feat: add self contained test-running ability to boxel-cli by jurgenwerk · Pull Request #4914 · cardstack/boxel

jurgenwerk · 2026-05-21T11:33:43Z

Summary

boxel test is now self-contained and local-only by default — no realm-server, no Vite dev server, no push step between writing code and running tests. The factory's agent loop tightens from write → realm push → boxel test to just write → boxel test, saving a tool-call turn per validation cycle.

What ships

Local-mode execution (boxel test with no args): the CLI starts an in-process HTTP server that hosts both the test page (from bundled-test-harness/) and the workspace's realm modules. .gts / .ts files are transpiled on demand via runtime-common's transpileJS. Workspace cards are read from cwd; base + skills realm sources are vendored under bundled-realms/.
Bundled test harness: pnpm build:test-harness copies the host's dev-mode dist/ into bundled-test-harness/, dropping only sourcemaps. ~60 MB unpacked / ~15 MB compressed. See "Why we ship the whole host dist" below for the rationale.
Single merged server: the asset server (test page + host bundle chunks) and the workspace transpiling server are one http.createServer listening on a random localhost port. Same origin for everything chromium fetches; module mounts at /workspace/, /base/, /skills/, the host bundle assets at /assets/..., and the test-page HTML at /.
Remote-realm fallback: boxel test --realm <url> opts back into the older remote-realm flow for testing cards already on a published realm.

Before vs after

Aspect	Pre-CS-11164	After this PR
Where harness comes from	Sibling `packages/host/dist/` (monorepo layout required)	`bundled-test-harness/` shipped in the npm tarball
`--realm` flag on `boxel test`	Required	Optional (defaults to local mode)
Realm-server needed at test time	Yes	No (in local mode)
`boxel realm push` before each test iteration	Required	Eliminated in local mode
Workable outside the monorepo	No	Yes
Agent tool-call turns per test iteration	2 (push + test)	1 (test)

Installation footprint

Step	Lands on disk
`npm install -g @cardstack/boxel-cli`	~75 MB CLI (CLI bundle + bundled-test-harness + bundled-realms + bundled-types)
User runs `boxel test` for the first time	+ ~150 MB chromium (via the separate `npx playwright install chromium` step the CLI's error message prompts; not auto-installed)
Total for a `boxel test` user	~225 MB

Why we ship the whole host dist (and not a pruned slim subset)

Spent meaningful time on this during the branch and rejected three different slim approaches. Recording the reasoning so future-us doesn't repeat it:

Attempt 1 — captured manifest of "what chromium actually fetches". Ran a card test with a network-request recorder, committed the resulting list of paths, had build-test-harness.ts copy only those files. Got the harness from 60 MB → 27 MB. Rejected because:

The committed manifest entries are content-hashed (runtime-common-Czdr1nwr.js). Any host content change re-hashes filenames, the committed paths 404, and the slim bundle silently misses chunks.
"Regenerate the manifest before publishing" is a manual step. Per-branch policy, automation-or-bust: tooling can't depend on humans remembering.
CI-side automation (capture-on-publish, Vite-manifest static graph, etc.) all add real moving parts for a single-digit-MB win in absolute terms.

Attempt 2 — bisecting minimizer over the captured set. Wrote a script that removed each captured entry, re-ran the test, kept the removal only if tests stayed green. ~232 entries. After 15 minutes of running and ~30 KEEP/REMOVED decisions, observed a ~7% success rate — almost everything tried as "obviously UI" (submode-switcher, modal-container, AI-assistant chunks, code-mode commands, etc.) was actually needed at runtime. Investigated why: it's Embroider's @embroider/virtual/compat-modules auto-glob. The host's app/app.ts does Resolver.withModules(compatModules), which materializes every .gts under app/services/, app/components/, app/commands/, app/routes/. The Ember resolver eagerly imports them at app boot — not because the test code uses them, but because the resolver walks the keyspace and pre-touches each entry. Removing any auto-globbed chunk fails the boot chain even though the test never executes that code. Rejected because the only way past the floor is to refactor the host (make monaco-service, ai-assistant-panel-service, etc. truly lazy / dynamic-import) — multi-day Embroider work that the ticket explicitly flagged as the bulk of the effort, and out of scope for what this PR is doing.

