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

## Problem

PR CI takes **10.4 minutes median** on GitHub-hosted runners; server and web tests share the critical path.

## Change

Reuse upstream's dependency, Chromium, and Electron cache steps without its private task-cache token or Blacksmith assumptions.

```diff
- Test Server: 3 shards
- Test Web: 1 job
+ Test Server: 6 shards
+ Test Web: 2 shards
```

Six server shards finish in **4:48–5:51**; two web shards finish in **4:59 / 5:34**.
Server test steps measure 209–273s (232s median), against the supplied three-shard median of 338s; web steps measure 256s / 265s, against 478s unsharded.
More web shards would not shorten this run: the unchanged pooled `Test` job is now the critical path at **8:43**, including 432s for mobile tests.
The approximate six-minute goal is not reached.
The workflow has 14 jobs including Changes and CI, below the org's 20 concurrent hosted-job limit.
The aggregate still depends on both matrix job IDs; ruleset 21768840 requires only `CI` and `merge-tree`, so no ruleset change is needed.

## Scope and approval

The maintainer explicitly approved this CI-only implementation and draft PR.
No `test:cached`, flaky-test fixes, ruleset changes, or upstream writes.

## Verification

- [x] Workflow passes actionlint
- [x] Browser/unit shard probe: 18 tests passed
- [x] Each probe shard runs one browser file
- [x] Each probe shard runs one unit file
- [x] No overlap across the four probe files
- [ ] `vp run fork:ci`: exit 1, Scripts failed

The committed-head preflight passed delta, stale-delete, fixup-fold, merge-tree, scan, knip, check, typecheck, and fork-tests (220 files, 1215 tests).
Scripts passed 596 tests and failed one unchanged Discord release workflow test at `scripts/notify-discord-release.test.ts:365`.
A subprocess trace shows the host shell fails before the fixture runs: `/home/user/.bashrc: line 255: path_prepend: command not found`, exit 127.
Full local web attempts also hit browser module/mock errors; the focused shard probe passed with one worker and Playwright's Ubuntu 24.04 x64 platform override on this Ubuntu 26.04 host.
### Hosted CI measurements

[Run 41293089108](https://github.com/RSI-Software/t3code-hyprws/actions/runs/41293089108) is **green on attempt 2**, with 16 checks passed and one unrelated preview check skipped.
Attempt 1 failed only the unchanged Mermaid SVG assertion; its full file passed 10/10 locally, and a governed job rerun passed all 276 shard files and 3677 tests without source changes.
The no-token dependency action succeeded; logs show its private task-cache step skipped, Chromium restored, and the ldd check took under one second.

The green run took **14:51 end-to-end including the failed attempt and rerun**; the rerun attempt itself took 5:08, reusing all other job results.
Attempt 1 took 9:01 but was red, so it is not a green-run speed claim.
The supplied historical median is 10:24; a concrete earlier green run, [41278753510](https://github.com/RSI-Software/t3code-hyprws/actions/runs/41278753510), took 11:50.
These are observations, not a controlled before/after benchmark or a new median.

Job wall times from the Actions jobs API:

| Job | Earlier green | This green run |
| --- | --- | --- |
| Changes | 0:05 | 0:07 |
| Check | 4:37 | 3:54 |
| Test | 6:05 | 8:43 |
| Test Scripts | 2:16 | 2:10 |
| Test Web / Web 1 | 7:18 | 4:59 |
| Test Web 2 | n/a | 5:34 |
| Test Server 1 | 10:17 | 4:49 |
| Test Server 2 | 7:58 | 4:51 |
| Test Server 3 | 6:54 | 5:06 |
| Test Server 4 | n/a | 5:11 |
| Test Server 5 | n/a | 4:48 |
| Test Server 6 | n/a | 5:51 |
| merge-tree | 0:20 | 0:18 |
| CI | 0:03 | 0:03 |

GPT Sol via Pi.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: qol
Co-authored-by: donjor <[email protected]>

## Problem

After #1985, the pooled **`Test` job is the critical path** at **8:43** (run 41293089108).
Inside it, **`apps/mobile` ran 432 s** (260 files) and finished about 20 s after every other package; the next longest took ~157 s.

## Change

- **New `Test Mobile` job:** mobile tests only
- **Setup:** no Electron, no libsecret
- **`Test` drops `@t3tools/mobile`**
- **`Test` skips** on a scripts or mobile-only change
- **`CI` aggregate** needs `test_mobile`

**Gating:** `changes` gains a fork-only `test_mobile` output, derived in the workflow from `test` and `test_packages`.
The shared `ci-changes.cjs` stays upstream's, and the job skips without taking a runner; `CI` already tolerates a skip whose gate is `"false"`.

`test_mobile` per change, from `ci-changes.cjs` output:

| Change | `test_packages` | `test_mobile` | `Test` |
| --- | --- | --- | --- |
| `apps/web` only | `@t3tools/scripts` | false | skips |
| `apps/mobile` only | mobile, scripts | true | skips |
| `apps/server` only | evals, scripts | false | runs |
| `packages/contracts` | includes mobile | true | runs |
| Workflow file | empty (all) | true | runs |
| Docs only | empty, `test=false` | false | skipped |

## Scope and approval

Fork CI tuning only; follows #1985. No product code.

## Verification

- `actionlint` on the workflow → exit 0
- `vp run check:changed` → exit 0
- `vp run fork:ci` → exit 0

**Wall times** (before: run 41293089108; after: run 41296032295):

| Job | Before | After |
| --- | --- | --- |
| Test | 8:43 | 5:11 |
| Test Mobile | n/a | 3:05 |
| Slowest Test Server | 5:51 | 5:21 |
| End to end | ~9:00 | 7:37 |

- **Critical path:** Test Server 6, 2:03 queued
- **Runner cap:** 5 jobs queued 1:36 to 2:44
- **Test Server 2 red:** unrelated test time bomb

Claude Opus 5.5, Claude Code harness in T3 Code.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: qol
Co-authored-by: donjor <[email protected]>

… changes since the pin (#1991)

Closes #1987

## Close conditions

The three copies above:

- [ ] **Each copy** carries a pin
- [ ] **`fork:scan` fails** a stale pin
- [ ] **Failure names** the file and commits
- [ ] **Tests:** stale, current, missing pin
- [ ] **Fork-sync doc:** adapt or no-change step

## Problem

The `hyprws-*.yml` copies of upstream workflows drift silently: upstream `ci.yml` has 26 unmirrored commits.

## Change

- **Pin:** `# fork-mirror: <path>@<sha>` in each header
- **Stale pin:** `fork:scan --mirror-pins` fails it
- **No pin:** fails too
- **Sync tip only:** flag set when `rebased`
- **Ordinary PRs:** never fail on a stale pin
- **No ledger:** the pin lives in the copy
- **Docs:** `fork-sync.md` *Workflow copies*
- **Skill:** fork-sync tip-work bullet

The guard #1151 deleted stays deleted.

Scan of this head against the current base:

```text
$ node scripts/fork-scan.ts --no-typecheck --mirror-pins --target ae521d1 --since origin/hyprws
Workflow mirrors, 1 finding(s):
  MIRROR  .github/workflows/hyprws-ci.yml: upstream .github/workflows/ci.yml changed in 26 commit(s) since pin 5d14c0e; adapt the copy or record no change, then pin ae521d1
  f6d3700 ci: server test shards survive a throttled task cache (pingdotgg#26332)
  67661cf chore: delete 18k lines of code that nothing reaches (pingdotgg#26238)
  ...
exit 1
```

## Pins

| Copy | Pin | Adapted |
| --- | --- | --- |
| `hyprws-ci.yml` | `5d14c0e9686` | none |
| `hyprws-release.yml` | `ae521d1d796` | `e6ca17001d0` |
| `hyprws-agent-context-budget.yml` | `ae521d1d796` | `da1eab01f71` |

- **CI pin:** the last upstream commit it matched
- **CI drift:** the next sync reviews it
- **`e6ca17001d0`:** prerelease drafts until assets land
- **`da1eab01f71`:** error names the evals rerun

#1985 took upstream cache steps only, so the CI pin stays.

