# Settings compendium Every Settings pane: section titles, control labels, and what they do. Labels match the in-app copy unless noted. **Keychain:** Some panes store secrets only in the macOS Keychain (called out below). ## Settings index (quick reference) {#settings-index} | Pane | Anchor | |------|--------| | [Profile](#profile) | `#profile` | | [You](#you) | `#you` | | [General](#general) | `#general` | | [LLMs](#llms) | `#llms` | | [Motion](#motion) | `#motion` | | [Advanced](#advanced) | `#advanced` | | [Access](#access) | `#access` | | [Network](#network) | `#network` | | [Messaging](#messaging) | `#messaging` | | [Audio](#audio) | `#audio` | | [Heartbeats](#heartbeats) | `#heartbeats` | | [Plugins](#plugins) | `#plugins` | | [Dashboard](#dashboard) | `#dashboard` | | [MCP](#mcp) | `#mcp` | | [Knowledge](#knowledge) | `#knowledge` | | [Kanban](#kanban) | `#kanban` | | [Mesh](#mesh) | `#mesh` | | [Diagnostics](#diagnostics) | `#diagnostics` | | [Software](#software) | `#software` | | [Pending workflow updates (sheet)](#pending-workflow-updates) | (under Advanced) | The **You** pane is the most important surface for **reviewing** what Mugi has learned. **Profile** is the everyday editor for name, tone, and assistant personality. Kanban and Knowledge also have substantial dedicated panes. --- ## Profile {#profile} Edit everyday identity for you and the assistant. Click **Save profile** after changes. ### You - **Name**, **What to call them**, **Pronouns**, **Timezone**, **Preferred tone** — Used in generated text and time reasoning. - **Notes** — Free-form space for anything else you want the assistant to remember about you. ### Assistant - **Name**, **Creature**, **Vibe**, **Emoji** — Core assistant identity fields used in every chat turn. ### Save - **Save profile** — Writes your edits to the running profile (required after changes; not auto-save on every keystroke). Background noticing may later **propose** updates to personal facts — those appear in **Settings → You** for review. --- ## You {#you} Review what Mugi has learned: proposals, connections, checkpoints. Badge on the Settings sidebar when proposals are pending. ### Who you are (summary) Summary of accepted facts. For name, pronouns, timezone, tone, and notes, use **Settings → Profile**. ### Connections & Threads Cross-topic links from conversations, Vault, and Knowledge corpora. ### What Mugi has noticed Recent observations before they become proposals. ### Suggestions from Mugi Proposed facts or updates with rationale, source, before/after diff, and **Accept** / **Reject**. ### History / Checkpoints Timeline of the personal model. **Mark this moment** saves a snapshot you can compare later. --- ## General {#general} ### Appearance - **Theme** — Segmented control: **Light**, **Dark**, **System**. Saves immediately and applies window chrome + app theme. - **Light theme pack** / **Dark theme pack** — Independent pickers for the color palette used in each appearance. Each pack is a small JSON file that supplies colors for Mugi’s status tokens (OK / info / warn / block) across trust, activity, and chat surfaces. Choosing the same pack on both sides is fine — packs ship both halves. - **Reveal user theme packs…** — Opens `~/.mugi/ThemePacks/` in Finder (creating it on first use). Drop a folder named `/mugi-theme.json` (one pack per folder) and click **Reload themes**; the new pack appears in both pickers. Each pack supplies `light` and `dark` color sections; invalid values are rejected in this pane. - **Reload themes** — Rediscovers bundled + user packs without relaunching the app. Use after editing a pack JSON or dropping a new one in the user folder. - **Reset to Default** — Sets both pickers back to the bundled **Default** pack. - **What themes can and cannot change:** theme packs recolor the **semantic status tokens** (used by trust, activity, and chat severity surfaces). They do not replace macOS window chrome, system