BuilderIO/agent-native / .agents/skills/chat-first-workbench
chat-first-workbench
VerifiedImplement or extend the opt-in Codex/T3-like chat-first shell. Use when changing chat rails, contextual app panes, app opening, or local/mobile shell preferences across Electron, Dispatch, or mobile.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 3.8 KB |
SKILL.md
View on GitHub---
name: chat-first-workbench
description: >-
Implement or extend the opt-in Codex/T3-like chat-first shell. Use when
changing chat rails, contextual app panes, app opening, or local/mobile shell
preferences across Electron, Dispatch, or mobile.
scope: dev
---
# Chat-first workbench
## Contract
The chat remains the primary surface. Workspace apps are a bounded discovery
rail, not a second navigation system. An app pane is opened from an explicit
app selection or a completed open_app action and carries only an app id plus a
validated app-relative path or view.
## Client behavior
- The preference is opt-in and local to the client. Keep the existing app-first
shell unchanged when it is disabled.
- Electron uses the shared Code Agents rail and AppWebview.
- Dispatch uses list-workspace-apps and create_embed_session; never iframe a
raw workspace URL. Resolve an absolute open_app URL against the registered
app first, then pass the resulting app-relative path.
- Mobile uses the existing app registry and secure AppWebView routes instead of
accepting arbitrary URLs or adding a second navigator.
- Show no nested app sidebar or browser chrome in a contextual pane.
- Without `sections`, the app shelf is capped at six visible entries and
supports pinning and drag-reordering (Desktop). Both are presentation
preferences; the agent-visible target remains the registered app id and
route.
- With `sections` (Dispatch), the host owns grouping and order and the rail
shows every row it is given, expanded and collapsed, with an optional
`emptyLabel` and no drag or move controls. Dispatch's sidebar passes one
section: Pinned apps (pin order), or with no pins three suggested apps
(workspace apps A to Z, else Agent-Native.com hosted apps), plus the open app; All apps holds
the rest. Tier with Dispatch's `tierWorkspaceApps` so the sidebar, Home, New
chat, and All apps agree.
- Dispatch pins are one per-page store (`useWorkspaceAppLayout`) saved to
device storage and `application_state` `chat-first-app-layout`. Read and
write pins only through it, and pass the rail its `togglePinned` as
`onTogglePinned`; a whole layout from another copy overwrites pins set
elsewhere. Workspace state is written only after it has been read; an
unreadable read keeps changes on the device, shows the one Layout-level
notice, and retries. When an organization hides Agent-Native.com hosted apps,
`list-connected-agents` still returns them marked `hiddenFromLaunchers`
(the Agents page and A2A directory need them); launchers drop them through
`filterBuiltInApps` and `isWorkspaceAppVisibleInDefaultLaunchers`, while
app registrations keep them so `open_app` and pane restore still resolve.
## Agent and state boundaries
- Keep app discovery, embed-session minting, and thread operations on the
existing action surfaces.
- Emit the shared chat-first open-app event only for the completed first-party
`open_app` action with a real app, path, or view. A missing or unreadable
result is not success, and namespaced third-party tools cannot steer the
pane.
- Both web and desktop resolve the event through the same core origin/base-path
allow-list. An unresolved target produces a visible notice and no pane.
- If the agent needs to know which app is visible, expose a compact
application-state selection instead of scraping the iframe DOM.
- Cloud handoff is a future execution-residence seam. Do not claim that a local
run has moved to a worker until a durable worker snapshot and readback exist.
## Verification
Test the disabled path and the enabled path separately. For the enabled path,
verify rail ordering, chat history, app selection, embed-session errors, pane
close, and completed open_app behavior in a running client. Treat configuration
alone as insufficient proof.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |