docs: fix inaccuracies found auditing docs against the code

Audited all repository docs against the v1.3.0 source and corrected
claims that no longer match the code:

- MISSION_API: command_source/trigger JSON shapes are internally tagged
  ({"type": ...}); inline uses `content` not `command`; remove the
  non-existent GET /api/control/tree row; SSE status uses
  `waiting_for_tool`; AutomationExecution exposes `variables_used`/`error`
  (not `started_at`/`result_message`).
- WORKSPACE_API/WORKSPACES: drop the non-existent `tools` field and `stdin`
  exec input; init log is /var/log/sandboxed-init.log; helper scripts are
  sandboxed-network-up / sandboxed-tailscale-up.
- BACKEND_API: update-config response message no longer mentions a restart.
- install-docker: compose service is `sandboxed-sh`, volume `sandboxed-data`;
  list all five MCP binaries; desktop stack is Wayland (Sway), not X11.
- install-native: remove the non-existent in-repo residential.json template
  path; fix helper script names; dev server runs on :3001.
- DESKTOP_SETUP: env var is DESKTOP_DISPLAY (not DESKTOP_DISPLAY_START).
- dashboard README: JWT stored in localStorage; `bun dev` already serves :3001.
- android README: minSdk is 33 (Android 13).
- PROVIDER_HEALTH_CHECK: Z.AI health model is glm-4.7-flash.
- STREAMING/HERMES/mobile-app-reliability: src/api/control.rs is now
  src/api/control/mod.rs.
- ASK_ASSISTANT_DESIGN: ask tool loop cap is 10 iterations.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Author: adrienlacombe

ASK_ASSISTANT_DESIGN.md

@@ -86,7 +86,7 @@ Store: `src/api/ask/store.rs` (`AskStore`). Global singleton via tokio
   `MetadataLlmConfig`; forwards `reasoning_effort` so reasoning models return
   visible content).
 - `mod.rs` — `run_ask_turn`: persist user msg → seed system prompt (recent
-  events + summary) → tool loop (≤6 iterations) → persist assistant answer.
+  events + summary) → tool loop (≤10 iterations) → persist assistant answer.
 - `http.rs` — handlers.
 
 Tools (read-on-demand so the 128K context ceiling never bites):

INSTALL.md

@@ -30,6 +30,6 @@ systemd-nspawn container isolation.
 | **Platform** | Any OS with Docker | Ubuntu 24.04 LTS |
 | **Setup time** | ~5 minutes | ~30 minutes |
 | **Container workspaces** | Yes (with `privileged: true`) | Yes (native systemd-nspawn) |
-| **Desktop automation** | Yes (headless Xvfb) | Yes (native X11 or Xvfb) |
+| **Desktop automation** | Yes (headless Wayland/Sway) | Yes (native X11 or Xvfb) |
 | **Performance** | Good (slight overhead on macOS) | Best (native Linux) |
 | **Updates** | `docker compose build && up -d` | Git pull + cargo build, or one-click from dashboard |

PROVIDER_HEALTH_CHECK.md

@@ -59,7 +59,7 @@ curl -X POST https://YOUR-BACKEND/api/ai/providers/cerebras/health
 
 The health check performs actual API validation for:
 - **Cerebras** - Tests with `llama-3.1-8b` model
-- **Z.AI** - Tests with `glm-4-flash` model
+- **Z.AI** - Tests with `glm-4.7-flash` model
 - **DeepInfra** - Tests with `Meta-Llama-3.1-8B-Instruct` model
 
 For OAuth-based providers (Anthropic, OpenAI, Google), it verifies credentials exist without making test API calls.

