Wave 1 — the multi-agent console (#909)

.github/workflows/ci.yml

@@ -2,6 +2,7 @@ name: CI
 
 on:
   pull_request:
+    types: [opened, synchronize, reopened, labeled, unlabeled]
   push:
     branches:
       - main
@@ -466,6 +467,68 @@ jobs:
 
       - name: Run e2e
         run: pnpm e2e
+        env:
+          # The workspace-playground config seeds this clean override from its
+          # committed fixtures, matching a fresh checkout instead of relying on
+          # mutable local playground state.
+          BORING_AGENT_WORKSPACE_ROOT: ${{ runner.temp }}/workspace-playground-e2e
+
+  full-app-e2e:
+    name: Full App E2E
+    needs: changes
+    if: github.event_name == 'push' || needs.changes.outputs.e2e == 'true' || contains(github.event.pull_request.labels.*.name, 'ci:e2e') || contains(github.event.pull_request.labels.*.name, 'ci:full') || contains(github.event.pull_request.labels.*.name, 'release-candidate') || startsWith(github.head_ref, 'release/')
+    runs-on: ubuntu-latest
+    services:
+      postgres:
+        image: postgres:16
+        env:
+          POSTGRES_USER: ubuntu
+          POSTGRES_PASSWORD: test
+          POSTGRES_DB: boring_ui_test
+        ports:
+          - 5432:5432
+        options: >-
+          --health-cmd pg_isready
+          --health-interval 10s
+          --health-timeout 5s
+          --health-retries 5
+    steps:
+      - name: Checkout
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+
+      - name: Setup pnpm
+        uses: pnpm/action-setup@0e279bb959325dab635dd2c09392533439d90093 # v6.0.8
+
+      - name: Setup Node
+        uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
+        with:
+          node-version: 22.19.0
+          cache: pnpm
+
+      - name: Install
+        run: pnpm install --frozen-lockfile
+
+      - name: Build full-app dependencies
+        run: pnpm --filter 'full-app^...' --workspace-concurrency=4 run build
+        env:
+          NODE_OPTIONS: --max-old-space-size=4096
+
+      - name: Cache Playwright browsers
+        uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
+        with:
+          path: ~/.cache/ms-playwright
+          key: ${{ runner.os }}-playwright-${{ hashFiles('pnpm-lock.yaml') }}
+          restore-keys: |
+            ${{ runner.os }}-playwright-
+
+      - name: Install Playwright browser
+        run: pnpm exec playwright install --with-deps chromium
+
+      - name: Run full-app e2e
+        run: pnpm --filter full-app run e2e
+        env:
+          FULL_APP_E2E_DATABASE_URL: postgres://ubuntu:[email protected]:5432/boring_ui_test
+          NODE_OPTIONS: --max-old-space-size=4096
 
   ui-review-fixture:
     name: UI Review
@@ -672,7 +735,7 @@ jobs:
     name: PR Fast Summary
     if: always() && github.event_name == 'pull_request'
     runs-on: ubuntu-latest
-    needs: [action-pins, lint, typecheck, unit-summary, invariants, bundle-budget, core-bundle-budget, e2e, remote-worker-smoke]
+    needs: [action-pins, lint, typecheck, unit-summary, invariants, bundle-budget, core-bundle-budget, e2e, full-app-e2e, remote-worker-smoke]
     steps:
       - name: Verify PR checks
         run: |
@@ -685,6 +748,7 @@ jobs:
             "bundle=${{ needs.bundle-budget.result }}" \
             "core-bundle=${{ needs.core-bundle-budget.result }}" \
             "e2e=${{ needs.e2e.result }}" \
+            "full-app-e2e=${{ needs.full-app-e2e.result }}" \
             "worker-smoke=${{ needs.remote-worker-smoke.result }}"; do
             name=${result%%=*}
             value=${result#*=}
@@ -699,7 +763,7 @@ jobs:
     name: Main Green Summary
     if: always() && github.event_name == 'push'
     runs-on: ubuntu-latest
-    needs: [action-pins, lint, typecheck, bundle-budget, core-bundle-budget, unit-summary, invariants, e2e, remote-worker-smoke]
+    needs: [action-pins, lint, typecheck, bundle-budget, core-bundle-budget, unit-summary, invariants, e2e, full-app-e2e, remote-worker-smoke]
     steps:
       - name: Verify main checks
         run: |
@@ -712,6 +776,7 @@ jobs:
             "unit=${{ needs.unit-summary.result }}" \
             "invariants=${{ needs.invariants.result }}" \
             "e2e=${{ needs.e2e.result }}" \
+            "full-app-e2e=${{ needs.full-app-e2e.result }}" \
             "worker-smoke=${{ needs.remote-worker-smoke.result }}"; do
             name=${result%%=*}
             value=${result#*=}