Attempt 3 — runtime fetch from GitHub Release. Split the harness into its own host-test-harness-v* release, downloaded on first boxel test per CLI version, cached under ~/.cache/boxel-cli/. Got the CLI tarball down to ~3 MB. Rejected because:

The host build still has to happen somewhere (the harness release workflow). Same coupling, just relocated.
Adds a fetcher + cache + sha256-verify + pin file + a new publish workflow + a "bump the pin" PR dance — meaningful infrastructure for a 60→3 MB tarball win when the user pays the difference under ~/.cache/ anyway.
First-run failure modes (offline, GH down, sha256 mismatch) are new surfaces with no equivalent in the bundled approach.

So we ship the full ~60 MB. The size context, honestly: a slim cut would save ~30 MB on disk for boxel test users. They're already paying 150 MB for chromium one-time, and the CLI install dominates only briefly until chromium downloads, after which it's ~25% of the total. Not nothing, but not worth the staleness/maintenance/infra cost the slim approaches require — and the only real way to a 5–10 MB harness is the host-side lazy-loading refactor anyway.

Files & flow

New: packages/boxel-cli/scripts/build-test-harness.ts — copies packages/host/dist/ minus sourcemaps into bundled-test-harness/. Refuses to proceed if tests/index.html is missing (catches accidental production-mode host builds).
New: packages/boxel-cli/scripts/build-realms.ts — vendors packages/base/ and packages/skills-realm/contents/ into bundled-realms/. Skills realm is optional (its contents are gitignored; CI doesn't have it; runtime never reaches /skills/ from a card test).
Changed: packages/boxel-cli/src/commands/test.ts — adds runTestsLocally; --realm becomes opt-in; the test-page server gains realm-mount routing and on-demand transpile.
Changed: packages/boxel-cli/scripts/build.ts — copies content_tag_bg.wasm next to dist/index.js post-build (esbuild inlines content-tag's wasm loader; the loader's readFileSync resolves against the CLI's dist dir).
Changed: .github/workflows/boxel-cli-publish.yml — adds pnpm --filter @cardstack/host build before the CLI build (the bundled-test-harness step reads host's dev dist).
Changed: .github/workflows/ci.yaml — boxel-cli-build job now depends on test-web-assets so the cached host dist is restored before the CLI build runs.
Changed: packages/boxel-cli/package.json — @playwright/test moves to dependencies (was devDependencies) so the headless-browser driver ships with a published install. bundled-test-harness/, bundled-realms/ added to files. Chromium itself is not auto-downloaded on install; users run npx playwright install chromium once.
Changed: runbook + software-factory-operations skill — rewritten around the local-mode default.

Other architecture decisions

Local server transpiles workspace cards in-process rather than pushing them to a realm-server first. Removes a tool-call turn per agent validation cycle.
One HTTP server, not two: the workspace module mounts share the same origin as the test page so chromium needs no CORS preflights and the loader's URL-mapping works out of the box.
Realm-server still required for boxel realm pull — the user syncs once from a realm-server (local, staging, or prod), then tests entirely locally. Validates the agent-token-efficiency goal: one command per validation cycle after the initial sync.

CS-11164 in Linear

Test plan

pnpm build produces ~60 MB bundled-test-harness/, ~1.7 MB bundled-realms/, ~2.4 MB CLI bundle.
pnpm lint (js + types) clean.
Local smoke: cd /tmp/cs-11164-check && boxel test from a workspace pulled via boxel realm pull (no monorepo on PATH, no realm-server running, no Vite) → 11 passed, 0 failed (~3 s).
Published-install smoke: pnpm pack + npm install -g <tarball>, which boxel resolves to the npm-installed binary, boxel test from the same workspace → same result. Tarball ~18 MB.
--debug shows the merged-server origin handling correctly and only emits chromium console + page errors.
BOXEL_TEST_HARNESS_DIR=<path> env override and --host-dist-dir <path> flag still resolve to the right dist.
CI green.