android_dashboard/README.md

@@ -8,7 +8,7 @@ The Android app is published on Zapstore:
 
 https://zapstore.dev/apps/sh.sandboxed.dashboard
 
-## What's in v0.2.0
+## What's in the app
 
 ### Bottom-tab screens
 
@@ -43,7 +43,7 @@ https://zapstore.dev/apps/sh.sandboxed.dashboard
 | --- | --- |
 | Language | Kotlin 2.0.21 |
 | Build | AGP 8.9.1, Gradle 8.11.1 |
-| SDK | `compileSdk` 36, `targetSdk` 36, `minSdk` 26 (Android 8.0) |
+| SDK | `compileSdk` 36, `targetSdk` 36, `minSdk` 33 (Android 13) |
 | UI | Jetpack Compose (BOM 2024.12.01), Material 3 with `material-icons-extended` |
 | Navigation | `androidx.navigation:navigation-compose` 2.9.8 |
 | State | ViewModel + StateFlow, Compose `collectAsState` |

backend/STREAMING.md

@@ -203,4 +203,4 @@ canonical row rather than the op log.
   reference implementation of the continuation rule.
 - `src/api/control_metrics.rs` records p50/p99 SSE chunk sizes so we
   can spot regressions in the streaming contract.
-- `src/api/control.rs::stream()` is the server-side mission filter.
+- `src/api/control/mod.rs::stream()` is the server-side mission filter.

dashboard/README.md

@@ -11,13 +11,13 @@ Developer-focused UI for monitoring and controlling the Sandboxed.sh backend.
 cd dashboard
 bun install
 
-# If the backend is on :3000, run the dashboard on :3001 to avoid port conflicts.
-PORT=3001 bun dev
+# The dev server runs on :3001 (backend stays on :3000, so no port conflict).
+bun dev
 ```
 
 Configure the backend URL via:
 - `NEXT_PUBLIC_API_URL` (defaults to `http://127.0.0.1:3000`)
 
 ## Auth
 
-If the backend reports `auth_required=true` from `GET /api/health`, the dashboard will prompt for credentials and store a JWT in `sessionStorage`. In multi-user mode (`auth_mode=multi_user`), it asks for username + password.
+If the backend reports `auth_required=true` from `GET /api/health`, the dashboard will prompt for credentials and store a JWT in `localStorage`. In multi-user mode (`auth_mode=multi_user`), it asks for username + password.

docs/BACKEND_API.md

@@ -91,6 +91,6 @@ vault. Grok accepts `cli_path` to override the CLI binary path.
 ```json
 {
   "ok": true,
-  "message": "Backend configuration updated. Restart Sandboxed.sh to apply runtime changes."
+  "message": "Backend configuration updated."
 }
 ```

docs/DESKTOP_SETUP.md

@@ -109,8 +109,8 @@ DESKTOP_ENABLED=true
 # Xvfb resolution (width x height)
 DESKTOP_RESOLUTION=1920x1080
 
-# Starting display number (will increment for concurrent sessions)
-DESKTOP_DISPLAY_START=99
+# X display to use for desktop automation
+DESKTOP_DISPLAY=:99
 ```
 
 ## Manual Testing

docs/HERMES_ASSISTANT_MIGRATION.md

@@ -12,7 +12,7 @@ assistant stack:
 - `src/api/telegram.rs` owns Telegram webhook routing, trigger filtering,
   per-chat mission creation, file download, streaming edits, Paloma commands,
   proactive mission cards, scheduled messages, workflow relays, and memory.
-- `src/api/control.rs` exposes the Telegram CRUD endpoints and routes webhook
+- `src/api/control/mod.rs` exposes the Telegram CRUD endpoints and routes webhook
   updates into `ControlCommand::UserMessage`.
 - `dashboard/src/app/assistant/page.tsx` is the new top-level UI surface,
   while `dashboard/src/app/settings/telegram/page.tsx` redirects for

docs/MISSION_API.md

@@ -181,7 +181,7 @@ GET /api/control/stream
 Server-Sent Events stream for real-time updates. Events have `event:` and `data:` fields.
 