### Release: no change

| Upstream commits | Reason |
| --- | --- |
| `f41518f` | launch check: untested release behaviour |
| `30f4d63` `fe27e8e` `e407f9b` `25159b6` `fd1c338` | test jobs: the gate needs green CI |
| `a9418a1` `6327560` `a43b003` | task cache: no secret, `setup-vp` |
| `1688f96` `783ccf0` `7479625` `bd0b9ed` `e74c668` | deploys: the fork has none |
| `e64766c` `569bd02` `14b47ee` `e9a5974` `c57a04b` | macOS, Windows: Linux only |
| `c13f7d9` | `.deb`, arm64: AppImage only |
| `6b450c7` `91cd91c` | npm: not published |
| `83d6583` | mobile: no fork build |
| `b438198` | Discord: no announce |
| `b5fc7be` | OpenAI notes: generated notes kept |
| `3060cc4` | channel defaults: fork resolves its own |
| `bed15b9` | skip in forks: copy is fork-owned |
| `28c32d2` | telemetry: defaults to `dev` here |
| `eb8f6f4` `c7f23c4` `0754920` `2c54f2f` `2f7616e` | CLI archives: already built |

## Scope and approval

The maintainer filed #1987 and assigned this scope.

## Verification

- [x] **Typecheck** (scripts) → exit 0
- [x] **Tests** (3 files) → exit 0, 27 passed
- [x] **check:changed** → exit 1
- [x] **Its two failures:** web browser timeouts
- [x] **No web code** in this diff
- [x] **Both pass** run alone
- [x] **fork:ci** (bwrap HOME workaround) → exit 0

Claude Opus 5.5, Claude Code in T3 Code.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: core
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

## Problem

PR CI takes **10.4 minutes median** on GitHub-hosted runners; server and web tests share the critical path.

## Change

Reuse upstream's dependency, Chromium, and Electron cache steps without its private task-cache token or Blacksmith assumptions.

```diff
- Test Server: 3 shards
- Test Web: 1 job
+ Test Server: 6 shards
+ Test Web: 2 shards
```

Six server shards finish in **4:48–5:51**; two web shards finish in **4:59 / 5:34**.
Server test steps measure 209–273s (232s median), against the supplied three-shard median of 338s; web steps measure 256s / 265s, against 478s unsharded.
More web shards would not shorten this run: the unchanged pooled `Test` job is now the critical path at **8:43**, including 432s for mobile tests.
The approximate six-minute goal is not reached.
The workflow has 14 jobs including Changes and CI, below the org's 20 concurrent hosted-job limit.
The aggregate still depends on both matrix job IDs; ruleset 21768840 requires only `CI` and `merge-tree`, so no ruleset change is needed.

## Scope and approval

The maintainer explicitly approved this CI-only implementation and draft PR.
No `test:cached`, flaky-test fixes, ruleset changes, or upstream writes.

## Verification

- [x] Workflow passes actionlint
- [x] Browser/unit shard probe: 18 tests passed
- [x] Each probe shard runs one browser file
- [x] Each probe shard runs one unit file
- [x] No overlap across the four probe files
- [ ] `vp run fork:ci`: exit 1, Scripts failed

The committed-head preflight passed delta, stale-delete, fixup-fold, merge-tree, scan, knip, check, typecheck, and fork-tests (220 files, 1215 tests).
Scripts passed 596 tests and failed one unchanged Discord release workflow test at `scripts/notify-discord-release.test.ts:365`.
A subprocess trace shows the host shell fails before the fixture runs: `/home/user/.bashrc: line 255: path_prepend: command not found`, exit 127.
Full local web attempts also hit browser module/mock errors; the focused shard probe passed with one worker and Playwright's Ubuntu 24.04 x64 platform override on this Ubuntu 26.04 host.
### Hosted CI measurements

[Run 41293089108](https://github.com/RSI-Software/t3code-hyprws/actions/runs/41293089108) is **green on attempt 2**, with 16 checks passed and one unrelated preview check skipped.
Attempt 1 failed only the unchanged Mermaid SVG assertion; its full file passed 10/10 locally, and a governed job rerun passed all 276 shard files and 3677 tests without source changes.
The no-token dependency action succeeded; logs show its private task-cache step skipped, Chromium restored, and the ldd check took under one second.

The green run took **14:51 end-to-end including the failed attempt and rerun**; the rerun attempt itself took 5:08, reusing all other job results.
Attempt 1 took 9:01 but was red, so it is not a green-run speed claim.
The supplied historical median is 10:24; a concrete earlier green run, [41278753510](https://github.com/RSI-Software/t3code-hyprws/actions/runs/41278753510), took 11:50.
These are observations, not a controlled before/after benchmark or a new median.

Job wall times from the Actions jobs API:

| Job | Earlier green | This green run |
| --- | --- | --- |
| Changes | 0:05 | 0:07 |
| Check | 4:37 | 3:54 |
| Test | 6:05 | 8:43 |
| Test Scripts | 2:16 | 2:10 |
| Test Web / Web 1 | 7:18 | 4:59 |
| Test Web 2 | n/a | 5:34 |
| Test Server 1 | 10:17 | 4:49 |
| Test Server 2 | 7:58 | 4:51 |
| Test Server 3 | 6:54 | 5:06 |
| Test Server 4 | n/a | 5:11 |
| Test Server 5 | n/a | 4:48 |
| Test Server 6 | n/a | 5:51 |
| merge-tree | 0:20 | 0:18 |
| CI | 0:03 | 0:03 |

GPT Sol via Pi.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: qol
Co-authored-by: donjor <[email protected]>

… changes since the pin (#1991)

Closes #1987

## Close conditions

The three copies above:

- [ ] **Each copy** carries a pin
- [ ] **`fork:scan` fails** a stale pin
- [ ] **Failure names** the file and commits
- [ ] **Tests:** stale, current, missing pin
- [ ] **Fork-sync doc:** adapt or no-change step

## Problem

The `hyprws-*.yml` copies of upstream workflows drift silently: upstream `ci.yml` has 26 unmirrored commits.

## Change

- **Pin:** `# fork-mirror: <path>@<sha>` in each header
- **Stale pin:** `fork:scan --mirror-pins` fails it
- **No pin:** fails too
- **Sync tip only:** flag set when `rebased`
- **Ordinary PRs:** never fail on a stale pin
- **No ledger:** the pin lives in the copy
- **Docs:** `fork-sync.md` *Workflow copies*
- **Skill:** fork-sync tip-work bullet

The guard #1151 deleted stays deleted.