🤖 Generated with Claude Code

`boxel test` now ships its own copy of the host's compiled test bundle in `bundled-test-harness/`, populated at release time by `scripts/build-test-harness.ts` from `packages/host/dist/`. The resolver prefers the bundled dir, falls back to the sibling `packages/host/dist/` for in-monorepo dev. `@playwright/test` moves from devDependencies to dependencies so a published install has the driver available. Tradeoff: the bundle is ~60MB. Stripping Monaco editor chunks looked tempting (drops it to ~40MB) but the host's Ember service container imports them during boot — 404s break test-page init silently with only a 5-minute `waitForFunction` timeout. Don't re-strip without auditing the host's eager imports. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

github-actions · 2026-05-21T11:53:47Z

Host Test Results

1 files 1 suites 1h 34m 3s ⏱️
2 724 tests 2 709 ✅ 15 💤 0 ❌
2 743 runs 2 728 ✅ 15 💤 0 ❌

Results for commit 361f82f.

Realm Server Test Results

1 files ±0 1 suites ±0 9m 39s ⏱️ +44s
1 482 tests ±0 1 482 ✅ ±0 0 💤 ±0 0 ❌ ±0
1 573 runs ±0 1 573 ✅ ±0 0 💤 ±0 0 ❌ ±0

Results for commit 361f82f. ± Comparison against earlier commit 94656aa.

Drops the realm-server proxy round-trip out of the agent loop. The test command now defaults to reading cards from a local workspace dir (cwd or [path] arg), starts a single in-process server that both hosts the existing test-page bundle and transpiles `.gts`/ `.ts` modules on demand via runtime-common's `transpileJS`. Base realm source is vendored into `bundled-realms/` (1.7MB) so a published install resolves `https://cardstack.com/base/...` imports without anything else on disk. `--realm <url>` is now opt-in for the older remote-realm flow. Post-build copies `content_tag_bg.wasm` next to `dist/index.js` — esbuild inlines content-tag's wasm loader, whose `readFileSync` then resolves against the CLI's dist/. The 60MB `bundled-test-harness/` still ships as-is; the slim extraction work follows in this same PR. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

`boxel test` now ships only the chunks chromium actually loads during a card test, captured into `scripts/test-harness-manifest.json` by running the test runner with `BOXEL_TEST_HARNESS_MANIFEST=<path>`. `build-test-harness.ts` copies just the manifest-listed files from `host/dist/` into `bundled-test-harness/`, dropping the AI assistant, code-mode, monaco workers, cytoscape, katex, and all the commands that card tests don't exercise. Cuts the harness from ~60MB (sourcemaps-stripped host dist) to ~27MB without touching the host's Embroider config. Smoke-tested against a multi-test sticky-note workspace from /tmp — 4 passed, manifest stable across renderCard variants. Regenerate the manifest when host adds runtime deps card tests load: the build script header documents the one-line capture command. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

The slim manifest filter still reads from `packages/host/dist/`, so `pnpm run build` in `packages/boxel-cli` needs the host's dev bundle on disk. Add the build step explicitly with a comment pointing at why production mode is wrong (strips test entry). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Copilot

Pull request overview

This PR makes boxel test self-contained by default by bundling a slimmed host QUnit harness and vendoring base/skills realm sources, then running tests via a single in-process HTTP server that serves both the test page assets and locally transpiled realm modules.

Changes:

Add a local-mode default for boxel test that serves workspace + bundled realms via an in-process server and runs QUnit in Playwright Chromium.
Introduce build scripts to vendor bundled-realms/ and copy a manifest-pruned bundled-test-harness/ from packages/host/dist/.
Update packaging/publish workflow so Playwright is a runtime dependency and the host build runs before publishing the CLI.

Reviewed changes

Copilot reviewed 12 out of 13 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
pnpm-lock.yaml	Moves `@playwright/test` to the boxel-cli runtime dependency set in the lockfile.
packages/software-factory/docs/runbook.md	Updates runbook to reflect local-only default `boxel test` behavior and `--realm` opt-in.
packages/software-factory/.agents/skills/software-factory-operations/SKILL.md	Updates skill docs/examples for new local-mode `boxel test [path]` flow.
packages/boxel-cli/tsconfig.json	Excludes bundled artifacts from TS compilation.
packages/boxel-cli/src/commands/test.ts	Implements local-mode test execution, unified test-page + module server, and manifest capture.
packages/boxel-cli/scripts/test-harness-manifest.json	Adds the committed allowlist of host dist assets to ship in the slim harness.
packages/boxel-cli/scripts/build.ts	Copies `content_tag_bg.wasm` into `dist/` so transpilation works in the bundled CLI.
packages/boxel-cli/scripts/build-test-harness.ts	Adds manifest-driven harness bundling from `packages/host/dist/`.
packages/boxel-cli/scripts/build-realms.ts	Adds realm vendoring script for `base/` and `skills-realm/contents/`.
packages/boxel-cli/package.json	Ships bundled realms/harness in published package; makes Playwright a dependency; wires new build steps.
packages/boxel-cli/.gitignore	Ignores new bundled artifact directories.
packages/boxel-cli/.eslintignore	Ignores new bundled artifact directories for linting.
.github/workflows/boxel-cli-publish.yml	Builds `@cardstack/host` before building/publishing the CLI so harness inputs exist.