materials, or SwiftUI semantic text colors — those continue to follow the **Theme** control above. Invalid hex values or pack ids are surfaced inline in this pane and the app falls back to the bundled default. ### Quick prompt panel - **Global shortcut** — Click the field, then press a shortcut. Must include **⌘**, **⌥**, or **⌃**. Requires **Input Monitoring** (and sometimes **Accessibility**) in System Settings for global use while other apps are focused. ### Council mode - **When to convene** — **Off**, **When I ask**, **Suggest on hard questions**, **Always** (Always shows a confirmation because it can run many model passes). - In chat, **`/council `** convenes the council pipeline even when this setting is **Off** ([slash commands](07-chat-search-and-confirmations.md#slash-commands)). - **Daily cap**, **Hourly cap**, **Batch time limit** — Rate limits when council is not Off. - **Append council transcript in chat** — Include seat-by-seat output in the thread when on. ### Grounding / fact-checking - **Verify Mugi's own claims** — **Off**, **Standard** (default), or **Ask first**. - **Standard** — when Mugi states a checkable fact it is not sure of, it may dispatch a background check (local SearXNG) and post **Confirmed**, **Quick correction**, or **Inconclusive** when done — without blocking your turn. - **Ask first** — offers to check before dispatching. - **Off** — no background checks; Mugi hedges in prose only. - **Allow background checks to use the public web** — Off by default; when on, weak local evidence may escalate to the public web (still opt-in per check path). - Mugi only ever checks **its own** claims, never yours. ### Results - **When Mugi delivers a page, chart, or video** — **Switch to Results**, **Switch only while in Chat**, or **Show a badge only**. - **Also open a separate preview window** — legacy floating preview in addition to the Results workspace. - **Notify when a result arrives in the background** — toast when Kanban, scheduled, or research work saves to Results while you are on another conversation. Full walkthrough: [15-results-workspace.md](15-results-workspace.md). ### Chat - **Show agent thinking** — Separate lighter bubbles for model reasoning. - **Show tool details in chat** — When off, tool calls appear in the status line and Activity feed (failures still show in chat). - **Show tabs in a vertical strip** — Optional vertical tab list for many open branches. ### Debug - **Enable debug logging** — Troubleshooting output under `~/.mugi/logs`. Debug logs can include conversation or tool excerpts. ### Setup - **Reset identity & templates…** — Resets identity/templates without wiping all conversations (confirmation required). - **Re-run Initial Setup** — Reinstalls runtime environments (Python, Kraken, Playwright, MLX, utility); keeps conversations and personality files; **Mugi restarts**. - **Full Reset…** — Removes **all** Mugi data; moves personality and conversations to **Trash** for recovery; **Mugi restarts**. Replay the onboarding carousel anytime via **Help → Welcome Tour…**. --- ## LLMs {#llms} The **Engines** tab picks which server powers chat. **Parameters** and **Memory & research** apply regardless of engine (context window reload behavior follows the active engine). ### Inference engine picker | Choice | Meaning | |--------|---------| | **MLX (mlx-vlm)** | Default. Mugi installs/manages mlx-vlm on port **8000**. Model download, LoRA, native tool probe, stack upgrade. | | **Exo (connect)** | Connect-only: you run Exo; Mugi probes `exo_backend_base_url` (default `http://127.0.0.1:52415/v1`). **Connect / Test** checks reachability, loaded model, and native tool calls. Mugi does not install Exo. | | **External API** | OpenAI-compatible remote URL + optional API key (LM Studio, Ollama, vLLM, cloud). | `inference_engine` in `app_prefs.json` is set on migration from older MLX/external toggles. Embeddings and the **utility model** always use the