.github/workflows/release-candidate.yml

@@ -0,0 +1,65 @@
+name: Release Candidate Built-Dist
+
+on:
+  pull_request:
+    types: [opened, synchronize, reopened, labeled]
+  push:
+    branches:
+      - main
+      - "release/**"
+
+permissions:
+  contents: read
+
+concurrency:
+  group: release-candidate-${{ github.event_name == 'pull_request' && github.event.pull_request.number || github.sha }}
+  cancel-in-progress: true
+
+jobs:
+  built-dist:
+    name: Release Candidate Built-Dist
+    if: github.event_name == 'push' || contains(github.event.pull_request.labels.*.name, 'release-candidate') || startsWith(github.head_ref, 'release/')
+    runs-on: ubuntu-latest
+    timeout-minutes: 30
+    steps:
+      - name: Checkout
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+
+      - name: Setup pnpm
+        uses: pnpm/action-setup@0e279bb959325dab635dd2c09392533439d90093 # v6.0.8
+
+      - name: Setup Node
+        uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
+        with:
+          node-version: 24
+          cache: pnpm
+
+      - name: Install
+        run: pnpm install --frozen-lockfile
+
+      - name: Test release controls
+        run: node --test scripts/atomic-release-tag.test.mjs scripts/require-release-candidate-check.test.mjs scripts/validate-release-resume.test.mjs
+
+      - name: Build package and plugin release artifacts
+        run: pnpm build:packages
+        env:
+          NODE_OPTIONS: --max-old-space-size=4096
+
+      - name: Cache Playwright browsers
+        uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
+        with:
+          path: ~/.cache/ms-playwright
+          key: ${{ runner.os }}-playwright-${{ hashFiles('pnpm-lock.yaml') }}
+          restore-keys: |
+            ${{ runner.os }}-playwright-
+
+      - name: Install Playwright browser
+        run: pnpm exec playwright install --with-deps chromium
+
+      - name: Run built-dist release-candidate smoke
+        run: |
+          env -u BORING_RC_BREAK_CSS -u BORING_RC_BREAK_FIRST_SEND \
+            pnpm --filter workspace-playground run test:e2e:release-candidate
+        env:
+          BORING_PLAYGROUND_DIST_ONLY: "1"
+          BORING_AGENT_WORKSPACE_ROOT: ${{ runner.temp }}/workspace-playground-release-candidate

AGENTS.md

@@ -37,6 +37,7 @@ Detailed coding practices, workflow, architecture, and package docs live under `
 | Need | Read |
 | --- | --- |
 | Project/package map | [`docs/README.md`](docs/README.md) |
+| **Single direction / sequencing spine** | [`docs/DIRECTION.md`](docs/DIRECTION.md) |
 | Coding rules | [`docs/kanzen/procedures/coding-rules.md`](docs/kanzen/procedures/coding-rules.md) |
 | Coding invariants | [`docs/kanzen/procedures/coding-invariants.md`](docs/kanzen/procedures/coding-invariants.md) |
 | Repo commands | [`docs/kanzen/procedures/repo-commands.md`](docs/kanzen/procedures/repo-commands.md) |

apps/full-app/.env.example

@@ -1,10 +1,10 @@
 # HTTP
 PORT=3000
-HOST=0.0.0.0
+HOST=127.0.0.1
 LOG_LEVEL=info
 
 # Database
-DATABASE_URL=postgres://postgres:postgres@localhost:5432/boring_ui_v2
+DATABASE_URL=postgres://boring:[email protected]:55439/boring_ui_v2
 