Files not reviewed (1)

pnpm-lock.yaml: Language not supported

Comments suppressed due to low confidence (1)

packages/boxel-cli/src/commands/test.ts:26

The comment above loadChromium() still says @playwright/test is a devDependency and won’t be present in a published install, but this PR moves @playwright/test to dependencies so it will be present at runtime. Please update/remove the comment to reflect the new packaging model (it’s still external to esbuild, but not a devDependency anymore).

// `@playwright/test` is a devDependency and external in our esbuild
// config, so it's not present in a published-from-npm install. Anything
// loaded at the top of this module would crash `boxel --help` for end
// users who never run `boxel test`. Resolved lazily inside the runner
// instead.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

    .command('test')
    .description(
-      "Run the realm's QUnit test suite (every `*.test.gts` file) in a headless Chromium driven against the host app. Monorepo-only: relies on the host app's compiled `dist/` being reachable from this CLI's location (or via TEST_HARNESS_HOST_DIST_PACKAGE_DIR).",
+      'Run every `*.test.gts` file in a workspace directory in a headless Chromium driven against the host app. Defaults to type-checking and serving cards from the local workspace (cwd or [path]); pass `--realm <url>` to test cards already on a remote realm instead.',


+ * `src/lib/local-module-server.ts`) mounts these directories at `/base/`
+ * and `/skills/` and transpiles `.gts` / `.ts` on demand. Cards that


+  for (let entry of entries) {
+    // Manifest entries are URL paths captured from the test runner.
+    // The CLI's local-mode realm mounts (`/workspace/`, `/base/`,
+    // `/skills/`) and the root request (`/`) are served at runtime,
+    // not from host/dist — skip those.
+    if (
+      entry === '/' ||
+      entry.startsWith('/workspace/') ||
+      entry.startsWith('/base/') ||
+      entry.startsWith('/skills/')
+    ) {
+      continue;
+    }
+    let rel = entry.replace(/^\//, '');
+    let src = join(HOST_DIST, rel);
+    let dst = join(OUT_DIR, rel);
+    try {
+      let st = statSync(src);
+      if (!st.isFile()) {
+        skipped.push(entry);
+        continue;
+      }
+    } catch {
+      skipped.push(entry);
+      continue;
+    }
+    mkdirSync(dirname(dst), { recursive: true });
+    copyFileSync(src, dst);
+    copied++;


- Fix `boxel test` --help wording: local mode doesn't typecheck, it serves cards via an in-process transpiling server. - Drop the stale `src/lib/local-module-server.ts` reference from `build-realms.ts` header; that file was merged into `test.ts`. - Harden `build-test-harness.ts`: fail loudly when files in `ALWAYS_INCLUDE` (`tests/index.html`) are missing — that's the early signal for a production-mode host build that strips test entries. Also refuse to copy `*.map` even if a stale manifest captured one. - Update the lazy-load-playwright comment to match the new packaging model — Playwright moved from devDependency to runtime dependency in the previous commit on this branch. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

The `BOXEL_TEST_HARNESS_MANIFEST` capture was recording every same-origin request — including `/workspace/sticky-note.test` and similar paths served from the runtime realm mounts. Those are useless in the committed manifest (the build script already skips them) and made it noisy/confusing. Filter at capture time using the runner's `realmMounts` config so only host/dist paths land in the manifest. Regenerated against the factory-test-cs11165-1xx workspace's 11-test sticky-note suite (was: 268 entries with 5 realm-mount cruft; now: 232 host-dist entries). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