MLX venv. ### MLX panel (when MLX is selected) - Model picker, download, install MLX stack, LoRA/adapters when shown. - **Native tool probe** — confirms `delta.tool_calls` from the loaded model. - Server log: **`~/.mugi/logs/server.log`** (MLX generation server). - **Text-only models** — Mugi disables Automatic Prompt Caching (APC) on launch for text-only checkpoints (APC assumes multimodal models). ### Exo panel (when Exo is selected) - Base URL, model id field, **Connect / Test** status. - Exo must already be running; use Qwen / Hermes-template models for reliable native tool calls. ### External API panel (when External is selected) - Base URL, API key (Keychain-backed where applicable), model id. - Mugi does not manage the remote process. ### Model Parameters - **Context Window** — Slider (large range); shows formatted token count; saves `llm_context_length`. Applies to the active chat engine after you stop dragging. - **Temperature** — Slider 0…2. - **Max Output Tokens** — Slider. - **Apply Context Window** — Reloads the active local model with the current context (disabled while a reload is in progress). ### Research Mode Limits - **Max Search Rounds**, **Max Read URLs**, **Min Scratch Appends**, **Min Scratch Size** — Stepped sliders with live values. - **Strict research phase hints** — Toggle. - **LLM-generated follow-up queries** — Toggle. Foreground **Research mode** spawns a durable Kanban research worker bound to the thread ([Research mode](07-chat-search-and-confirmations.md#research-mode)). ### Agent loop limits (advanced) These keys live in **`~/.mugi/app_prefs.json`**. They are not required for normal use; raise them only when you need longer tool-heavy runs or more guardrail retries. | Key | Default | Role | |-----|---------|------| | `agent_max_tool_rounds` | 35 | Maximum tool-calling rounds per chat run before stop or the max-rounds summary round (clamped 1–500). | | `agent_guard_retries_total` | 8 | Shared budget for harness-injected retries (e.g. task-plan continuation, max-rounds summary; clamped 0–100). | | `task_plan_continuation_retries` | 3 | How many times to nudge when a recorded task plan still has open steps before allowing a final answer (clamped 0–50). | ### Memory & Recall - **Retrieval profile** — **Conservative** / **Balanced**. - **Memory write safety** — **Off — no guardrails** / **Warn — flag risky writes** / **Insights only — require structured sections** / **Block — reject risky writes**. Captions explain password/API noise handling. - **Semantic session search (utility model)** — Toggle; requires utility model when available. --- ## Network {#network} ### SearXNG Search Engine - **Use local SearXNG (Docker)** — When on, **Base URL** shows a fixed `http://127.0.0.1:` (read-only). When off, **Base URL** is editable if you run SearXNG elsewhere yourself. - **Port** — Used with local Docker (defaults restored if invalid). - **Fallback URL** — Default example in UI is `https://searx.party`; used when the primary search endpoint is unavailable. - **Status:** — Shows **running** (green) or other states; **Refresh** updates status. ### Docker Controls - Buttons: **Setup**, **Start**, **Stop**, **Restart**, **Remove** — Each POSTs to `/preferences/searxng/`; errors show in red below. **Setup** / **Remove** may reload preferences on success. --- ## Advanced {#advanced} ### Shell Execution - **Allow shell command execution** — Turning **on** starts an inline flow: caption **Authentication required to enable shell access** with **Confirm** (device owner auth) and **Cancel**. **Confirm** saves `allow_shell_execute: true`; failure or cancel turns the toggle back off. ### CLI Tools - **Enable CLI tools (brew, ffmpeg, run_installed_cli)** — Saves `include_cli_tools`. Caption describes `~/.mugi/homebrew` and `~/.mugi/cli-venvs`, and that shell must be allowed to install new programs. - **Install into Mugi Homebrew** — **Formula** text field, optional **Tap**, **Install