Scan of this head against the current base:

```text
$ node scripts/fork-scan.ts --no-typecheck --mirror-pins --target ae521d1 --since origin/hyprws
Workflow mirrors, 1 finding(s):
  MIRROR  .github/workflows/hyprws-ci.yml: upstream .github/workflows/ci.yml changed in 26 commit(s) since pin 5d14c0e; adapt the copy or record no change, then pin ae521d1
  f6d3700 ci: server test shards survive a throttled task cache (pingdotgg#26332)
  67661cf chore: delete 18k lines of code that nothing reaches (pingdotgg#26238)
  ...
exit 1
```

## Pins

| Copy | Pin | Adapted |
| --- | --- | --- |
| `hyprws-ci.yml` | `5d14c0e9686` | none |
| `hyprws-release.yml` | `ae521d1d796` | `e6ca17001d0` |
| `hyprws-agent-context-budget.yml` | `ae521d1d796` | `da1eab01f71` |

- **CI pin:** the last upstream commit it matched
- **CI drift:** the next sync reviews it
- **`e6ca17001d0`:** prerelease drafts until assets land
- **`da1eab01f71`:** error names the evals rerun

#1985 took upstream cache steps only, so the CI pin stays.

### Release: no change

| Upstream commits | Reason |
| --- | --- |
| `f41518f` | launch check: untested release behaviour |
| `30f4d63` `fe27e8e` `e407f9b` `25159b6` `fd1c338` | test jobs: the gate needs green CI |
| `a9418a1` `6327560` `a43b003` | task cache: no secret, `setup-vp` |
| `1688f96` `783ccf0` `7479625` `bd0b9ed` `e74c668` | deploys: the fork has none |
| `e64766c` `569bd02` `14b47ee` `e9a5974` `c57a04b` | macOS, Windows: Linux only |
| `c13f7d9` | `.deb`, arm64: AppImage only |
| `6b450c7` `91cd91c` | npm: not published |
| `83d6583` | mobile: no fork build |
| `b438198` | Discord: no announce |
| `b5fc7be` | OpenAI notes: generated notes kept |
| `3060cc4` | channel defaults: fork resolves its own |
| `bed15b9` | skip in forks: copy is fork-owned |
| `28c32d2` | telemetry: defaults to `dev` here |
| `eb8f6f4` `c7f23c4` `0754920` `2c54f2f` `2f7616e` | CLI archives: already built |

## Scope and approval

The maintainer filed #1987 and assigned this scope.

## Verification

- [x] **Typecheck** (scripts) → exit 0
- [x] **Tests** (3 files) → exit 0, 27 passed
- [x] **check:changed** → exit 1
- [x] **Its two failures:** web browser timeouts
- [x] **No web code** in this diff
- [x] **Both pass** run alone
- [x] **fork:ci** (bwrap HOME workaround) → exit 0

Claude Opus 5.5, Claude Code in T3 Code.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: core
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

## Problem

PR CI takes **10.4 minutes median** on GitHub-hosted runners; server and web tests share the critical path.

## Change

Reuse upstream's dependency, Chromium, and Electron cache steps without its private task-cache token or Blacksmith assumptions.

```diff
- Test Server: 3 shards
- Test Web: 1 job
+ Test Server: 6 shards
+ Test Web: 2 shards
```

Six server shards finish in **4:48–5:51**; two web shards finish in **4:59 / 5:34**.
Server test steps measure 209–273s (232s median), against the supplied three-shard median of 338s; web steps measure 256s / 265s, against 478s unsharded.
More web shards would not shorten this run: the unchanged pooled `Test` job is now the critical path at **8:43**, including 432s for mobile tests.
The approximate six-minute goal is not reached.
The workflow has 14 jobs including Changes and CI, below the org's 20 concurrent hosted-job limit.
The aggregate still depends on both matrix job IDs; ruleset 21768840 requires only `CI` and `merge-tree`, so no ruleset change is needed.

## Scope and approval

The maintainer explicitly approved this CI-only implementation and draft PR.
No `test:cached`, flaky-test fixes, ruleset changes, or upstream writes.

## Verification

- [x] Workflow passes actionlint
- [x] Browser/unit shard probe: 18 tests passed
- [x] Each probe shard runs one browser file
- [x] Each probe shard runs one unit file
- [x] No overlap across the four probe files
- [ ] `vp run fork:ci`: exit 1, Scripts failed

The committed-head preflight passed delta, stale-delete, fixup-fold, merge-tree, scan, knip, check, typecheck, and fork-tests (220 files, 1215 tests).
Scripts passed 596 tests and failed one unchanged Discord release workflow test at `scripts/notify-discord-release.test.ts:365`.
A subprocess trace shows the host shell fails before the fixture runs: `/home/user/.bashrc: line 255: path_prepend: command not found`, exit 127.
Full local web attempts also hit browser module/mock errors; the focused shard probe passed with one worker and Playwright's Ubuntu 24.04 x64 platform override on this Ubuntu 26.04 host.
### Hosted CI measurements

[Run 41293089108](https://github.com/RSI-Software/t3code-hyprws/actions/runs/41293089108) is **green on attempt 2**, with 16 checks passed and one unrelated preview check skipped.
Attempt 1 failed only the unchanged Mermaid SVG assertion; its full file passed 10/10 locally, and a governed job rerun passed all 276 shard files and 3677 tests without source changes.
The no-token dependency action succeeded; logs show its private task-cache step skipped, Chromium restored, and the ldd check took under one second.

The green run took **14:51 end-to-end including the failed attempt and rerun**; the rerun attempt itself took 5:08, reusing all other job results.
Attempt 1 took 9:01 but was red, so it is not a green-run speed claim.
The supplied historical median is 10:24; a concrete earlier green run, [41278753510](https://github.com/RSI-Software/t3code-hyprws/actions/runs/41278753510), took 11:50.
These are observations, not a controlled before/after benchmark or a new median.

Job wall times from the Actions jobs API:

| Job | Earlier green | This green run |
| --- | --- | --- |
| Changes | 0:05 | 0:07 |
| Check | 4:37 | 3:54 |
| Test | 6:05 | 8:43 |
| Test Scripts | 2:16 | 2:10 |
| Test Web / Web 1 | 7:18 | 4:59 |
| Test Web 2 | n/a | 5:34 |
| Test Server 1 | 10:17 | 4:49 |
| Test Server 2 | 7:58 | 4:51 |
| Test Server 3 | 6:54 | 5:06 |
| Test Server 4 | n/a | 5:11 |
| Test Server 5 | n/a | 4:48 |
| Test Server 6 | n/a | 5:51 |
| merge-tree | 0:20 | 0:18 |
| CI | 0:03 | 0:03 |

GPT Sol via Pi.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: qol
Co-authored-by: donjor <[email protected]>

## Problem

After #1985, the pooled **`Test` job is the critical path** at **8:43** (run 41293089108).
Inside it, **`apps/mobile` ran 432 s** (260 files) and finished about 20 s after every other package; the next longest took ~157 s.

## Change

- **New `Test Mobile` job:** mobile tests only
- **Setup:** no Electron, no libsecret
- **`Test` drops `@t3tools/mobile`**
- **`Test` skips** on a scripts or mobile-only change
- **`CI` aggregate** needs `test_mobile`

**Gating:** `changes` gains a fork-only `test_mobile` output, derived in the workflow from `test` and `test_packages`.
The shared `ci-changes.cjs` stays upstream's, and the job skips without taking a runner; `CI` already tolerates a skip whose gate is `"false"`.

`test_mobile` per change, from `ci-changes.cjs` output:

| Change | `test_packages` | `test_mobile` | `Test` |
| --- | --- | --- | --- |
| `apps/web` only | `@t3tools/scripts` | false | skips |
| `apps/mobile` only | mobile, scripts | true | skips |
| `apps/server` only | evals, scripts | false | runs |
| `packages/contracts` | includes mobile | true | runs |
| Workflow file | empty (all) | true | runs |
| Docs only | empty, `test=false` | false | skipped |

## Scope and approval

Fork CI tuning only; follows #1985. No product code.