`BOXEL_TEST_RUNNER_TIMEOUT_MS` overrides the `page.waitForFunction` ceiling (default 5 min). Used by the manifest-minimization script to fail fast on certain-failure removals without burning the full CI-friendly default. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

The captured manifest needed manual regeneration whenever the host's chunks re-hashed (which happens on basically any host content change). That made correctness depend on a maintainer remembering to refresh the JSON before publishing — silent staleness, easy to forget, real maintenance burden. Going with the "no manifest" approach: copy `packages/host/dist/` into `bundled-test-harness/` wholesale, dropping only sourcemaps. 60 MB unpacked, ~15 MB on the wire. Next to the 150 MB Playwright chromium download every `boxel test` user already needs, the size delta is noise. Zero possibility of staleness because there is no snapshot to drift. Removes: - `scripts/test-harness-manifest.json` (committed allowlist) - `BOXEL_TEST_HARNESS_MANIFEST` env capture path in test.ts - All the manifest filtering / required-file guards in `build-test-harness.ts` Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Stop bundling `packages/host/dist/` inside the boxel-cli npm tarball. Instead, host test harness ships as a separate GH Release tarball that boxel-cli downloads on first `boxel test` per CLI version and caches under `~/.cache/boxel-cli/host-test-harness/<version>/`. Why: bundling at CLI build time tied every boxel-cli publish to a host build, and shipping the full ~60 MB inside every CLI install was overkill for a one-time-download. The release-then-fetch shape decouples the two: the host test harness gets cut on demand (workflow_dispatch by a maintainer), boxel-cli embeds a pinned version + sha256, and `boxel test` resolves through the fetcher. Resolution order in the fetcher: 1. --host-dist-dir flag → use as-is 2. BOXEL_TEST_HARNESS_DIR env → use as-is (for CI) 3. Monorepo sibling packages/host/dist/ → use directly (so in-repo devs never download) 4. Local cache hit at the pinned version → use 5. Download from GH release, sha256-verify, extract, use New files: - .github/workflows/host-test-harness-publish.yml manual workflow_dispatch; builds host, tars dist minus sourcemaps, creates a host-test-harness-v<ver> release. - packages/boxel-cli/host-test-harness.json version + sha256 pin (currently the 0.0.0-placeholder until a maintainer cuts the first release). - packages/boxel-cli/src/lib/host-test-harness-fetcher.ts the resolver / downloader / sha256-verifier / cache manager. Removed: - packages/boxel-cli/scripts/build-test-harness.ts - bundled-test-harness/ tracking in .gitignore, .eslintignore, tsconfig.json, package.json files array, build chain, clean target. - The pnpm --filter @cardstack/host build step in the boxel-cli publish workflow. New CLI flags: --refresh-harness, --offline-tarball <path>. Smoke-tested in-monorepo (resolves to sibling, 11/11 passing). Placeholder-pin guard verified by stashing the sibling and re-running — surfaces a clear "no release cut yet" error. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

This reverts commit 1f69ddf.

Two unrelated CI failures revealed by the latest PR run: 1. `boxel-cli-build` job had no `host/dist/` because it didn't depend on `test-web-assets`. Add the dependency and the same download+restore steps the other CLI jobs use, so the host bundle is available when `build:test-harness` reads from it. 2. `build:realms` exited 1 on missing `packages/skills-realm/contents/`. That directory is gitignored (cloned separately via `pnpm --filter @cardstack/skills-realm skills:setup`) and isn't present in CI workspaces. Card tests don't actually load anything from `/skills/` at runtime — that mount is for AI-assistant scaffolding — so mark `skills` as a non-required realm and skip with a log when missing. Verified locally by stashing `packages/skills-realm/contents/` and running `pnpm build:realms` — produces the slightly smaller bundled-realms output (1.41 MB without skills vs 1.70 MB with). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…f-contained-qunit-test-harness-into-boxel-cli

Added earlier on this branch so the manifest-minimization script could fail-fast at 30 s per iteration. We abandoned the manifest approach, so the knob no longer has a consumer. Restore the simple 5-minute waitForFunction default. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

The 'manifest-listed chunks' / '~27MB despite the host producing 60MB' wording was left over from when we shipped a manifest-pruned slim cut. We reverted to the full host dev dist; only sourcemaps are dropped now. Update the comment to match.

jurgenwerk · 2026-05-22T15:55:30Z

My idea was to try to slim down the build needed for the harness but doing this looks like a significant endeavor in the embroider territory, so I opted into simply bundling the development host build into the cli (the description has info on the CLI size impact) that has the test runner and related support.