Formula** button (disabled if shell disabled or empty formula). Caption: binaries under `~/.mugi/homebrew/bin`. ### Kraken OCR - **Install Kraken OCR environment** / **Remove Kraken OCR environment** — One-time venv under `~/.mugi/kraken-venv`. - **Recognition model path (optional)** — Text field + **Choose…** (`.mlmodel`). Empty = auto default. ### Browser Automation - **Use Playwright for full-page screenshots** — Toggle. - **Browser for screenshots** — Picker (installed Chrome/Edge/Brave vs bundled Chromium). Captions for **Screen Recording** permission. - **Status:** Installed / Not installed; **Install Playwright** or **Reinstall Playwright** — venv `~/.mugi/playwright-venv`. ### Backend environment - **Update backend packages** — Installs/upgrades packages in the **main** backend venv (not Kraken/Playwright venvs). ### Parallel utility model - **Enable embedded utility model** — Toggle; **Model** picker (Qwen, Bonsai, etc.). - **Status** line, **Model:** repo line, error text if load fails. - Buttons: **Download model**, **Upgrade utility MLX**, **Refresh Status**, **Unload from memory** — as enabled by state. Prism/Bonsai caption when applicable. ### Active Hours - Caption: silences heartbeat/idle/curiosity outside windows; **cron** and **inbound webhooks** are **not** gated. - **Schedule** — **Always active** vs **Custom schedule** (segmented). - Custom: **Start** / **End** time pickers; **Days** toggles (weekday buttons); overnight ranges supported; empty weekdays = always active. ### Inbound Webhooks - **Accept inbound webhooks** — Master toggle. Caption: HMAC per source; **secrets in Keychain only**. - **Copy webhook URL** — Disabled when webhooks off. - **Generate primary token** / **Rotate primary token** — Keychain-backed primary token for `X-Mugi-Webhook-Token`. - Per-source UI: add sources, **Notify only** vs **Full turn** modes (see in-app table). New sources get HMAC once; copy into external services. Details: [Inbound webhooks](inbound-webhooks.md). ### Outbound Webhooks - HTTPS targets by **slug**; HMAC in Keychain; cron jobs reference slugs in **Deliver-to**. Caption explains POST body and `X-Mugi-Signature`. ### File Watcher - Enable + list of rules; add rule with **rule name**, **path**, **glob**, **Notify only** / **Full turn**, **Recursive**, **Debounce**, **Add rule**. Stored under `[file_watcher]` in `mugi_config.toml` (caption). ### Personal noticing - **Personal pattern noticing & connections journal** — Master toggle for cross-topic noticing from conversations (and optionally Vault). - **Surface personal pattern notes in On my mind** — Pattern follow-ups in the proactive inbox (requires personal noticing). - **Include Vault notebook in noticing** — Vault excerpts participate in connection proposals (requires personal noticing). - Connection facts → **Settings → You**; pattern follow-ups → **On my mind** (at most once per week per pattern). ### Weekly self-review - **Enable ledger self-review** — Background review of task health (separate from personal noticing); writes insights and may open follow-ups. Off by default. ### Evolution mode (advanced) — disclosure Inside **Evolution mode (advanced)**: - **Evolution mode** — Constitution, git, plugin tools, **background consciousness**. Caption: **restart the background agent** after changing so the consciousness heartbeat picks it up. - **Workflow learning (text evolution)** — Background evaluation of workflows; proposals need approval. - **Pause workflow learning** — Pauses analysis only. - **Pending workflow updates:** count + **Review** → opens [Pending workflow updates](#pending-workflow-updates) sheet. --- ## Audio {#audio} ### Text-to-Speech - **Speak agent replies** — Master toggle. - **TTS engine** — **Apple built-in voices**, **Supertonic** (when installed), or **Chatterbox** (optional — install from **Settings → Software**). - Engine-specific voice pickers, **Preview**, speed/quality controls