| name | openforge-app-operator |
| description | Launch, click through, and smoke-check the OpenForge Electron desktop app, Rust sidecar, plugins, terminal, and CLI bridge while validating fixes. |
openforge-app-operator
Use this when a fix needs validation in the running OpenForge app rather than only unit tests. Keep default smoke checks read-only and non-destructive: do not create, update, delete, move, or start tasks/agents unless the task explicitly requires that behavior or the user approves it.
Start the dev app
From the repository root:
pnpm i
pnpm electron:dev
pnpm electron:dev starts Vite, builds/launches the Rust sidecar, builds Electron main/preload, then opens the desktop shell.
If you need automation or screenshots, run the same app with an Electron remote debugging port after building the Electron main process:
pnpm exec vite --host 127.0.0.1
pnpm electron:build
OPENFORGE_BACKEND_PORT=17645 \
OPENFORGE_HTTP_PORT=17645 \
OPENFORGE_SIDECAR_PATH=/path/to/openforge-sidecar \
OPENFORGE_ELECTRON_SIDECAR=1 \
ELECTRON_RENDERER_URL=http://127.0.0.1:1420 \
pnpm exec electron --remote-debugging-port=9224 .
Use a free backend/debugging port if those are already taken.
For plugin-page audits, build bundled plugin artifacts before launch so built-in views can register:
pnpm dev:plugin-artifacts
When using a custom dev launch, isolate app data and user data so audits can safely create projects or enable plugins without mutating the installed app:
OPENFORGE_APP_DATA_DIR="$PWD/.openforge-dev/sidecar-app-data" \
OPENFORGE_ELECTRON_USER_DATA_DIR="$PWD/.openforge-dev/user-data" \
OPENFORGE_BACKEND_PORT=17645 \
OPENFORGE_HTTP_PORT=17645 \
OPENFORGE_SIDECAR_PATH=/path/to/openforge-sidecar \
OPENFORGE_ELECTRON_SIDECAR=1 \
ELECTRON_RENDERER_URL=http://127.0.0.1:1420 \
pnpm exec electron --remote-debugging-port=9224 .
Before restarting custom CDP launches, stop stale Vite/Electron/sidecar processes for the same worktree/ports. A stale sidecar can make the new Electron instance fail readiness with an invalid backend authorization token.
Confirm the CLI bridge
In a second shell, use the installed launcher only:
openforge project list
"$HOME/.openforge/bin/openforge" project list
If the dev app uses a non-default bridge port, set it before invoking the launcher:
OPENFORGE_HTTP_PORT=17645 openforge project list
Do not call the underlying cli.js directly.
Read-only CLI checks are safe defaults. Discover IDs first, then use the relevant project/task for the current app context:
openforge project list
openforge task list --project-id <project-id> --state doing
openforge task get --task-id <current-task-id>
Skip task-specific lookups when no current task id is known. Avoid CLI commands that mutate data (task create, task update, deletes, provider starts) unless the validation specifically requires them.
Basic click-through map
Click around only enough to prove the changed area works. Prefer observation over mutation.
Sidebar and projects
- Click project/avatar dots in the left rail and confirm the selected project changes or project context remains coherent.
- Click the main navigation icons: Board, Files, PR/GitHub review, Skills, Terminal, Task Schedules, and Settings where present.
- Check that the active icon, page title, and content agree after navigation.
- For audits that need plugin pages, do not stop at the empty/default app state. Add or select a real project and enable built-in plugins for that project when the user has approved state mutation.
Board and task detail
- Board should load with task columns/counts and no blocking renderer errors.
- Select an existing task card and confirm the detail pane updates with prompt, summary, comments/review context, and status.
- Use “Open full view” only as a read-only navigation check; return to the board afterward.
- If task context menus are involved, verify the menu opens and contains Start Task, Move to, and Delete, but do not choose destructive actions.
Settings
- Open Settings and confirm global and project settings render.
- For provider or GitHub settings changes, verify fields/sections appear, but do not overwrite credentials during smoke validation.
Skills
- Open Skills (
Cmd+L when shortcuts are enabled) and confirm project/personal skills are grouped by source.
- Select the edited skill and verify name, description, source, and markdown content render.
- For skill discovery changes, also run:
find .agents/skills -maxdepth 2 -name SKILL.md | sort
PR/GitHub review
- Open the PR/GitHub review view and confirm assigned PRs, diffs, comments, or empty states render.
