Feat/core migration image generation

[Vect0rM]

Describe Your Changes

Fixes Issues

• Closes #
• Closes #

Self Checklist

• Added relevant comments, esp in complex areas
• Updated docs (for bug fixes / features)
• Created issues for follow-up changes or refactoring needed

[Vect0rM]

Second pass, on 3439bde. Of the five things from last round, three are closed — and the two that aren't are the same two I'd still like before merge, plus one new one the repo's own gate found.

The headline: base (835f663) is now an ancestor of the head, so this isn't a merge at all — it fast-forwards feat/atomic-chat-core-migration by 281 commits. git merge-tree reports 0 conflicts, against 21 last round.

Verified

Measured on the PR head and on the base branch in the same container.

One note for anyone reproducing, because it cost me a run: yarn test needs yarn build:core first. Without core/dist, 106 test files fail to collect on Failed to resolve entry for package "@janhq/core" and it looks like the branch is on fire. It isn't — with core built it's clean.

• make typecheck — exit 0.
• yarn lint — 0 errors, 30 warnings (all react-refresh/only-export-components and react-hooks/exhaustive-deps).
• Full yarn test — 390 files passed, 1 skipped; 4508 passed, 16 skipped. Exit 0. I ran the base-branch baseline in the same container rather than quoting last round's numbers: 311 files / 3201 passed. So +79 test files and +1,307 tests, nothing else disturbed, nothing red.
• No CI, still — get_check_runs on 3439bde returns total_count: 0, fourth round running.

Two things I checked because they're the ones that fail silently, and both are right:

The squash-merge-back didn't eat anything. This branch carries #300's original commits and merges main's squash of the same work (21a2896). git diff 21a2896^1 21a2896 is empty — the squash brought in nothing and dropped nothing. That's the correct resolution of the case that usually reverts things without anyone noticing.

c537970 survived main landing on top of it. 3439bde rewrote 141 lines of web-app/src/lib/diffusion/models.ts from main, over the same function c537970 had just fixed. The .filter((e) => e.required && !e.present) is still there at line 290 — I checked the line, not just that the file was touched.

Closed since last round

1. The merge, and the e2e suite. 1d836ae Reapply "Feat/app e2e harness" undid the revert. All the files I was worried about are on the head: 40 under tests/e2e/, plus src-tauri/src/core/e2e.rs and src-tauri/tauri.e2e.conf.json. Nothing got resolved by deletion.

2. The sdcpp pin. 47525cb is in.

5. The Remote & LAN split. Settled by #300 landing on main as a5ff5f8.

Still open

3. The dist-js files are still there

Unchanged — same four files, same commit (09778f0), still no Cargo.toml, no src/, no package.json, no .gitignore:

src-tauri/plugins/tauri-plugin-foundation-models/dist-js/index.cjs    2,702 B
src-tauri/plugins/tauri-plugin-foundation-models/dist-js/index.d.ts   1,475 B
src-tauri/plugins/tauri-plugin-foundation-models/dist-js/index.js     2,296 B
src-tauri/plugins/tauri-plugin-foundation-models/dist-js/types.d.ts     187 B

Base has zero files under that path. main has the plugin as a 26-file crate. This branch is the only place the compiled output exists, without the source it was compiled from.

One new detail that makes it worth deleting rather than shrugging at — the repo has a test asserting this plugin is gone, tests/desktop-legacy-path.test.mjs:

test('the Foundation Models plugin is gone', () => {
  assert.throws(() =>
    statSync(join(ROOT, 'src-tauri', 'plugins', 'tauri-plugin-foundation-models', 'src'))
  )

It passes — because it probes src/, and dist-js/ is what survived. So we assert the plugin is gone while shipping its build output.

I did check whether it breaks anything before asking again: src-tauri/plugins/package.json globs **, which only matches directories containing a package.json, so the plugin build skips it. It's dead weight, not a hazard. git rm -r the directory.