for Supertonic; Chatterbox voice profiles when installed. ### Voice Input - **Insert text while speaking**, **Push-to-talk mode**, **Prefer on-device recognition** — Dictation behavior; use with **Edit → Voice Input (⌘⇧V)**. - **Enable Full Duplex voice mode (Beta)** — Continuous listen/speak loop with barge-in and partial transcript options when on. - **Enable Voice Conversation pane (Beta)** — Adds **Voice** to the left mode rail (⌃⌘3) with idle timeout and max session duration steppers. --- ## Motion {#motion} - **Motion preference** — **Calm**, **Standard**, or **Vivid** — global animation intensity (respects macOS Reduce Motion). - **Per-surface toggles** — Subagent indicators, Activity feed highlight, inertial chat auto-scroll, terminal streaming animation, voice mic pulse. --- ## Dashboard {#dashboard} - **Global shortcut** — **⇧⌥D** toggles the **Dashboard panel** (**View → Dashboard**). Same Input Monitoring requirements as the Quick Panel. When a shell-approval alert is open, **⇧⌘D** is reserved for **Deny shell command** instead. - **Open Dashboard Plugins Folder** / **Reload Dashboard Plugins** — Dashboard plugins live under `~/.mugi/dashboard_plugins`. - Per-plugin settings when plugins are installed. User guide: [Dashboard panel](08-right-panels.md#dashboard-panel). --- ## Kanban {#kanban} Configuration only — open the board from the **toolbar Kanban button** or **View → Kanban Board**. How dispatcher, workers, and coordinator mode fit together: [08-right-panels.md § Kanban](08-right-panels.md#kanban-overlay) and [Security overview §8](security/harness_threat_model.md#8-kanban-and-task-completion). - **Kanban enabled** — Master kill switch. When off, tasks cannot be claimed and the dispatcher idles. - **Mesh delegation** — Allow agents to tag tasks mesh-eligible; surface inbound mesh delegations on the board. - **Limits & lifecycle** — Tick interval (how often the dispatcher scans the board), claim TTL, deadlines, failure limits, retention, create rate limits. - **Concurrency profiles** — Maximum simultaneous **workers** per assignee profile (general, research, extraction, etc.). A profile cap of `2` means at most two research workers at once. **Not in this pane:** coordinator mode (automatic when a chat thread creates a multi-task batch) and Council mode (**Settings → General**). --- ## Software {#software} Optional runtimes not bundled in the main app: - **Chatterbox TTS** — Install/remove optional voice-cloning runtime; voice profile count; install log under `~/.mugi/logs/`. - **Hyperframes** (when shown) — Optional video tooling with dependency checks and install/remove flows. After installing Chatterbox, choose it under **Settings → Audio → TTS engine**. --- ## Messaging {#messaging} ### Message Gateway - **Enable messaging gateway** — `_config_messaging.enabled`. - **Rate limit:** `0…120` seconds (stepper). ### Discord - **Enable Discord bot** - **Bot token** — `SecureField`. **Keychain:** service `mugi-discord`, account `bot_token` — saved on **Save** / auto-save paths; empty token deletes keychain entry. - **Channel IDs (comma-separated)** - **Command prefix** — Default `!` - **Require @mention** - **Save** (⌘S) — Persists messaging config + token to Keychain. - **Test Connection** — Alert **Discord Test** with success or error string. --- ## Heartbeats {#heartbeats} ### Continuity - **Continuity prompts when idle** — Maps to `idle_in_chat_tick_enabled`; posts `idleTickConfigChanged` so idle behavior updates. Caption: internal prompt after quiet chat; disable to stop **foreground** continuity runs. ### Intervals - **Standard interval:** `1…1440` **min** (stepper) — heartbeat cadence. ### Features - **Memory maintenance** — Toggle + interval stored in heartbeat config. - **Growth check** — Toggle. - **Notification sound** — Toggle. ### Watchlist (collector) - **Enable watchlist monitoring** — When on: **Check interval** `30…1440` min (step 30). - Caption: re-fetch