fix: explain /bash enablement when disabled (#722) (thanks @vrknetha)

CHANGELOG.md

@@ -1,120 +1,48 @@
 # Changelog
 
-## 2026.1.13-1
+## 2025.1.12 (Unreleased)
-
-### Changes
-- Cron: accept ISO timestamps for one-shot schedules (UTC) and allow optional delete-after-run; wired into CLI + macOS editor.
-- Gateway: add Tailscale binary discovery, custom bind mode, and probe auth retry for password changes. (#740 — thanks @jeffersonwarrior)
-- Agents: add compaction mode config with optional safeguard summarization for long histories. (#700 — thanks @thewilloftheshadow)
-- Agents: support per-agent model fallbacks via `agents.list[].model`. (#583 — thanks @mitschabaude-bot)
-- Tools: add tool profiles plus group shorthands for tool policy allow/deny (global, per-agent, sandbox).
-
-### Fixes
-- Gateway: honor `CLAWDBOT_LAUNCHD_LABEL` / `CLAWDBOT_SYSTEMD_UNIT` overrides when checking or restarting the daemon.
-- Tools: harden tool input schemas for strict providers (no unions; flattened actions; enum helpers + tests).
-- Discord: allow clearing channel parents via message tool without nullable schema inputs.
-
-## 2026.1.12-4
-
-### Changes
-- Models/Moonshot: add Kimi K2 0905 + turbo/thinking variants to the preset + docs. (#818 — thanks @mickahouan)
-- Memory: allow custom OpenAI-compatible embedding endpoints for memory search (remote baseUrl/apiKey/headers). (#819 — thanks @mukhtharcm)
-- Auth: add Chutes OAuth (PKCE + refresh + onboarding choice). (#726 — thanks @FrieSei)
-- Agents: make workspace bootstrap truncation configurable (default 20k) and warn when files are truncated.
-
-### Fixes
-- Typing: keep typing indicators alive during tool execution. (#450, #447 — thanks @thewilloftheshadow)
-- Cron: coerce enabled patches so disabling jobs persists correctly. (#205 — thanks @thewilloftheshadow)
-- Control UI: keep chat scroll position unless user is near the bottom. (#217 — thanks @thewilloftheshadow)
-- Fallback: treat credential validation failures ("no credentials found", "no API key found") as auth errors that trigger model fallback. (#822 — thanks @sebslight)
-- Telegram: preserve forum topic thread ids, including General topic replies. (#727 — thanks @thewilloftheshadow)
-- Telegram: persist polling update offsets across restarts to avoid duplicate updates. (#739 — thanks @thewilloftheshadow)
-- Discord: avoid duplicate message/reaction listeners on monitor reloads. (#744 — thanks @thewilloftheshadow)
-- System events: include local timestamps when events are injected into prompts. (#245 — thanks @thewilloftheshadow)
-- Cron: accept `jobId` aliases for cron update/run/remove params in gateway validation. (#252 — thanks @thewilloftheshadow)
-- Models/Google: normalize Gemini 3 model ids to preview variants before runtime selection. (#795 — thanks @thewilloftheshadow)
-- Models/Google: strip Gemini CLI tool call/response ids in patched provider handling. (#783 — thanks @ananth-vardhan-cn)
-- TUI: keep the last streamed response instead of replacing it with "(no output)". (#747 — thanks @thewilloftheshadow)
-- Slack: accept slash commands with or without leading `/` for custom command configs. (#798 — thanks @thewilloftheshadow)
-- Onboarding/Configure: refuse to proceed with invalid configs; run `clawdbot doctor` first to avoid wiping custom fields. (#764 — thanks @mukhtharcm)
-- Onboarding: quote Windows browser URLs when launching via `cmd start` to preserve OAuth query params. (#794 — thanks @roshanasingh4)