 # Auth
 BETTER_AUTH_SECRET=0000000000000000000000000000000000000000000000000000000000000000
@@ -14,9 +14,18 @@ CORS_ORIGINS=http://localhost:3000,http://localhost:5173
 # Workspace settings encryption (32-byte hex)
 WORKSPACE_SETTINGS_ENCRYPTION_KEY=aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
 
-# Mail transport
-[email protected]
-MAIL_TRANSPORT_URL=console://
+# Local developer login. This is ignored when NODE_ENV=production.
+ENABLE_DEV_LOGIN=1
+
+# Keep email verification disabled for the zero-edit local flow. Uncomment both
+# values to exercise verification/password-reset mail through the console transport.
+# [email protected]
+# MAIL_TRANSPORT_URL=console://
+SEND_WELCOME_EMAIL=false
+
+# Keep local workspaces and Pi sessions outside the checkout.
+BORING_AGENT_WORKSPACE_ROOT=/tmp/boring-full-app-workspaces
+BORING_AGENT_SESSION_ROOT=/tmp/boring-full-app-pi-sessions
 
 # Optional
 # Generic MCP/Sources plugin. COMPOSIO_API_KEY is read server-side only; never expose it through VITE_*.

apps/full-app/README.md

@@ -42,15 +42,24 @@ The smoke boots a local ephemeral `127.0.0.1` Fastify listener, registers `regis
 
 ## Run (local dev)
 
+The committed smoke is the zero-edit proof from a clean checkout after `pnpm install`:
+
+```bash
+pnpm --filter full-app smoke:dev-console
+```
+
+That one command allocates free loopback ports for an isolated Postgres 16 service, the Fastify backend, and Vite; waits for Postgres; builds the app dependencies; applies all migrations using the unedited `.env.example` as its base; starts the full dev app; signs in through the allocated Vite `/dev-login`; reaches the authenticated workspace console; and requires HTTP 202 from one addressed prompt request. It also gives the run unique temporary workspace and session roots and constructs a hermetic child environment, so ambient mail, runtime-mode, or model settings cannot change the proof. It intercepts only browser model discovery so the prompt can be submitted without storing a provider credential; it does not claim a live model response. The script terminates its child process groups, removes its uniquely named Compose service and volume, verifies ownership of its script-created temporary root, and then removes that root when it exits.
+
+For an interactive session, copy the working development defaults without editing them, start Postgres on the example's fixed `127.0.0.1:55439` port, migrate, and run Fastify/Vite on the example's fixed ports 3000/5173:
+
 ```bash
-# from repo root, after `pnpm install`
-cp apps/full-app/.env.example apps/full-app/.env   # then fill in values
-# bring up Postgres, then apply migrations:
+cp apps/full-app/.env.example apps/full-app/.env
+docker compose -f apps/full-app/docker-compose.dev.yml up -d --wait
 pnpm --filter full-app migrate
 pnpm --filter full-app dev
 ```
 
-Open `http://localhost:5173`.
+Open `http://localhost:5173/dev-login`. The `.env` defaults bind the backend and Vite to loopback, enable only the non-production dev-login helper, and intentionally leave mail unset so local accounts are not blocked on email verification. The helper also rejects non-loopback clients, including clients forwarded through the Vite dev proxy. Uncomment both mail values when testing verification flows. Stop Postgres with `docker compose -f apps/full-app/docker-compose.dev.yml down`; its named `postgres-data` volume persists for the next interactive run. Use `down --volumes` only when intentionally resetting local data. Override only the interactive Postgres host port with `BORING_DEV_POSTGRES_PORT`; the smoke always supplies an allocated value and removes its own uniquely named volume.
 