## Verification

- `actionlint` on the workflow → exit 0
- `vp run check:changed` → exit 0
- `vp run fork:ci` → exit 0

**Wall times** (before: run 41293089108; after: run 41296032295):

| Job | Before | After |
| --- | --- | --- |
| Test | 8:43 | 5:11 |
| Test Mobile | n/a | 3:05 |
| Slowest Test Server | 5:51 | 5:21 |
| End to end | ~9:00 | 7:37 |

- **Critical path:** Test Server 6, 2:03 queued
- **Runner cap:** 5 jobs queued 1:36 to 2:44
- **Test Server 2 red:** unrelated test time bomb

Claude Opus 5.5, Claude Code harness in T3 Code.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: qol
Co-authored-by: donjor <[email protected]>

… changes since the pin (#1991)

Closes #1987

## Close conditions

The three copies above:

- [ ] **Each copy** carries a pin
- [ ] **`fork:scan` fails** a stale pin
- [ ] **Failure names** the file and commits
- [ ] **Tests:** stale, current, missing pin
- [ ] **Fork-sync doc:** adapt or no-change step

## Problem

The `hyprws-*.yml` copies of upstream workflows drift silently: upstream `ci.yml` has 26 unmirrored commits.

## Change

- **Pin:** `# fork-mirror: <path>@<sha>` in each header
- **Stale pin:** `fork:scan --mirror-pins` fails it
- **No pin:** fails too
- **Sync tip only:** flag set when `rebased`
- **Ordinary PRs:** never fail on a stale pin
- **No ledger:** the pin lives in the copy
- **Docs:** `fork-sync.md` *Workflow copies*
- **Skill:** fork-sync tip-work bullet

The guard #1151 deleted stays deleted.

Scan of this head against the current base:

```text
$ node scripts/fork-scan.ts --no-typecheck --mirror-pins --target ae521d1 --since origin/hyprws
Workflow mirrors, 1 finding(s):
  MIRROR  .github/workflows/hyprws-ci.yml: upstream .github/workflows/ci.yml changed in 26 commit(s) since pin 5d14c0e; adapt the copy or record no change, then pin ae521d1
  f6d3700 ci: server test shards survive a throttled task cache (pingdotgg#26332)
  67661cf chore: delete 18k lines of code that nothing reaches (pingdotgg#26238)
  ...
exit 1
```

## Pins

| Copy | Pin | Adapted |
| --- | --- | --- |
| `hyprws-ci.yml` | `5d14c0e9686` | none |
| `hyprws-release.yml` | `ae521d1d796` | `e6ca17001d0` |
| `hyprws-agent-context-budget.yml` | `ae521d1d796` | `da1eab01f71` |

- **CI pin:** the last upstream commit it matched
- **CI drift:** the next sync reviews it
- **`e6ca17001d0`:** prerelease drafts until assets land
- **`da1eab01f71`:** error names the evals rerun

#1985 took upstream cache steps only, so the CI pin stays.

### Release: no change

| Upstream commits | Reason |
| --- | --- |
| `f41518f` | launch check: untested release behaviour |
| `30f4d63` `fe27e8e` `e407f9b` `25159b6` `fd1c338` | test jobs: the gate needs green CI |
| `a9418a1` `6327560` `a43b003` | task cache: no secret, `setup-vp` |
| `1688f96` `783ccf0` `7479625` `bd0b9ed` `e74c668` | deploys: the fork has none |
| `e64766c` `569bd02` `14b47ee` `e9a5974` `c57a04b` | macOS, Windows: Linux only |
| `c13f7d9` | `.deb`, arm64: AppImage only |
| `6b450c7` `91cd91c` | npm: not published |
| `83d6583` | mobile: no fork build |
| `b438198` | Discord: no announce |
| `b5fc7be` | OpenAI notes: generated notes kept |
| `3060cc4` | channel defaults: fork resolves its own |
| `bed15b9` | skip in forks: copy is fork-owned |
| `28c32d2` | telemetry: defaults to `dev` here |
| `eb8f6f4` `c7f23c4` `0754920` `2c54f2f` `2f7616e` | CLI archives: already built |

## Scope and approval

The maintainer filed #1987 and assigned this scope.

## Verification

- [x] **Typecheck** (scripts) → exit 0
- [x] **Tests** (3 files) → exit 0, 27 passed
- [x] **check:changed** → exit 1
- [x] **Its two failures:** web browser timeouts
- [x] **No web code** in this diff
- [x] **Both pass** run alone
- [x] **fork:ci** (bwrap HOME workaround) → exit 0

Claude Opus 5.5, Claude Code in T3 Code.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: core
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

## Problem

PR CI takes **10.4 minutes median** on GitHub-hosted runners; server and web tests share the critical path.

## Change

Reuse upstream's dependency, Chromium, and Electron cache steps without its private task-cache token or Blacksmith assumptions.

```diff
- Test Server: 3 shards
- Test Web: 1 job
+ Test Server: 6 shards
+ Test Web: 2 shards
```

Six server shards finish in **4:48–5:51**; two web shards finish in **4:59 / 5:34**.
Server test steps measure 209–273s (232s median), against the supplied three-shard median of 338s; web steps measure 256s / 265s, against 478s unsharded.
More web shards would not shorten this run: the unchanged pooled `Test` job is now the critical path at **8:43**, including 432s for mobile tests.
The approximate six-minute goal is not reached.
The workflow has 14 jobs including Changes and CI, below the org's 20 concurrent hosted-job limit.
The aggregate still depends on both matrix job IDs; ruleset 21768840 requires only `CI` and `merge-tree`, so no ruleset change is needed.

## Scope and approval

The maintainer explicitly approved this CI-only implementation and draft PR.
No `test:cached`, flaky-test fixes, ruleset changes, or upstream writes.

## Verification

- [x] Workflow passes actionlint
- [x] Browser/unit shard probe: 18 tests passed
- [x] Each probe shard runs one browser file
- [x] Each probe shard runs one unit file
- [x] No overlap across the four probe files
- [ ] `vp run fork:ci`: exit 1, Scripts failed

The committed-head preflight passed delta, stale-delete, fixup-fold, merge-tree, scan, knip, check, typecheck, and fork-tests (220 files, 1215 tests).
Scripts passed 596 tests and failed one unchanged Discord release workflow test at `scripts/notify-discord-release.test.ts:365`.
A subprocess trace shows the host shell fails before the fixture runs: `/home/user/.bashrc: line 255: path_prepend: command not found`, exit 127.
Full local web attempts also hit browser module/mock errors; the focused shard probe passed with one worker and Playwright's Ubuntu 24.04 x64 platform override on this Ubuntu 26.04 host.
### Hosted CI measurements

[Run 41293089108](https://github.com/RSI-Software/t3code-hyprws/actions/runs/41293089108) is **green on attempt 2**, with 16 checks passed and one unrelated preview check skipped.
Attempt 1 failed only the unchanged Mermaid SVG assertion; its full file passed 10/10 locally, and a governed job rerun passed all 276 shard files and 3677 tests without source changes.
The no-token dependency action succeeded; logs show its private task-cache step skipped, Chromium restored, and the ldd check took under one second.

The green run took **14:51 end-to-end including the failed attempt and rerun**; the rerun attempt itself took 5:08, reusing all other job results.
Attempt 1 took 9:01 but was red, so it is not a green-run speed claim.
The supplied historical median is 10:24; a concrete earlier green run, [41278753510](https://github.com/RSI-Software/t3code-hyprws/actions/runs/41278753510), took 11:50.
These are observations, not a controlled before/after benchmark or a new median.