 **Event types**:
-- `status` — control state changed (`idle`, `running`, `tool_waiting`)
+- `status` — control state changed (`idle`, `running`, `waiting_for_tool`)
 - `user_message` — user message received
 - `assistant_message` — agent response complete
 - `thinking` — agent reasoning (streaming)
@@ -208,7 +208,6 @@ data: {"id":"uuid","content":"Done!","success":true,"cost_cents":5,"model":"clau
 | `/api/control/missions/:id/tree` | GET | Get agent tree for mission |
 | `/api/control/missions/current` | GET | Get current active mission |
 | `/api/control/missions/:id/resume` | POST | Resume interrupted mission |
-| `/api/control/tree` | GET | Get live agent tree |
 | `/api/control/progress` | GET | Get execution progress |
 
 ## Automations
@@ -240,8 +239,8 @@ POST /api/control/missions/:id/automations
 **Body**:
 ```json
 {
-  "command_source": {"library": {"name": "my-command"}},
-  "trigger": {"interval": {"seconds": 300}},
+  "command_source": {"type": "library", "name": "my-command"},
+  "trigger": {"type": "interval", "seconds": 300},
   "variables": {"key": "value"},
   "retry_config": {
     "max_retries": 3,
@@ -252,14 +251,19 @@ POST /api/control/missions/:id/automations
 }
 ```
 
+Both `command_source` and `trigger` are internally tagged: a `"type"`
+discriminator selects the variant and its fields sit alongside it.
+
 **Trigger types**:
-- `{"interval": {"seconds": 300}}` — Run every N seconds
-- `{"webhook": {"config": {"webhook_id": "optional-uuid"}}}` — Trigger via webhook
-- `"agent_finished"` — Trigger after each agent turn completes
+- `{"type": "interval", "seconds": 300}` — Run every N seconds
+- `{"type": "cron", "expression": "0 8 * * *", "timezone": "Europe/Paris"}` — Cron schedule (timezone defaults to UTC)
+- `{"type": "webhook", "config": {"webhook_id": "optional-uuid"}}` — Trigger via webhook
+- `{"type": "agent_finished"}` — Trigger after each agent turn completes
 
 **Command sources**:
-- `{"library": {"name": "command-name"}}` — Use a library command
-- `{"inline": {"command": "echo hello"}}` — Inline shell command
+- `{"type": "library", "name": "command-name"}` — Use a library command
+- `{"type": "local_file", "path": "scripts/run.sh"}` — Command from a file in the mission workspace
+- `{"type": "inline", "content": "echo hello"}` — Inline shell command
 
 **Response**: `Automation` object.
 
@@ -280,8 +284,8 @@ PATCH /api/control/automations/:id
 **Body** (all fields optional):
 ```json
 {
-  "command_source": {"library": {"name": "new-command"}},
-  "trigger": {"interval": {"seconds": 600}},
+  "command_source": {"type": "library", "name": "new-command"},
+  "trigger": {"type": "interval", "seconds": 600},
   "variables": {"key": "new-value"},
   "retry_config": {"max_retries": 5},
   "active": false
@@ -320,8 +324,8 @@ GET /api/control/missions/:id/automation-executions
 {
   "id": "uuid",
   "mission_id": "uuid",
-  "command_source": {"library": {"name": "my-command"}},
-  "trigger": {"interval": {"seconds": 300}},
+  "command_source": {"type": "library", "name": "my-command"},
+  "trigger": {"type": "interval", "seconds": 300},
   "variables": {"key": "value"},
   "active": true,
   "created_at": "2025-01-13T10:00:00Z",
@@ -345,9 +349,9 @@ GET /api/control/missions/:id/automation-executions
   "trigger_source": "interval",
   "status": "success",
   "webhook_payload": null,
-  "started_at": "2025-01-13T10:05:00Z",
+  "variables_used": {"key": "value"},
   "completed_at": "2025-01-13T10:05:05Z",
-  "result_message": "Command completed successfully",
+  "error": null,
   "retry_count": 0
 }
 ```