-- Gateway/Auth: allow Tailscale Serve identity headers to satisfy token auth when `allowTailscale` is enabled. (#823 — thanks @roshanasingh4)
-- Anthropic: merge consecutive user turns (preserve newest metadata) before validation to avoid “Incorrect role information” errors. (#804 — thanks @ThomsenDrake)
-- Discord/Slack: centralize reply-thread planning so auto-thread replies stay in the created thread without parent reply refs.
-- Telegram: respect account-scoped bindings when webhook mode is enabled. (#821 — thanks @gumadeiras)
-- Update: run `clawdbot doctor --non-interactive` during updates to avoid TTY hangs. (#781 — thanks @ronyrus)
-- Browser tools: treat explicit `maxChars: 0` as unlimited while keeping the default limit only when omitted. (#796 — thanks @gabriel-trigo)
-- Tools: allow Claude/Gemini tool param aliases (`file_path`, `old_string`, `new_string`) while enforcing required params at runtime. (#793 — thanks @hsrvc)
-- Gemini: downgrade tool-call history missing `thought_signature` to avoid INVALID_ARGUMENT errors. (#793 — thanks @hsrvc)
-- Messaging: enforce context isolation for message tool sends across providers (normalized targets + tests). (#793 — thanks @hsrvc)
-- Auto-reply: re-evaluate reasoning tag enforcement on fallback providers to prevent leaked reasoning. (#810 — thanks @mcinteerj)
-- Tools/Gemini: drop null-only union variants while cleaning tool schemas to avoid Cloud Code Assist schema errors. (#782 — thanks @AbhisekBasu1)
-- Connections UI: polish multi-account account cards in the Connections view. (#816 — thanks @steipete)
-- Gemini: strip Claude `msg_*` thought_signature fields from session history to avoid base64 decode errors. (#805 — thanks @marcmarg)
-
-## 2026.1.12-3
-
-### Changes
-- Sandbox: drop legacy `memory` tool-policy shorthand; require explicit `group:memory`.
-
-### Fixes
-- Telegram: tolerate mocked bots missing native-command APIs (`setMyCommands`, `command`) during tests.
-- Auto-reply: fix streaming block reply media handling (no redeclared/use-before-declare vars).
-
-## 2026.1.12-2
-
-### Changes
-- Subagents: add config to set default sub-agent model (`agents.defaults.subagents.model` + per-agent override); still overridden by `sessions_spawn.model`.
-- Plugins: restore full voice-call plugin parity (Telnyx/Twilio, streaming, inbound policies, tools/CLI).
-- Sandbox: support tool-policy groups in `tools.sandbox.tools` (e.g. `group:memory`, `group:fs`) to reduce config churn.
-
-### Fixes
-- Models/MiniMax: strip malformed tool invocation XML (`<invoke>...</invoke>` and `</minimax:tool_call>`) from assistant text to prevent tool call leaks into user messages. (#809 — thanks @latitudeki5223)
-- Tools/Models: MiniMax vision now uses the Coding Plan VLM endpoint (`/v1/coding_plan/vlm`) so the `image` tool works with MiniMax keys (also accepts `@/path/to/file.png`-style inputs).
-- Gateway/macOS: reduce noisy loopback WS "closed before connect" logs during tests.
-- Auto-reply: resolve ambiguous `/model` fuzzy matches by picking the best candidate instead of erroring.
-
-## 2026.1.12-1
-
-### Changes
-- Heartbeat: raise default `ackMaxChars` to 300 so any `HEARTBEAT_OK` replies with short padding stay internal (fewer noisy heartbeat posts on providers).
-- Onboarding: normalize API key inputs (strip `export KEY=...` wrappers) so shell-style entries paste cleanly.
-
-## 2026.1.12 (Unreleased)
 