Job wall times from the Actions jobs API:

| Job | Earlier green | This green run |
| --- | --- | --- |
| Changes | 0:05 | 0:07 |
| Check | 4:37 | 3:54 |
| Test | 6:05 | 8:43 |
| Test Scripts | 2:16 | 2:10 |
| Test Web / Web 1 | 7:18 | 4:59 |
| Test Web 2 | n/a | 5:34 |
| Test Server 1 | 10:17 | 4:49 |
| Test Server 2 | 7:58 | 4:51 |
| Test Server 3 | 6:54 | 5:06 |
| Test Server 4 | n/a | 5:11 |
| Test Server 5 | n/a | 4:48 |
| Test Server 6 | n/a | 5:51 |
| merge-tree | 0:20 | 0:18 |
| CI | 0:03 | 0:03 |

GPT Sol via Pi.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: qol
Co-authored-by: donjor <[email protected]>

## Problem

After #1985, the pooled **`Test` job is the critical path** at **8:43** (run 41293089108).
Inside it, **`apps/mobile` ran 432 s** (260 files) and finished about 20 s after every other package; the next longest took ~157 s.

## Change

- **New `Test Mobile` job:** mobile tests only
- **Setup:** no Electron, no libsecret
- **`Test` drops `@t3tools/mobile`**
- **`Test` skips** on a scripts or mobile-only change
- **`CI` aggregate** needs `test_mobile`

**Gating:** `changes` gains a fork-only `test_mobile` output, derived in the workflow from `test` and `test_packages`.
The shared `ci-changes.cjs` stays upstream's, and the job skips without taking a runner; `CI` already tolerates a skip whose gate is `"false"`.

`test_mobile` per change, from `ci-changes.cjs` output:

| Change | `test_packages` | `test_mobile` | `Test` |
| --- | --- | --- | --- |
| `apps/web` only | `@t3tools/scripts` | false | skips |
| `apps/mobile` only | mobile, scripts | true | skips |
| `apps/server` only | evals, scripts | false | runs |
| `packages/contracts` | includes mobile | true | runs |
| Workflow file | empty (all) | true | runs |
| Docs only | empty, `test=false` | false | skipped |

## Scope and approval

Fork CI tuning only; follows #1985. No product code.

## Verification

- `actionlint` on the workflow → exit 0
- `vp run check:changed` → exit 0
- `vp run fork:ci` → exit 0

**Wall times** (before: run 41293089108; after: run 41296032295):

| Job | Before | After |
| --- | --- | --- |
| Test | 8:43 | 5:11 |
| Test Mobile | n/a | 3:05 |
| Slowest Test Server | 5:51 | 5:21 |
| End to end | ~9:00 | 7:37 |

- **Critical path:** Test Server 6, 2:03 queued
- **Runner cap:** 5 jobs queued 1:36 to 2:44
- **Test Server 2 red:** unrelated test time bomb

Claude Opus 5.5, Claude Code harness in T3 Code.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: qol
Co-authored-by: donjor <[email protected]>

… changes since the pin (#1991)

Closes #1987

## Close conditions

The three copies above:

- [ ] **Each copy** carries a pin
- [ ] **`fork:scan` fails** a stale pin
- [ ] **Failure names** the file and commits
- [ ] **Tests:** stale, current, missing pin
- [ ] **Fork-sync doc:** adapt or no-change step

## Problem

The `hyprws-*.yml` copies of upstream workflows drift silently: upstream `ci.yml` has 26 unmirrored commits.

## Change

- **Pin:** `# fork-mirror: <path>@<sha>` in each header
- **Stale pin:** `fork:scan --mirror-pins` fails it
- **No pin:** fails too
- **Sync tip only:** flag set when `rebased`
- **Ordinary PRs:** never fail on a stale pin
- **No ledger:** the pin lives in the copy
- **Docs:** `fork-sync.md` *Workflow copies*
- **Skill:** fork-sync tip-work bullet

The guard #1151 deleted stays deleted.

Scan of this head against the current base:

```text
$ node scripts/fork-scan.ts --no-typecheck --mirror-pins --target ae521d1 --since origin/hyprws
Workflow mirrors, 1 finding(s):
  MIRROR  .github/workflows/hyprws-ci.yml: upstream .github/workflows/ci.yml changed in 26 commit(s) since pin 5d14c0e; adapt the copy or record no change, then pin ae521d1
  f6d3700 ci: server test shards survive a throttled task cache (pingdotgg#26332)
  67661cf chore: delete 18k lines of code that nothing reaches (pingdotgg#26238)
  ...
exit 1
```

## Pins

| Copy | Pin | Adapted |
| --- | --- | --- |
| `hyprws-ci.yml` | `5d14c0e9686` | none |
| `hyprws-release.yml` | `ae521d1d796` | `e6ca17001d0` |
| `hyprws-agent-context-budget.yml` | `ae521d1d796` | `da1eab01f71` |

- **CI pin:** the last upstream commit it matched
- **CI drift:** the next sync reviews it
- **`e6ca17001d0`:** prerelease drafts until assets land
- **`da1eab01f71`:** error names the evals rerun

#1985 took upstream cache steps only, so the CI pin stays.

### Release: no change

| Upstream commits | Reason |
| --- | --- |
| `f41518f` | launch check: untested release behaviour |
| `30f4d63` `fe27e8e` `e407f9b` `25159b6` `fd1c338` | test jobs: the gate needs green CI |
| `a9418a1` `6327560` `a43b003` | task cache: no secret, `setup-vp` |
| `1688f96` `783ccf0` `7479625` `bd0b9ed` `e74c668` | deploys: the fork has none |
| `e64766c` `569bd02` `14b47ee` `e9a5974` `c57a04b` | macOS, Windows: Linux only |
| `c13f7d9` | `.deb`, arm64: AppImage only |
| `6b450c7` `91cd91c` | npm: not published |
| `83d6583` | mobile: no fork build |
| `b438198` | Discord: no announce |
| `b5fc7be` | OpenAI notes: generated notes kept |
| `3060cc4` | channel defaults: fork resolves its own |
| `bed15b9` | skip in forks: copy is fork-owned |
| `28c32d2` | telemetry: defaults to `dev` here |
| `eb8f6f4` `c7f23c4` `0754920` `2c54f2f` `2f7616e` | CLI archives: already built |

## Scope and approval

The maintainer filed #1987 and assigned this scope.

## Verification

- [x] **Typecheck** (scripts) → exit 0
- [x] **Tests** (3 files) → exit 0, 27 passed
- [x] **check:changed** → exit 1
- [x] **Its two failures:** web browser timeouts
- [x] **No web code** in this diff
- [x] **Both pass** run alone
- [x] **fork:ci** (bwrap HOME workaround) → exit 0

Claude Opus 5.5, Claude Code in T3 Code.

## Fork trailers

Fork-Domain: fork-meta
Fork-Tier: core
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

Upstream's new generated-snapshot test in client-runtime takes over 5 s on the pooled GitHub-hosted runner.

Squashes:

- 141f543fc03 ci(fork): pooled package tests get the same widened per-test budget as web
- 4525592d405 fix(ci): provision browser and fold replay prerequisites (#1666) (#1666)
- 806c939a862 fix(ci): run hyprws CI on stacked pull requests (#1715) (#1715)
- e14e7e73a75 perf(ci): fork CI skips unaffected jobs and reruns of green commits (#1926) (#1926)
- 7660a7f1eb6 fix(ci): pin Playwright test jobs to ubuntu-24.04 (#1972) (#1972)
- f45cff55f89 perf(ci): cache dependencies and shard web and server tests (#1985) (#1985)
- 7530a35f2bc perf(ci): run mobile tests in their own job (#1989) (#1989)
- 0497930b3eb ci(fork-meta): hyprws CI checks generated GraphQL and boots the built web client

Fork-Domain: fork-meta
Fork-Tier: qol
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

Upstream's new generated-snapshot test in client-runtime takes over 5 s on the pooled GitHub-hosted runner.

