fix(ci): stop fanning out the whole cockpit e2e matrix for unrelated changes

[blove]

PR #932 changed exactly three files under apps/website/src and ran 41 Cockpit — e2e lanes. Two independent scoping defects stacked.

1. The cockpit shell claimed the e2e matrix

apps/cockpit/project.json carried scope:cockpit-e2e. But the matrix dispatches nx e2e for the standalone Angular cap apps under cockpit/**, and:

• 0 of 40 caps depend on the apps/cockpit Next.js shell (checked against the nx project graph)
• no workflow anywhere runs the shell's own e2e target

So the tag could never select real work — only over-select. And it did: apps/cockpit statically depends on apps/website, so a website-only PR made the shell nx-affected and flipped cockpit_e2e true. From that run's CI scope log:

Affected projects (3):
  website [scope:website, scope:website-e2e, type:app]
  cockpit [scope:cockpit, scope:cockpit-deploy-smoke, scope:cockpit-e2e, scope:cockpit-examples, type:app]
  scripts [scope:scripts-tests, type:tooling]
cockpit_e2e=true

2. A cap was only matchable by its Angular name

A cap is two independent nx projects — the Angular app that owns the e2e target and the python backend — with no edge between them. cockpit-matrix.mjs matched affected projects on cap.angular only, so a python-only cap change attributed nothing, which main() reads as "nx found no cap" and answers with the full fleet.

Measured, before → after

change	lanes before	lanes after
website-only (#932 base..head)	41	0 — the dispatcher's if: gate is now false
python-only cap change	40	1 — the cap that changed
libs/chat change	40	40 — unchanged, no coverage lost

(40 vs 41 is drift in the cap count since that run, not an effect of this change.)

Notes

• The python sibling's nx name is read from its own project.json rather than derived from the path. A convention-derived guess would fail open to the full fleet without ever reporting that it guessed.
• The emitted matrix JSON keeps its exact {angular, python} shape; pythonName is internal to attribution.
• The shell still builds and tests on a website change — it consumes that code. Only the cap e2e matrix is descoped.

Tests

Both new spec groups were confirmed failing before the fix:

not ok 1 - apps/cockpit is not tagged scope:cockpit-e2e
not ok 2 - a website-only change leaves cockpit_e2e false
# SyntaxError: ... does not provide an export named 'isCapAffected'

After: node --test over all six suites 125/125, nx test scripts vitest 71/71 (baseline), nx lint scripts clean. No new spec file, so scripts/vite.config.mts exclude and the ci.yml node --test invocation need no change.

Left alone deliberately

scope:cockpit-examples and scope:cockpit-deploy-smoke on the shell are also broad, but those jobs are cheap and already self-scope (Cockpit — build all examples correctly logged "No affected cockpit angular projects" on #932). Separately: apps/cockpit has an e2e target that no workflow runs — worth a look, but not this PR.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
threadplane	Canceled		Sep 1, 2026 10:35pm UTC

[github-actions]

Automated approval: this PR received an intelligent (AI) code review. See the review comments on this PR.

[github-actions]

Claude finished @blove's task in 0s —— View job

---

I'll analyze this and get back to you.

[github-actions]

Claude finished @blove's task in 0s —— View job

---

I'll analyze this and get back to you.