URLs and notify on content change. ### RSS / Atom Feeds - **Enable feeds monitoring** — When on: **Check interval** `30…1440` min (step 30). - List of feeds with label + URL; **+** / **−** bar to add (sheet) or remove selection. - **Add RSS / Atom Feed** sheet: **Feed URL**, **Label (optional)**. - Bottom: **Save Heartbeat Settings** — flushes heartbeat JSON to preferences. --- ## Plugins {#plugins} Top caption: plugins live in **`~/.mugi/plugins`**; use **Reload** after editing files. - **Open Plugins Folder** — Reveals `~/.mugi/plugins` in Finder. - **Reload Plugins** — Rescans installed plugins on disk. ### Installed plugins Each row: **DisclosureGroup** with status dot, **name**, **Enabled** / **Disabled**. Expanded: - **Slug**, **Version**, **Author**, **Load error**, **Load warning(s)**, **Route mount error** (when present). - **Disable** / **Enable** — Toggles a plugin without uninstalling it. - **View in Finder** — `~/.mugi/plugins/`. - **Remove** — Confirmation: disables plugin and **removes approval**; **does not delete files on disk**. ### Plugin settings If the plugin registered a schema: - **Plugin settings** — Dynamic fields from `settings_schema.fields`. - **Save Settings** — `PUT preferences/plugins//settings`; values merged into **`~/.mugi/app_prefs.json`** (caption in UI). --- ## Access {#access} Top banner: **dangerous paths** (system dirs, `.ssh`, `.aws`, `.config`, keychain files, etc.) and **`~/.mugi/plugins`** cannot be allowed. ### Allowed Directories - List of paths with **minus** to remove. - **Add Directory…** — macOS **folder** picker; duplicate path shows an error in-form. ### Calendar - Caption; **Request Calendar Access…** — async permission via `CalendarService`. ### Reminders - Caption; **Request Reminders Access…** — reminders permission. ### Email - Caption: IMAP; **Password is stored in the system Keychain.** - **IMAP server**, **Port**, **Email or username**, **Password** (`SecureField`). - **Save** (⌘S) — **Keychain:** service `mugi-imap`, account `password`. - **Test connection** — Alert with result. --- ## Mesh {#mesh} ### Mugi Mesh - Caption: local discovery and encrypted collaboration. - **Send files via Mesh…** — Opens the same flow as chat **`/mesh-send`**: encrypted LAN file transfer to a connected peer (explicit file picker + consent). Size caps and `file_transfer_enabled` are configured in **Settings → Mesh** or in `mugi_config.toml` under `~/.mugi`. - **Refresh Mesh Status** — Reloads peer/session lists. ### Known Peers - Per peer: **hostname**, **fingerprint**, **Remove**, **Allow task offload** (toggle). Help on toggle: delegated subtasks via `contact_peer(is_task=true)` only if you trust the peer. - Footer: only enable offload for trusted peers. ### Discovered Peers - Optional **connectStatusMessage**; each row: hostname, `address:port`, fingerprint, **Connect**. ### Pending Requests - **Allow** / **Deny** for each request (can also use **macOS notifications** — see [05-notifications.md](05-notifications.md)). ### Active Sessions - Lists `peer_hostname` and `peer_address:peer_port`. --- ## MCP {#mcp} ### Agent - **Enable MCP tools for the agent** — `mcp_enabled`. When **off**, config stays on disk but the assistant does **not** load or call MCP tools; **discovery and health checks still run** (caption). ### MCP Servers - Trust warning; config path **`~/.mugi/mcp_servers.json`**. ### Status - Messages for SDK missing, MCP disabled, invalid JSON. - **SDK installed**, **Config exists**, **Cached tools**, **Health checked** (timestamp). - Per server: health dot, **id**, **transport**, **N tools**, summary, **Copy error** on health errors. - **Refresh Status**, **Check Server Health**, **Open .mugi Folder**. ### Discover Tools - **Discover Now** — May open sheet with tools/errors; first **npx** runs can be slow (caption). ### Config Editor - **Import conflict policy** — Segmented: **Rename** / **Skip** / **Overwrite**. - Monospaced **JSON** editor; **Save** when dirty (see app for exact button labels). - Import from common paths: `~/.cursor/mcp.json`, `~/.claude/mcp.json`, `~/.config/mcp/mcp.json` (when present). - **Add Server** wizard — HTTP vs stdio, preflight, preview (full flow in-app). ### URL schemes - `mugi://mcp/add?...