 ### Hosted automation scheduler
 
@@ -77,13 +86,15 @@ available as an operational fallback when the internal scheduler is enabled.
 
 | Script | What it does |
 |--------|--------------|
-| `dev` | Build agent/workspace/core, then `tsx src/server/dev.ts` (Vite :5173 + Fastify) |
-| `build` | Build packages, then `build-app.mts` (frontend → `dist/front`, server → `dist/server`) |
+| `build:deps` | Topologically build the full-app dependency selector with bounded concurrency |
+| `dev` | Build dependencies, load `.env` when present, then start Vite :5173 + Fastify |
+| `build` | Build dependencies, then `build-app.mts` (frontend → `dist/front`, server → `dist/server`) |
 | `start` | `node dist/server/main.js` (prod, listens on `PORT`) |
 | `start:worker` | `node dist/server/agent-worker.js` — provider-neutral remote-worker process |
-| `migrate` | `tsx src/server/migrate.ts` — apply DB migrations |
+| `migrate` | Load `.env` when present, then apply core and automation DB migrations |
 | `typecheck` / `lint` | `tsc --noEmit` (`lint` is an alias of `typecheck`) |
 | `e2e` / `e2e:smoke` | Playwright against `e2e/playwright.config.ts` (the two scripts are identical) |
+| `smoke:dev-console` | Zero-edit browser prompt proof using dynamically allocated loopback ports and `.env.example` defaults |
 | `smoke:mcp-managed-agent` | `node --import tsx scripts/managed-agent-mcp-smoke.ts` — local stock MCP client smoke for `/mcp/managed-agent` |
 | `smoke:remote-worker` | `node scripts/remote-worker-smoke.mjs` — local remote-worker contract/isolation smoke |
 
@@ -117,13 +128,12 @@ From `.env.example` and code. Required for a working server:
 | `DATABASE_URL` | Postgres connection string |
 | `BETTER_AUTH_SECRET`, `BETTER_AUTH_URL` | Auth secret + base URL |
 | `WORKSPACE_SETTINGS_ENCRYPTION_KEY` | 32-byte hex; encrypts per-workspace settings |
-| `MAIL_FROM`, `MAIL_TRANSPORT_URL` | Mail transport (`console://` for dev) |
 
 Common optional:
 
 | Var | Default | Notes |
 |-----|---------|-------|
-| `PORT` / `HOST` / `LOG_LEVEL` | `3000` / `0.0.0.0` / `info` | HTTP server |
+| `PORT` / `HOST` / `LOG_LEVEL` | `3000` / `0.0.0.0` / `info` | HTTP server. The local `.env.example` narrows `HOST` to `127.0.0.1`. |
 | `CORS_ORIGINS` | `http://localhost:3000,http://localhost:5173` | Allowed origins |
 | `BORING_PLUGIN_AUTHORING` | `0` | `1` installs the plugin-authoring surface |
 | `BORING_AUTOMATION_INTERNAL_SCHEDULER` | `true` | Set to `false` only when an external scheduler owns hosted Automation wake-ups |
@@ -142,6 +152,10 @@ Common optional:
 | `COMPOSIO_API_KEY` | — | Optional server-only managed connector credential resolved by the app's boring-mcp managed connector secret resolver. Do not create a `VITE_*` mirror. |
 | `BORING_MCP_MAX_READONLY_INPUT_BYTES` | `65536` | Governed read-only MCP call input limit. |
 
+### Agent fleet composition
+
+Both full-app entrypoints pass `createFullAppAgentFleetComposition()` into `createCoreWorkspaceAgentServer`. That `agents` option uses the same configured-agent shape as workspace-playground: each entry has an `agentTypeId`, distinct `definition.instructions`, and its own encoded `model.preferred`, accompanied by `defaultAgentTypeId` and the app-owned `fleetCompiler`. Full-app currently configures `default` and `dummy`; use `BORING_AGENT_DEFAULT_MODEL` and `BORING_AGENT_DUMMY_MODEL` to override their preferred models. Host compatibility tools remain scoped into every addressed binding, while an app that wants per-agent plugin selection should list the preflighted plugin IDs in each agent's `plugins` bindings.
+
 ### Local dev login
 