Squashes:

- 141f543fc03 ci(fork): pooled package tests get the same widened per-test budget as web
- 4525592d405 fix(ci): provision browser and fold replay prerequisites (#1666) (#1666)
- 806c939a862 fix(ci): run hyprws CI on stacked pull requests (#1715) (#1715)
- e14e7e73a75 perf(ci): fork CI skips unaffected jobs and reruns of green commits (#1926) (#1926)
- 7660a7f1eb6 fix(ci): pin Playwright test jobs to ubuntu-24.04 (#1972) (#1972)
- f45cff55f89 perf(ci): cache dependencies and shard web and server tests (#1985) (#1985)
- 7530a35f2bc perf(ci): run mobile tests in their own job (#1989) (#1989)
- 0497930b3eb ci(fork-meta): hyprws CI checks generated GraphQL and boots the built web client

Fork-Domain: fork-meta
Fork-Tier: qol
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

Upstream's new generated-snapshot test in client-runtime takes over 5 s on the pooled GitHub-hosted runner.

Squashes:

- 141f543fc03 ci(fork): pooled package tests get the same widened per-test budget as web
- 4525592d405 fix(ci): provision browser and fold replay prerequisites (#1666) (#1666)
- 806c939a862 fix(ci): run hyprws CI on stacked pull requests (#1715) (#1715)
- e14e7e73a75 perf(ci): fork CI skips unaffected jobs and reruns of green commits (#1926) (#1926)
- 7660a7f1eb6 fix(ci): pin Playwright test jobs to ubuntu-24.04 (#1972) (#1972)
- f45cff55f89 perf(ci): cache dependencies and shard web and server tests (#1985) (#1985)
- 7530a35f2bc perf(ci): run mobile tests in their own job (#1989) (#1989)
- 0497930b3eb ci(fork-meta): hyprws CI checks generated GraphQL and boots the built web client

Fork-Domain: fork-meta
Fork-Tier: qol
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

Upstream's new generated-snapshot test in client-runtime takes over 5 s on the pooled GitHub-hosted runner.

Squashes:

- 141f543fc03 ci(fork): pooled package tests get the same widened per-test budget as web
- 4525592d405 fix(ci): provision browser and fold replay prerequisites (#1666) (#1666)
- 806c939a862 fix(ci): run hyprws CI on stacked pull requests (#1715) (#1715)
- e14e7e73a75 perf(ci): fork CI skips unaffected jobs and reruns of green commits (#1926) (#1926)
- 7660a7f1eb6 fix(ci): pin Playwright test jobs to ubuntu-24.04 (#1972) (#1972)
- f45cff55f89 perf(ci): cache dependencies and shard web and server tests (#1985) (#1985)
- 7530a35f2bc perf(ci): run mobile tests in their own job (#1989) (#1989)
- 0497930b3eb ci(fork-meta): hyprws CI checks generated GraphQL and boots the built web client

Fork-Domain: fork-meta
Fork-Tier: qol
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

Upstream's new generated-snapshot test in client-runtime takes over 5 s on the pooled GitHub-hosted runner.

Squashes:

- 141f543fc03 ci(fork): pooled package tests get the same widened per-test budget as web
- 4525592d405 fix(ci): provision browser and fold replay prerequisites (#1666) (#1666)
- 806c939a862 fix(ci): run hyprws CI on stacked pull requests (#1715) (#1715)
- e14e7e73a75 perf(ci): fork CI skips unaffected jobs and reruns of green commits (#1926) (#1926)
- 7660a7f1eb6 fix(ci): pin Playwright test jobs to ubuntu-24.04 (#1972) (#1972)
- f45cff55f89 perf(ci): cache dependencies and shard web and server tests (#1985) (#1985)
- 7530a35f2bc perf(ci): run mobile tests in their own job (#1989) (#1989)
- 0497930b3eb ci(fork-meta): hyprws CI checks generated GraphQL and boots the built web client

Fork-Domain: fork-meta
Fork-Tier: qol
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

Upstream's new generated-snapshot test in client-runtime takes over 5 s on the pooled GitHub-hosted runner.

Squashes:

- 141f543fc03 ci(fork): pooled package tests get the same widened per-test budget as web
- 4525592d405 fix(ci): provision browser and fold replay prerequisites (#1666) (#1666)
- 806c939a862 fix(ci): run hyprws CI on stacked pull requests (#1715) (#1715)
- e14e7e73a75 perf(ci): fork CI skips unaffected jobs and reruns of green commits (#1926) (#1926)
- 7660a7f1eb6 fix(ci): pin Playwright test jobs to ubuntu-24.04 (#1972) (#1972)
- f45cff55f89 perf(ci): cache dependencies and shard web and server tests (#1985) (#1985)
- 7530a35f2bc perf(ci): run mobile tests in their own job (#1989) (#1989)
- 0497930b3eb ci(fork-meta): hyprws CI checks generated GraphQL and boots the built web client

Fork-Domain: fork-meta
Fork-Tier: qol
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

Upstream's new generated-snapshot test in client-runtime takes over 5 s on the pooled GitHub-hosted runner.

Squashes:

- 141f543fc03 ci(fork): pooled package tests get the same widened per-test budget as web
- 4525592d405 fix(ci): provision browser and fold replay prerequisites (#1666) (#1666)
- 806c939a862 fix(ci): run hyprws CI on stacked pull requests (#1715) (#1715)
- e14e7e73a75 perf(ci): fork CI skips unaffected jobs and reruns of green commits (#1926) (#1926)
- 7660a7f1eb6 fix(ci): pin Playwright test jobs to ubuntu-24.04 (#1972) (#1972)
- f45cff55f89 perf(ci): cache dependencies and shard web and server tests (#1985) (#1985)
- 7530a35f2bc perf(ci): run mobile tests in their own job (#1989) (#1989)
- 0497930b3eb ci(fork-meta): hyprws CI checks generated GraphQL and boots the built web client

Fork-Domain: fork-meta
Fork-Tier: qol
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

Upstream's new generated-snapshot test in client-runtime takes over 5 s on the pooled GitHub-hosted runner.

Squashes:

- 141f543fc03 ci(fork): pooled package tests get the same widened per-test budget as web
- 4525592d405 fix(ci): provision browser and fold replay prerequisites (#1666) (#1666)
- 806c939a862 fix(ci): run hyprws CI on stacked pull requests (#1715) (#1715)
- e14e7e73a75 perf(ci): fork CI skips unaffected jobs and reruns of green commits (#1926) (#1926)
- 7660a7f1eb6 fix(ci): pin Playwright test jobs to ubuntu-24.04 (#1972) (#1972)
- f45cff55f89 perf(ci): cache dependencies and shard web and server tests (#1985) (#1985)
- 7530a35f2bc perf(ci): run mobile tests in their own job (#1989) (#1989)
- 0497930b3eb ci(fork-meta): hyprws CI checks generated GraphQL and boots the built web client

Fork-Domain: fork-meta
Fork-Tier: qol
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

Upstream's new generated-snapshot test in client-runtime takes over 5 s on the pooled GitHub-hosted runner.

