feature: Workspace Research And Planning

.gitignore

@@ -78,3 +78,5 @@ _workspace*
 /.nexus-brain/
 /.nexus-collab/
 
+/.adw/config.yaml
+/.adw/templates/.prefs.json

CLAUDE.md

@@ -63,7 +63,7 @@ Keep the mental model high-level:
 - `src/hooks/` — reusable editor and data hooks
 - `src/types/` — shared type definitions
 - `docs/tasks/` — task-specific plans and notes
-- `packages/` — auxiliary packages such as `nexus-acp-bridge`
+- `packages/` — auxiliary packages such as `nexus-acp-bridge` (see `packages/nexus-acp-bridge/CLAUDE.md` for package-scoped guidance)
 
 ---
 

README.md

@@ -28,6 +28,14 @@ Nexus is a visual workflow editor for designing, composing, and exporting AI wor
 - Export generated files as a ZIP or write them directly into a target folder
 - Include generated `run-<workflow>.sh` and `run-<workflow>.bat` helper scripts with exported workflow artifacts
 
+### 🔎 Workspace Research and Planning
+
+- Open a workspace-native Research surface from every workspace dashboard at `/workspace/[id]/research`
+- Create research spaces from Research Brief, PRD, Implementation Plan, and Decision Log templates
+- Capture tile-based notes, tasks, sources, relationships, syntheses, and promote curated findings into Brain
+- Import/export portable `.nodepad` files and copy/export research as markdown
+- Research editing stays local-first when optional AI connector features are disconnected
+
 ### 📝 Content and Agent Authoring
 
 - Fullscreen editing for prompts and documents
@@ -119,16 +127,48 @@ bun run docker:down
 
 ### 3. Use AI features (optional)
 
-AI features require a running [OpenCode](https://github.com/nichochar/opencode) server.
+AI features require a running [OpenCode](https://github.com/nichochar/opencode) server **or** a compatible bridge endpoint such as the bundled ACP bridge.
+
+Nexus now defaults its connection UI to the bundled ACP bridge endpoint (`http://127.0.0.1:4080`). The connect dialog includes quick-start presets for:
+
+- Claude Code via the ACP bridge
+- OpenCode via the ACP bridge
+- Direct OpenCode server mode
 
 ```bash
 bun add -g opencode-ai
 opencode serve --cors http://localhost:3000
 ```
 
-Then connect from the Nexus header.
+Then connect from the Nexus header. If you prefer the bundled bridge, use one of the ACP presets below instead.
+
+### Optional: run the bundled ACP bridge
+
+The repository also includes a minimal ACP bridge under `packages/nexus-acp-bridge/`. It exposes the subset of the OpenCode-style HTTP/SSE API that Nexus currently uses, so you can point Nexus at the bridge URL instead of a direct OpenCode server.
+
+```bash
+bun run nexus-acp-bridge
+```
+
+That command auto-loads the checked-in defaults in `packages/nexus-acp-bridge/.env.defaults`, which currently select the `claude-code` preset. Customise it with simple flags:
+
+```bash
+bun run nexus-acp-bridge --agent claude --cors http://localhost:3000
+bun run nexus-acp-bridge --agent codex
+bun run nexus-acp-bridge --agent opencode
+```
+
+The first time you run `--agent claude`, the bridge auto-vendors `@agentclientprotocol/claude-agent-acp`. Pass `--no-auto-setup` to opt out. Other supported flags include `--port`, `--host`, and repeatable `--project-dir`. Environment variables still override everything.
+
+By default the bridge listens on:
+
+```text
+http://127.0.0.1:4080
+```
+
+You can configure it with the environment variables documented in `packages/nexus-acp-bridge/README.md` and the example file at `packages/nexus-acp-bridge/examples/.env.claude.example`.
 
-OpenCode is only required for AI-powered features such as:
+An OpenCode server or compatible bridge is only required for AI-powered features such as:
 
 - AI workflow generation
 - AI prompt generation and editing
@@ -334,6 +374,7 @@ The timer starts after the initial refresh on server startup completes. Manual r
 | `Ctrl/Cmd + Alt + E` | Export workflow JSON |
 | `Ctrl/Cmd + Alt + G` | Open generate/export dialog |
 | `Ctrl/Cmd + Alt + A` | AI generate workflow |
+| `Ctrl/Cmd + Alt + I` | Toggle AI side-kick |
 | `Ctrl/Cmd + Alt + P` | Preview generated output |
 | `H` / `V` | Hand tool / Selection tool |
 | `?` | View all shortcuts |

docs/tasks/ai-sidekick-acp-ux-caa41bd1/e2e-ai-sidekick-acp-ux-caa41bd1.md

@@ -0,0 +1,37 @@
+# E2E Specification — AI Side-Kick ACP UX
+
+## User Story
+
+Validate that a workflow author can open the persistent AI side-kick, ask it to inspect the canvas, apply Nexus workflow actions, approve or deny destructive actions, and respond to forwarded ACP permission requests without leaving the editor.
+
+## Test Steps
+
+> Browser driving is reserved for the E2E pipeline and should use `playwright-cli` there only.
+
+1. Open the side-kick via the header button and capture screenshot `sidekick-open-empty`.
+2. Send `What can you tell me about this canvas?` and assert an assistant text response appears with no Nexus action card.
+3. Send `Add a Prompt node named Draft Prompt and connect it after Start` using a mocked/deterministic bridge response, then assert an `addNode` card is `done`, a `connectNodes` card is `done`, and the canvas contains `Draft Prompt`; capture `sidekick-action-success`.
+4. Select the created node and send `Delete this node`, assert a destructive action card appears with `Allow once`, `Allow always`, and `Deny`; capture `sidekick-approval-card`.
+5. Click `Deny`, assert the node remains and the card status is `denied`.
+6. Repeat delete, click `Allow once`, assert the node is removed and status is `done`.
+7. Trigger a mocked forwarded ACP permission request, assert the permission card displays option buttons, choose `allow_once`, and assert it becomes resolved; capture `sidekick-permission-resolved`.
+8. Click `New conversation`, assert message history clears and side-kick remains open; capture `sidekick-new-conversation`.
+
+## Success Criteria
+
+- Side-kick opens from the header and remains anchored in the bottom-right editor area.
+- Text-only assistant responses render as assistant messages with no action cards.
+- Nexus action cards show action names, arguments, and terminal `done` or `error` state.
+- Destructive actions show exactly `Allow once`, `Allow always`, and `Deny` while awaiting approval.
+- Denying a destructive action leaves the target node on the canvas and marks the card `denied`.
+- Allowing once removes the target node and marks the card `done`.
+- Forwarded ACP permission cards show option buttons and transition to resolved after choosing an option.
+- New conversation clears visible message history, resets approvals, and keeps the panel open.
+
+## Screenshot Capture Points
+
+- `sidekick-open-empty`
+- `sidekick-action-success`
+- `sidekick-approval-card`
+- `sidekick-permission-resolved`
+- `sidekick-new-conversation`