- For GitHub polling/comment changes, verify status text and errors are visible without submitting reviews unless requested.
Files
- Open Files and expand a project tree if available.
- Select a safe text/markdown file and confirm the viewer renders content.
- In a freshly seeded dev app, Files may be absent until the File Viewer plugin is enabled for the active project.
- Do not edit or delete files from the app during a smoke check.
Terminal and provider runtime
- Open Terminal or task terminal areas only when the change touches shells, PTY, provider sessions, hooks, or streaming events.
- For shell-tab changes, verify output streams, exit state, tab switching, and stale event filtering.
- Do not start or stop agents unless the task explicitly requires provider lifecycle validation.
- Remember provider/runtime cautions: frontend invoke payloads use camelCase keys, multi-shell PTY state is shell-key scoped (
${taskId}-shell-${index}), and terminal lifecycle ownership belongs in src/lib/terminalPool.ts.
Shortcuts and focus
- Use app shortcuts when validating navigation:
Cmd+H Board, Cmd+G PR Review, Cmd+L Skills, Cmd+, Settings.
- Check plain-key vim bindings only when relevant and never while an input is focused.
- Confirm Escape/back navigation returns to a stable view when that behavior is in scope.
Screenshots and CDP
Prefer native screenshots when available:
screencapture -x /tmp/openforge-dev-app.png
If native screenshots are unavailable in the harness, use Electron/Chromium DevTools Protocol via the remote debugging port:
- Start Electron with
--remote-debugging-port=<port>.
- Before CDP calls, poll
http://127.0.0.1:<debug-port>/json/list until a page target exists.
- Verify the OpenForge HTTP bridge port responds before app assertions.
- Connect to the page target WebSocket.
- Use
Runtime.evaluate, Input.dispatchMouseEvent, and Page.captureScreenshot for read-only click-through and screenshots.
Initial ECONNREFUSED during startup is not an app failure until readiness polling times out. Record the screenshot path and the views clicked.
For approved stateful audits, CDP can use the Electron preload bridge instead of brittle form clicking. Use camelCase payload keys exactly as renderer IPC wrappers do:
const invoke = window.openforge.invoke.bind(window.openforge)
const projectPath = '/absolute/path/to/project-worktree'
const projectName = 'openforge/KVG-1450'
const projects = await invoke('get_projects', null)
const project = projects.find((row) => row.path === projectPath)
?? await invoke('create_project', { name: projectName, path: projectPath })
const plugins = await invoke('list_plugins', null)
for (const plugin of plugins) {
await invoke('set_plugin_enabled', { projectId: project.id, pluginId: plugin.id, enabled: true })
}
const enabled = await invoke('get_enabled_plugins', { projectId: project.id })
After enabling plugins, reload the renderer and verify the rail contains Files, Pull Requests, Skills, Terminal, and Task Schedules. Capture one screenshot/text summary per reachable page, not just the initial Board/Settings state.
Conditional verification gates
Choose gates based on changed files and risk:
- Frontend/shared TypeScript:
pnpm exec tsc --noEmit and focused pnpm test <path-or-pattern> or pnpm exec vitest run <path-or-pattern>. Do not use separator forms such as pnpm test -- <path> or pnpm test -- --run <path>.
- Broad frontend behavior:
pnpm test.
- Rust/backend sidecar:
cd src-tauri && cargo test; use cargo check for quick Rust-only validation.
- IPC/Electron bridge: IPC contract, backend bridge, preload, event forwarder, and TypeScript tests.
- Plugins/Skills view: Skills view tests, plugin SDK tests, plugin protocol/views tests, and
pnpm build:plugins when packaging/runtime changes.
- Terminal/PTY/provider runtime: terminal pool/session tests plus app runtime smoke checks for shells and exits.
- UI-only changes: validate behavior/accessibility; do not assert Tailwind/daisyUI class names.
Cleanup
- Stop
pnpm electron:dev with Ctrl+C.
- If you launched separate Vite/Electron/CDP processes, kill only the dev processes you started.
- Ensure provider agents, shells, or sidecars started for validation are stopped before finishing.
Reporting
Include:
- Files changed.
- Commands run and pass/fail status.
- CLI bridge check and port used.
- App sections clicked and what was observed.
- Screenshot path if captured.
- Cleanup performed.
- Any skipped gates with a brief reason.
- Open risks requiring a human, credentials, network, or broader environment.