Squashes:

- 141f543fc03 ci(fork): pooled package tests get the same widened per-test budget as web
- 4525592d405 fix(ci): provision browser and fold replay prerequisites (#1666) (#1666)
- 806c939a862 fix(ci): run hyprws CI on stacked pull requests (#1715) (#1715)
- e14e7e73a75 perf(ci): fork CI skips unaffected jobs and reruns of green commits (#1926) (#1926)
- 7660a7f1eb6 fix(ci): pin Playwright test jobs to ubuntu-24.04 (#1972) (#1972)
- f45cff55f89 perf(ci): cache dependencies and shard web and server tests (#1985) (#1985)
- 7530a35f2bc perf(ci): run mobile tests in their own job (#1989) (#1989)
- 0497930b3eb ci(fork-meta): hyprws CI checks generated GraphQL and boots the built web client

Fork-Domain: fork-meta
Fork-Tier: qol
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

Upstream's new generated-snapshot test in client-runtime takes over 5 s on the pooled GitHub-hosted runner.

Squashes:

- 141f543fc03 ci(fork): pooled package tests get the same widened per-test budget as web
- 4525592d405 fix(ci): provision browser and fold replay prerequisites (#1666) (#1666)
- 806c939a862 fix(ci): run hyprws CI on stacked pull requests (#1715) (#1715)
- e14e7e73a75 perf(ci): fork CI skips unaffected jobs and reruns of green commits (#1926) (#1926)
- 7660a7f1eb6 fix(ci): pin Playwright test jobs to ubuntu-24.04 (#1972) (#1972)
- f45cff55f89 perf(ci): cache dependencies and shard web and server tests (#1985) (#1985)
- 7530a35f2bc perf(ci): run mobile tests in their own job (#1989) (#1989)
- 0497930b3eb ci(fork-meta): hyprws CI checks generated GraphQL and boots the built web client

Fork-Domain: fork-meta
Fork-Tier: qol
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

Upstream's new generated-snapshot test in client-runtime takes over 5 s on the pooled GitHub-hosted runner.

Squashes:

- 141f543fc03 ci(fork): pooled package tests get the same widened per-test budget as web
- 4525592d405 fix(ci): provision browser and fold replay prerequisites (#1666) (#1666)
- 806c939a862 fix(ci): run hyprws CI on stacked pull requests (#1715) (#1715)
- e14e7e73a75 perf(ci): fork CI skips unaffected jobs and reruns of green commits (#1926) (#1926)
- 7660a7f1eb6 fix(ci): pin Playwright test jobs to ubuntu-24.04 (#1972) (#1972)
- f45cff55f89 perf(ci): cache dependencies and shard web and server tests (#1985) (#1985)
- 7530a35f2bc perf(ci): run mobile tests in their own job (#1989) (#1989)
- 0497930b3eb ci(fork-meta): hyprws CI checks generated GraphQL and boots the built web client

Fork-Domain: fork-meta
Fork-Tier: qol
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

Upstream's new generated-snapshot test in client-runtime takes over 5 s on the pooled GitHub-hosted runner.

Squashes:

- 141f543fc03 ci(fork): pooled package tests get the same widened per-test budget as web
- 4525592d405 fix(ci): provision browser and fold replay prerequisites (#1666) (#1666)
- 806c939a862 fix(ci): run hyprws CI on stacked pull requests (#1715) (#1715)
- e14e7e73a75 perf(ci): fork CI skips unaffected jobs and reruns of green commits (#1926) (#1926)
- 7660a7f1eb6 fix(ci): pin Playwright test jobs to ubuntu-24.04 (#1972) (#1972)
- f45cff55f89 perf(ci): cache dependencies and shard web and server tests (#1985) (#1985)
- 7530a35f2bc perf(ci): run mobile tests in their own job (#1989) (#1989)
- 0497930b3eb ci(fork-meta): hyprws CI checks generated GraphQL and boots the built web client

Fork-Domain: fork-meta
Fork-Tier: qol
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

Upstream's new generated-snapshot test in client-runtime takes over 5 s on the pooled GitHub-hosted runner.

Squashes:

- 141f543fc03 ci(fork): pooled package tests get the same widened per-test budget as web
- 4525592d405 fix(ci): provision browser and fold replay prerequisites (#1666) (#1666)
- 806c939a862 fix(ci): run hyprws CI on stacked pull requests (#1715) (#1715)
- e14e7e73a75 perf(ci): fork CI skips unaffected jobs and reruns of green commits (#1926) (#1926)
- 7660a7f1eb6 fix(ci): pin Playwright test jobs to ubuntu-24.04 (#1972) (#1972)
- f45cff55f89 perf(ci): cache dependencies and shard web and server tests (#1985) (#1985)
- 7530a35f2bc perf(ci): run mobile tests in their own job (#1989) (#1989)
- 0497930b3eb ci(fork-meta): hyprws CI checks generated GraphQL and boots the built web client

Fork-Domain: fork-meta
Fork-Tier: qol
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

Upstream's new generated-snapshot test in client-runtime takes over 5 s on the pooled GitHub-hosted runner.

Squashes:

- 141f543fc03 ci(fork): pooled package tests get the same widened per-test budget as web
- 4525592d405 fix(ci): provision browser and fold replay prerequisites (#1666) (#1666)
- 806c939a862 fix(ci): run hyprws CI on stacked pull requests (#1715) (#1715)
- e14e7e73a75 perf(ci): fork CI skips unaffected jobs and reruns of green commits (#1926) (#1926)
- 7660a7f1eb6 fix(ci): pin Playwright test jobs to ubuntu-24.04 (#1972) (#1972)
- f45cff55f89 perf(ci): cache dependencies and shard web and server tests (#1985) (#1985)
- 7530a35f2bc perf(ci): run mobile tests in their own job (#1989) (#1989)
- 0497930b3eb ci(fork-meta): hyprws CI checks generated GraphQL and boots the built web client

Fork-Domain: fork-meta
Fork-Tier: qol
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

Upstream's new generated-snapshot test in client-runtime takes over 5 s on the pooled GitHub-hosted runner.

Squashes:

- 141f543fc03 ci(fork): pooled package tests get the same widened per-test budget as web
- 4525592d405 fix(ci): provision browser and fold replay prerequisites (#1666) (#1666)
- 806c939a862 fix(ci): run hyprws CI on stacked pull requests (#1715) (#1715)
- e14e7e73a75 perf(ci): fork CI skips unaffected jobs and reruns of green commits (#1926) (#1926)
- 7660a7f1eb6 fix(ci): pin Playwright test jobs to ubuntu-24.04 (#1972) (#1972)
- f45cff55f89 perf(ci): cache dependencies and shard web and server tests (#1985) (#1985)
- 7530a35f2bc perf(ci): run mobile tests in their own job (#1989) (#1989)
- 0497930b3eb ci(fork-meta): hyprws CI checks generated GraphQL and boots the built web client

Fork-Domain: fork-meta
Fork-Tier: qol
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

Upstream's new generated-snapshot test in client-runtime takes over 5 s on the pooled GitHub-hosted runner.

