Skip to content

feat(desktop): migrate CookieMonster workflows to OpenCode V2 - #100

Merged
trmosala merged 106 commits into
devfrom
v2-runtime
Nov 8, 2026
Merged

trmosala merged 106 commits into
devfrom
v2-runtime

Conversation

@trmosala

@trmosala trmosala commented Nov 7, 2026

Copy link
Copy Markdown
Owner

Issue for this PR

User-requested CookieMonster V2 migration, workflow sidebar, and split chats; no separate issue.

Type of change

  • New feature
  • Bug fix
  • Refactor / code improvement

What does this PR do?

CookieMonster now runs the pinned OpenCode 2.2.5 runtime in Electron's Node utility process. It keeps the Working, Snoozed and Settled sidebar workflow and two-chat drag-to-split interface while porting browser tools, scheduling, managed skills and the After Effects adapter to V2. This includes the preceding upstream 1.19.5 update and workflow work.

Startup imports history into a separate V2 database and preserves the originals. Follow-up prompts apply the selected model, variant and agent before admission. CI now includes the protected dev branch and uses GitHub-hosted runners available to this fork.

How did you verify your code works?

  • Root typecheck: 33/33 tasks. Production Chromium workflow/split regressions: 27/27.
  • Actual V2 server integration: 17 tests; runtime/AE: 7; compatibility: 15; desktop: 39; browser tools: 108; post-rebase WPP bridge: 500.
  • Built and packaged the complete Electron app with native dependencies. Observed native prompts, interruption, real input requests, workflow persistence, split chat interaction and model/agent switching using an explicitly local scripted model.
  • Reproduced all three Sol findings before fixing them. Sol independently reran 39 tests and approved the fixes.

Limits: the broader app suite has 854 passing tests and 10 existing localization failures reproduced before migration. A broad browser-unit run exhausted memory. Hosted free inference returned HTTP 429; successful native inference uses local fixtures. Live WPP SSO, macOS AE/aerender and signed macOS packaging remain unverified. The Linux unpacked build is validation evidence, not a release target.

Screenshots / recordings

Native Electron UI with a local fixture model:

Split chats

V2 model and agent selection

Checklist

  • I have tested my changes locally
  • I have not included unrelated changes in this PR

Reviewed with sol-loop: 2 rounds with gpt-6.1-sol. Sol verdict: approved on high (implementation head 461308c; subsequent CI routing and screenshot commit contains no runtime changes).

Created with GPT-6 Astra and GPT-6.1 Sol in Codex, and Claude Opus 5.5 in Claude Agent through T3 Code.

opencode and others added 30 commits September 30, 2026 22:39
Co-authored-by: rekram1-node <rekram1-node@users.noreply.github.com>
Co-authored-by: adamdotdevin <2363879+adamdotdevin@users.noreply.github.com>
opencode and others added 24 commits October 28, 2026 06:17
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A malformed /api/fs/find response threw inside the fulfilled handler, which
the sibling rejection handler could not catch, and the palette's resource
error took down the whole session page.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Mock V2 forms, permission decisions, archive-by-update, fs/find, info and
  location routes; send declared error statuses the 2.2.5 client keeps.
- Open the collapsed sidebar Projects section and use its Open project action.
- Follow CM3's deliberate side-panel changes from dev (no automatic Review
  tab; header and its tablist live inside the panel) and its dark landing.
- Scroll pinned-footer-hidden projects into view before dragging.
- Wait for split navigation and snooze to settle before acting, removing
  races that typed into an outgoing composer or woke a not-yet-snoozed chat.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
test-browser/cm-loading.test.ts launches a real Chromium, but only the e2e
job installed browsers, so unit (linux) failed with a missing executable.
Reuse the e2e job's cached install steps.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Turbo's strict env mode dropped the variable, so the browser test looked in
~/.cache/ms-playwright instead of the cached install location.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A single 220ms window between screenshots stalled on the Windows runner,
giving identical frames. Keep sampling for up to ~2s until a frame differs.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Windows Chromium does not restart an SVG <image>'s inner CSS animation
after it returns from display:none, so the frame diff after toggling
reduced motion never changed there. Check frames on first paint and
assert the motion image is restored once reduced motion is off.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Windows headless Chromium returned identical captures of the animated SVG
<image> for over 2s, both on first paint and after reduced motion is
cleared. Keep every other loader assertion on all platforms.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The new-session column grew from 25% down with no bottom bound, so the
CM3 banner, heading and starters slid under the bottom-pinned provider
tip, which intercepted clicks on "Trace a bug". Bound the column above
the tip and let it scroll on short windows.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…tpApi exerciser

Routes merged from dev had no scenarios, failing --fail-on-missing.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Without a system rg, core's ripgrep tests fall back to downloading and
Expand-Archive-ing ripgrep through PowerShell, which fails or exceeds the
5s test timeout on windows-2025.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
CookieMonster-branded builds show a full-width film of the new session
workflow and split view once on the first launch of 2.2.5 (fresh installs
included), in place of the upstream changelog for that version. It honours
the existing release-notes setting and reuses its strings.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The shelf assertions are vertical placement and expansion, which text
direction does not affect; keep one width per direction and drop the
debug screenshots.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
opencode's external-directory and read tests strip the drive letter from
a temp path, which resolves against the D: checkout while TEMP defaults
to C: on windows-2025 runners.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Without a system rg, core's tests download ripgrep from GitHub releases
at test time, so a releases outage (504s) fails glob, grep and skill
tests unrelated to the change under test.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.