 For local development only, the dev server can expose a one-click login helper:
@@ -153,10 +167,10 @@ ENABLE_DEV_LOGIN=1 pnpm --filter full-app dev
 Then open:
 
 ```txt
-http://localhost:3000/dev-login
+http://localhost:5173/dev-login
 ```
 
-The route signs in `DEV_LOGIN_EMAIL` (default `[email protected]`) or creates it if missing, sets the normal Better Auth session cookie, and redirects to `/`. The core dev server proxies `/dev-login` from the frontend port to the API server. The route is unavailable unless `ENABLE_DEV_LOGIN=1` and is ignored in `NODE_ENV=production`.
+The route signs in `DEV_LOGIN_EMAIL` (default `[email protected]`) or creates it if missing, sets the normal Better Auth session cookie, and redirects to `/`. Use the Vite URL above so the relative redirect lands on the frontend rather than the API-only development port. The core dev server proxies `/dev-login` to the API server while preserving the observed client address for the route's loopback-only check. The route rejects non-loopback clients, is unavailable unless `ENABLE_DEV_LOGIN=1`, and is ignored in `NODE_ENV=production`. Email verification follows ordinary mail configuration: the unedited development env leaves mail unset, while production and explicit mail-enabled testing retain the normal verification flow.
 
 ## Container reference
 

apps/full-app/docker-compose.dev.yml

@@ -0,0 +1,19 @@
+services:
+  postgres:
+    image: postgres:16-alpine
+    environment:
+      POSTGRES_USER: boring
+      POSTGRES_PASSWORD: boring
+      POSTGRES_DB: boring_ui_v2
+    ports:
+      - "127.0.0.1:${BORING_DEV_POSTGRES_PORT:-55439}:5432"
+    volumes:
+      - postgres-data:/var/lib/postgresql/data
+    healthcheck:
+      test: ["CMD-SHELL", "pg_isready -U boring -d boring_ui_v2"]
+      interval: 1s
+      timeout: 5s
+      retries: 30
+
+volumes:
+  postgres-data:

apps/full-app/e2e/google-auth-webserver.sh

@@ -12,11 +12,12 @@ cleanup() {
 trap cleanup EXIT
 
 printf '[features]\ngoogle_oauth = true\n' > "$CONFIG_PATH"
+ln -s "$APP_DIR/node_modules" "$CONFIG_DIR/node_modules"
 
 cd "$APP_DIR"
 pnpm --filter @hachej/boring-core exec tsup --no-dts
 pnpm --filter @hachej/boring-core exec sh -c "cp src/front/theme.css dist/front/theme.css"
 pnpm migrate
 pnpm build
 cd "$CONFIG_DIR"
-exec env NODE_ENV=production node "$APP_DIR/dist/server/main.js"
+exec env NODE_ENV=production BORING_APP_CONFIG_PATH="$CONFIG_PATH" node "$APP_DIR/dist/server/main.js"

apps/full-app/e2e/playwright.config.ts

@@ -34,12 +34,19 @@ const webServerEnv = Object.fromEntries(
     MAIL_TRANSPORT_URL: mailTransportUrl,
     PORT: String(apiPort),
     CSP_ENABLED: 'true',
+    BORING_ALLOW_UNSAFE_AGENT_MODE: '1',
   }).filter((entry): entry is [string, string] => typeof entry[1] === 'string'),
 )
 
 export default defineConfig({
   testDir: '.',
-  testMatch: ['smoke.spec.ts', 'csp.spec.ts', 'workspace-lifecycle.spec.ts', 'google-signup.spec.ts', 'runtime-readiness.spec.ts'],
+  testMatch: [
+    'smoke.spec.ts',
+    'csp.spec.ts',
+    'workspace-lifecycle.spec.ts',
+    'google-signup.spec.ts',
+    'runtime-readiness.spec.ts',
+  ],
   fullyParallel: false,
   workers: 1,
   retries: isCI ? 1 : 0,