Squashes:

- 141f543fc03 ci(fork): pooled package tests get the same widened per-test budget as web
- 4525592d405 fix(ci): provision browser and fold replay prerequisites (#1666) (#1666)
- 806c939a862 fix(ci): run hyprws CI on stacked pull requests (#1715) (#1715)
- e14e7e73a75 perf(ci): fork CI skips unaffected jobs and reruns of green commits (#1926) (#1926)
- 7660a7f1eb6 fix(ci): pin Playwright test jobs to ubuntu-24.04 (#1972) (#1972)
- f45cff55f89 perf(ci): cache dependencies and shard web and server tests (#1985) (#1985)
- 7530a35f2bc perf(ci): run mobile tests in their own job (#1989) (#1989)
- 0497930b3eb ci(fork-meta): hyprws CI checks generated GraphQL and boots the built web client

Fork-Domain: fork-meta
Fork-Tier: qol
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

Upstream's new generated-snapshot test in client-runtime takes over 5 s on the pooled GitHub-hosted runner.

Squashes:

- 141f543fc03 ci(fork): pooled package tests get the same widened per-test budget as web
- 4525592d405 fix(ci): provision browser and fold replay prerequisites (#1666) (#1666)
- 806c939a862 fix(ci): run hyprws CI on stacked pull requests (#1715) (#1715)
- e14e7e73a75 perf(ci): fork CI skips unaffected jobs and reruns of green commits (#1926) (#1926)
- 7660a7f1eb6 fix(ci): pin Playwright test jobs to ubuntu-24.04 (#1972) (#1972)
- f45cff55f89 perf(ci): cache dependencies and shard web and server tests (#1985) (#1985)
- 7530a35f2bc perf(ci): run mobile tests in their own job (#1989) (#1989)
- 0497930b3eb ci(fork-meta): hyprws CI checks generated GraphQL and boots the built web client

Fork-Domain: fork-meta
Fork-Tier: qol
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

Upstream's new generated-snapshot test in client-runtime takes over 5 s on the pooled GitHub-hosted runner.

Squashes:

- 141f543fc03 ci(fork): pooled package tests get the same widened per-test budget as web
- 4525592d405 fix(ci): provision browser and fold replay prerequisites (#1666) (#1666)
- 806c939a862 fix(ci): run hyprws CI on stacked pull requests (#1715) (#1715)
- e14e7e73a75 perf(ci): fork CI skips unaffected jobs and reruns of green commits (#1926) (#1926)
- 7660a7f1eb6 fix(ci): pin Playwright test jobs to ubuntu-24.04 (#1972) (#1972)
- f45cff55f89 perf(ci): cache dependencies and shard web and server tests (#1985) (#1985)
- 7530a35f2bc perf(ci): run mobile tests in their own job (#1989) (#1989)
- 0497930b3eb ci(fork-meta): hyprws CI checks generated GraphQL and boots the built web client

Fork-Domain: fork-meta
Fork-Tier: qol
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

Upstream's new generated-snapshot test in client-runtime takes over 5 s on the pooled GitHub-hosted runner.

Squashes:

- 141f543fc03 ci(fork): pooled package tests get the same widened per-test budget as web
- 4525592d405 fix(ci): provision browser and fold replay prerequisites (#1666) (#1666)
- 806c939a862 fix(ci): run hyprws CI on stacked pull requests (#1715) (#1715)
- e14e7e73a75 perf(ci): fork CI skips unaffected jobs and reruns of green commits (#1926) (#1926)
- 7660a7f1eb6 fix(ci): pin Playwright test jobs to ubuntu-24.04 (#1972) (#1972)
- f45cff55f89 perf(ci): cache dependencies and shard web and server tests (#1985) (#1985)
- 7530a35f2bc perf(ci): run mobile tests in their own job (#1989) (#1989)
- 0497930b3eb ci(fork-meta): hyprws CI checks generated GraphQL and boots the built web client

Fork-Domain: fork-meta
Fork-Tier: qol
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

Upstream's new generated-snapshot test in client-runtime takes over 5 s on the pooled GitHub-hosted runner.

Squashes:

- 141f543fc03 ci(fork): pooled package tests get the same widened per-test budget as web
- 4525592d405 fix(ci): provision browser and fold replay prerequisites (#1666) (#1666)
- 806c939a862 fix(ci): run hyprws CI on stacked pull requests (#1715) (#1715)
- e14e7e73a75 perf(ci): fork CI skips unaffected jobs and reruns of green commits (#1926) (#1926)
- 7660a7f1eb6 fix(ci): pin Playwright test jobs to ubuntu-24.04 (#1972) (#1972)
- f45cff55f89 perf(ci): cache dependencies and shard web and server tests (#1985) (#1985)
- 7530a35f2bc perf(ci): run mobile tests in their own job (#1989) (#1989)
- 0497930b3eb ci(fork-meta): hyprws CI checks generated GraphQL and boots the built web client

Fork-Domain: fork-meta
Fork-Tier: qol
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

Upstream's new generated-snapshot test in client-runtime takes over 5 s on the pooled GitHub-hosted runner.

Squashes:

- 141f543fc03 ci(fork): pooled package tests get the same widened per-test budget as web
- 4525592d405 fix(ci): provision browser and fold replay prerequisites (#1666) (#1666)
- 806c939a862 fix(ci): run hyprws CI on stacked pull requests (#1715) (#1715)
- e14e7e73a75 perf(ci): fork CI skips unaffected jobs and reruns of green commits (#1926) (#1926)
- 7660a7f1eb6 fix(ci): pin Playwright test jobs to ubuntu-24.04 (#1972) (#1972)
- f45cff55f89 perf(ci): cache dependencies and shard web and server tests (#1985) (#1985)
- 7530a35f2bc perf(ci): run mobile tests in their own job (#1989) (#1989)
- 0497930b3eb ci(fork-meta): hyprws CI checks generated GraphQL and boots the built web client

Fork-Domain: fork-meta
Fork-Tier: qol
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

Upstream's new generated-snapshot test in client-runtime takes over 5 s on the pooled GitHub-hosted runner.

Squashes:

- 141f543fc03 ci(fork): pooled package tests get the same widened per-test budget as web
- 4525592d405 fix(ci): provision browser and fold replay prerequisites (#1666) (#1666)
- 806c939a862 fix(ci): run hyprws CI on stacked pull requests (#1715) (#1715)
- e14e7e73a75 perf(ci): fork CI skips unaffected jobs and reruns of green commits (#1926) (#1926)
- 7660a7f1eb6 fix(ci): pin Playwright test jobs to ubuntu-24.04 (#1972) (#1972)
- f45cff55f89 perf(ci): cache dependencies and shard web and server tests (#1985) (#1985)
- 7530a35f2bc perf(ci): run mobile tests in their own job (#1989) (#1989)
- 0497930b3eb ci(fork-meta): hyprws CI checks generated GraphQL and boots the built web client

Fork-Domain: fork-meta
Fork-Tier: qol
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>

Upstream's new generated-snapshot test in client-runtime takes over 5 s on the pooled GitHub-hosted runner.

Squashes:

- 141f543fc03 ci(fork): pooled package tests get the same widened per-test budget as web
- 4525592d405 fix(ci): provision browser and fold replay prerequisites (#1666) (#1666)
- 806c939a862 fix(ci): run hyprws CI on stacked pull requests (#1715) (#1715)
- e14e7e73a75 perf(ci): fork CI skips unaffected jobs and reruns of green commits (#1926) (#1926)
- 7660a7f1eb6 fix(ci): pin Playwright test jobs to ubuntu-24.04 (#1972) (#1972)
- f45cff55f89 perf(ci): cache dependencies and shard web and server tests (#1985) (#1985)
- 7530a35f2bc perf(ci): run mobile tests in their own job (#1989) (#1989)
- 0497930b3eb ci(fork-meta): hyprws CI checks generated GraphQL and boots the built web client

Fork-Domain: fork-meta
Fork-Tier: qol
Fork-Upstreamable: no
Co-authored-by: donjor <[email protected]>