This is hard because a card test imports Application, which imports compat-modules, which has 138 top-level static import statements for every file under app/. Removing any one of those chunks causes the ES module graph to abort at app.ts evaluation — before Ember boots, before QUnit loads, before any test code runs.

Claude recommends we could do this:

Don't use compat-modules. Replace it in a test-only app-test.ts with a manually-curated registry: {
'services/loader-service': LoaderService, 'services/realm-info-service': RealmInfoService, ... } — only what the test path actually looks up at runtime

I could try this, but perhaps bundling the host development dist into boxel cli for purposes of running the tests isn't such a big deal? On the other hand it's a lot of bloat, one other option I could think of is to publish the build as a package and let the boxel-cli download (and cache it) it only when running boxel test.

habdelra · 2026-05-22T21:52:44Z

+ * Copies the dev-mode host build wholesale, dropping only sourcemaps.
+ * Tried a manifest-driven slim cut (~27 MB instead of ~60 MB), but
+ * the manifest needed manual regen whenever the host's chunks
+ * re-hashed, and we don't want correctness to depend on human
+ * upkeep — see `feedback_no_manual_maintenance` in memory. The full
+ * dist is ~60 MB unpacked and ~15 MB on the wire; next to the
+ * 150 MB Playwright chromium install every `boxel test` user
+ * already needs, it's noise.


that's an interesting approach. like we chatted about the other day, it would be interesting if in rthe future we could spearate this into its own npm package (boxel-cli-test-support), that would prompt the user to install if you try to run the test validations.

jurgenwerk and others added 3 commits May 21, 2026 14:30

jurgenwerk requested a review from Copilot May 22, 2026 07:36

Copilot started reviewing on behalf of jurgenwerk May 22, 2026 07:37 View session

Copilot AI reviewed May 22, 2026

View reviewed changes

jurgenwerk and others added 7 commits May 22, 2026 09:46

Revert "Fetch host test harness lazily at boxel test runtime (CS-11164)"

2f81ac3

This reverts commit 1f69ddf.

jurgenwerk changed the title ~~Bundle self-contained QUnit test harness into boxel-cli (CS-11164)~~ feat: bundle self-contained QUnit test harness into boxel-cli (CS-11164) May 22, 2026

jurgenwerk and others added 3 commits May 22, 2026 12:31

Merge remote-tracking branch 'origin/main' into cs-11164-bundle-a-sel…

c2e816d

…f-contained-qunit-test-harness-into-boxel-cli

jurgenwerk marked this pull request as ready for review May 22, 2026 15:56

jurgenwerk requested review from a team and habdelra May 22, 2026 15:56

jurgenwerk changed the title ~~feat: bundle self-contained QUnit test harness into boxel-cli (CS-11164)~~ feat: add self contained test-running ability to boxel-cli May 22, 2026

habdelra reviewed May 22, 2026

View reviewed changes

habdelra approved these changes May 22, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: add self contained test-running ability to boxel-cli#4914

feat: add self contained test-running ability to boxel-cli#4914
jurgenwerk wants to merge 14 commits into
mainfrom
cs-11164-bundle-a-self-contained-qunit-test-harness-into-boxel-cli

jurgenwerk commented May 21, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented May 21, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

jurgenwerk commented May 22, 2026 •

edited

Loading

Uh oh!

habdelra May 22, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

		* `src/lib/local-module-server.ts`) mounts these directories at `/base/`
		* and `/skills/` and transpiles `.gts` / `.ts` on demand. Cards that

Conversation

jurgenwerk commented May 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

What ships

Before vs after

Installation footprint

Why we ship the whole host dist (and not a pruned slim subset)

Files & flow

Other architecture decisions

Test plan

Uh oh!

github-actions Bot commented May 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Host Test Results

Realm Server Test Results

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

jurgenwerk commented May 22, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

habdelra May 22, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

jurgenwerk commented May 21, 2026 •

edited

Loading

github-actions Bot commented May 21, 2026 •

edited

Loading

jurgenwerk commented May 22, 2026 •

edited

Loading