` and `mugi-app://` — Prefills add-server from another app. ### Setup wizard - **Import from Cursor** posts `importCursorMCPConfig` to open MCP settings with import. --- ## Knowledge {#knowledge} Point Mugi at a folder or repo; it digests the material into a searchable corpus (synthesis, per-file notes, topic map, claims graph). Original files are never modified. Compiled output lives under `~/.mugi/workspace/knowledge//`. ### Ingest profiles - **docs** — prose, PDFs, Markdown, Office files (default for document folders). - **repo** — code trees; skips `.git`, `node_modules`, build artifacts, etc. Change profile and re-digest anytime; compiled output is replaced. ### Library vs Settings pane | Surface | Use | |---------|-----| | **Knowledge Library** (books icon on mode rail) | Browse, search, Ask with citations, **Discuss this** into chat | | **Settings → Knowledge** | Add/remove corpora, re-digest, staleness check, export ZIP, status | Interrupted ingest shows **Continue ingest** in the Library. **Pause ingest** checkpoints cleanly. Search and Ask need a complete digest; partial source notes may still be browsable. ### Staleness **Check Staleness** counts changed files (no LLM). **Re-digest** updates only new/changed sources when hashes differ. Optional folder watch re-runs when files change. ### You pane and chat Overlaps between chat, Vault, and corpus claims can appear as proposals in **Settings → You**. **Discuss this** on a source injects that note plus context into the current thread, same pattern as Vault. Workflow: [14-workflows-and-examples.md](14-workflows-and-examples.md) §4. ### Corpus actions - **Open Notes** — reveal compiled folder in Finder - **Re-digest** — refresh (skips unchanged files) - **Check Staleness** — count changes only - **Remove** — deletes compiled tree only; source folder untouched Folders outside `~/.mugi/workspace` need a one-time **Allow digest outside workspace** grant (security-scoped bookmark). Errors show on the corpus row and in the Library banner. --- ## Diagnostics {#diagnostics} ### Memory & Knowledge Inspector - **Memory & Knowledge Inspector toolbar + SSE** — When on, adds a **Memory Inspector** toolbar button and right pane showing per-turn context assembly. Off by default. ### System Diagnostics - **Run Diagnostics** — Refreshes health checks **and** recent background runs. Shortcut **⌘⇧R** when the pane is focused. - **Summary banner** — All checks passed, needs attention, or autonomy warnings; elapsed time shown. - **Autonomy posture** — Disclosure with OK/warning bullets when present. - **Doctor notes** — Warning list from the health report. - **Checks** — Individual pass/fail rows (models, Docker, permissions, etc.). - Empty state before first run: loading indicator, then **Run Diagnostics**. ### Recent background runs - **Refresh** — Updates the task history list only. - **Filter by run kind** — All, Cron, Webhooks, Heartbeats, Idle, Browser. - **Filter by run status** — All, Skipped, Problems. - Skipped runs are often normal (active hours, kill switches). --- ## Pending workflow updates {#pending-workflow-updates} **Entry:** **Settings → Advanced → Evolution mode (advanced) → Workflow learning → Review** (enabled when pending count > 0). **Sheet: Pending Workflow Updates** - **Dismiss All** — Reject all pending workflow proposals. - **Done** — Close sheet. - List of pending workflow improvements; tap a row for before/after diff and accept/reject. --- ## Cross-references - First-line diagnostics: [06-diagnostics-and-basic-troubleshooting.md](06-diagnostics-and-basic-troubleshooting.md) - Webhook tunneling: [Inbound webhooks](inbound-webhooks.md)