4. The description is still the template

## Describe Your Changes

-

666 files, 281 commits, three unchecked boxes — and since this fast-forwards the integration branch, it's the record for the whole line. Most of what it needs is in this comment and the last one.

New: make verify-fast fails, and it's the pre-push hook

.husky/pre-push runs make verify-fast. On this head it fails:

node scripts/check-test-quality.mjs
Test-quality guard found new false-confidence patterns:
- [call-only-assertions] web-app/src/containers/__tests__/DownloadManagement.validation-toast.test.tsx

Both tests in that file assert only that a mock was called — toast.loading at :78 and toast.dismiss at :104 — with no observable outcome, and the file isn't in tests/test-quality-allowlist.json.

Worth saying where it came from, because it reflects well on the branch rather than badly: it arrived from main at 41ae3a8 in the v2.0.44 merge. I ran the same guard on origin/main — main fails it on 9 files; this branch is down to 1, because 09f6fcc, bf1605b and a543b9d went back and converted the others. This is the one straggler that landed after that sweep.

09f6fcc is the pattern to copy — one real assertion alongside the mock one (store state, or the DOM) is all the guard wants. Allowlisting it would also close the gate, but the branch has earned the better fix everywhere else.

For the record, the other two verify-fast targets that failed here are my environment, not the branch: test-hardening-contracts fails 6 of 60 on ENOENT for resources/bin/jan-cli and resources/pre-install, and test-coverage-critical dies on Couldn't find the node_modules state file because I never ran yarn install inside extensions/. Both need make install-and-build, which I can't complete here. test-telemetry-props passes.

Checked and fine

• The telemetry consent gate. TELEMETRY_ENABLED defaults to true, which I went looking at sideways for a privacy-first app — but the comment's claim holds: useAnalytic.ts:64 defaults productAnalytic: true, so the Rust gate matches rather than overriding. AnalyticProvider's effect pushes the persisted value on mount, the core is launched --telemetry on|off from the single consent_enabled() source, and only the desktop supervisor ever calls launch_and_wait — so the default-on window is bounded by webview mount, exactly the window the ADR says Rust panics already have. Documented, precedented, and the ADR is unusually good about it.
• ZCode. Patching a third-party agent's config file is already house practice (openclaw_patch_config), the ADR names Apache-2.0, and the lock/backup/refuse-don't-repair handling is more careful than it needed to be. zcode.png is the 13th bundled integration mark, so no new licensing question — I checked the folder before raising one.

Nits

• Prettier. 256 of the 602 changed text files fail --check. Before treating that as a number: 77 of them fail identically on the base branch, and prettier is enforced nowhere — not in .husky/pre-commit (yarn lint --fix --quiet), not in verify-fast, not in CI. So it's house-wide drift, not a gate. What's actually this branch's: 23 files that are clean on base and dirty on head, which is the list worth a prettier --write:

  web-app/src/lib/messages.ts
  web-app/src/lib/agent-file-links.ts
  web-app/src/containers/RenderMarkdown.tsx
  web-app/src/containers/SetupScreen.tsx
  web-app/src/hooks/useAgentApprovalActions.ts
  …and 18 more
  

  The other 157 are new files, take them or leave them.
• Two commits are in twice. 254c29a/8b20b0b and 8647144/bab21a3 are pairs with identical patch-ids — cherry-picked onto this line and also present on the sdcpp line, joined at c11c4c0. I checked the result rather than assuming a clean join: model-logo.ts has one Krea rule, recommended-models.json parses with no duplicate keys. History noise, nothing to fix.
• yarn test still doesn't reach extensions/. Root vitest.config.ts lists only ./core and ./web-app as projects, so 26 test files under extensions/ never run in the root suite — 4 of them new here, including both loadCancel suites. Same gap as the last two rounds, now with more tests sitting behind it.

Delete the dist-js directory, give that one test an observable outcome, and write the description — after that I have nothing. The consent plumbing and the ZCode config merge are both better documented than they needed to be 🧩

---

Generated by Claude Code