 ### Highlights
-- Memory: add vector search for agent memories (Markdown-only scope) with SQLite index, chunking, lazy sync + file watch, and per-agent enablement/fallback.
+- Memory: add vector search for agent memories (Markdown-only) with SQLite index, chunking, lazy sync + file watch, and per-agent enablement/fallback.
+- Plugins: restore full voice-call plugin parity (Telnyx/Twilio, streaming, inbound policies, tools/CLI).
+- Models: add Synthetic provider plus Moonshot Kimi K2 0905 + turbo/thinking variants (with docs).
+- Cron: one-shot schedules accept ISO timestamps (UTC) with optional delete-after-run; cron jobs can target a specific agent (CLI + macOS/Control UI).
+- Agents: add compaction mode config with optional safeguard summarization and per-agent model fallbacks.
 
-### Changes
+### New & Improved
-- Agents: strengthen memory recall guidance (memory_search mandatory for past work/preferences; system prompt injects conditional recall section; memory_get now described as safe snippet fetch).
+- Memory: add custom OpenAI-compatible embedding endpoints; support OpenAI/local `node-llama-cpp` embeddings with per-agent overrides and provider metadata in tools/CLI.
-- Browser: add `scrollintoview` action to scroll refs into view before click/type.
+- Memory: new `clawdbot memory` CLI plus `memory_search`/`memory_get` tools with snippets + line ranges; index stored under `~/.clawdbot/memory/{agentId}.sqlite` with watch-on-by-default.
-- Memory: embedding providers support OpenAI or local `node-llama-cpp`; config adds defaults + per-agent overrides, provider/fallback metadata surfaced in tools/CLI.
+- Agents: strengthen memory recall guidance; make workspace bootstrap truncation configurable (default 20k) with warnings; add default sub-agent model config.
-- CLI/Tools: new `clawdbot memory` commands plus `memory_search`/`memory_get` tools returning snippets + line ranges and provider info.
+- Tools/Sandbox: add tool profiles + group shorthands; support tool-policy groups in `tools.sandbox.tools`; drop legacy `memory` shorthand; allow Docker bind mounts via `docker.binds`.
-- Runtime: memory index stored under `~/.clawdbot/memory/{agentId}.sqlite` with watch-on-by-default; inline status replies now stay auth-gated while inline prompts continue to the agent.
+- Tools: add browser `scrollintoview` action; allow Claude/Gemini tool param aliases; allow thinking `xhigh` for GPT-5.2/Codex with safe downgrades.
-- Doctor: rebuild Control UI assets when protocol schema is newer to avoid stale UI connect errors. (#786) — thanks @meaningfool.
+- Gateway/CLI: add Tailscale binary discovery, custom bind mode, and probe auth retry; add `clawdbot dashboard` auto-open flow; default native slash commands to `"auto"` with per-provider overrides.
-- Cron: allow jobs to target a specific agent and expose agent selection in the macOS app + Control UI.
+- Auth/Onboarding: add Chutes OAuth (PKCE + refresh + onboarding choice); normalize API key inputs; default TUI onboarding to `deliver: false`.
-- Discord: add `discord.allowBots` to permit bot-authored messages (still ignores its own messages) with docs warning about bot loops. (#802) — thanks @zknicker.
+- Providers: add `discord.allowBots`; trim legacy MiniMax M2 from default catalogs; route MiniMax vision to the Coding Plan VLM endpoint (also accepts `@/path/to/file.png` inputs).
-- CLI/Onboarding: `clawdbot dashboard` prints/copies the tokenized Control UI link and opens it; onboarding now auto-opens the dashboard with your token and keeps the link in the summary.
+- Gateway: allow Tailscale Serve identity headers to satisfy token auth; rebuild Control UI assets when protocol schema is newer.
-- Commands: native slash commands now default to `"auto"` (on for Discord/Telegram, off for Slack) with per-provider overrides (`discord/telegram/slack.commands.native`) and docs updated.
+- Heartbeat: default `ackMaxChars` to 300 so short `HEARTBEAT_OK` replies stay internal.
-- Sandbox: allow Docker bind mounts via `docker.binds`; merges global + per-agent binds (per-agent ignored under shared scope) for custom host paths. (#790 — thanks @akonyer)
-- Models: add Synthetic provider (Anthropic-compatible) and trim legacy MiniMax M2 from default catalogs. (#811 — thanks @siraht)
 
 ### Fixes
-- Auto-reply: inline `/status` now honors allowlists (authorized stripped + replied inline; unauthorized leaves text for the agent) to match command gating tests.
+- Models/Providers: treat credential validation failures as auth errors to trigger fallback; normalize `${ENV_VAR}` apiKey values and auto-fill missing provider keys; preserve explicit GitHub Copilot provider config + agent-dir auth profiles.
-- Models: normalize `${ENV_VAR}` apiKey config values and auto-fill missing provider `apiKey` from env/auth when custom provider models are configured (fixes MiniMax “Unknown model” on fresh installs).
+- Gemini: normalize Gemini 3 ids to preview variants; strip Gemini CLI tool call/response ids; downgrade missing `thought_signature`; strip Claude `msg_*` thought_signature fields to avoid base64 decode errors.
-- Models/Tools: include `MiniMax-VL-01` in implicit MiniMax provider so image pairing uses a real vision model.
+- MiniMax: strip malformed tool invocation XML; include `MiniMax-VL-01` in implicit provider for image pairing.
-- Telegram: show typing indicator in General forum topics. (#779) — thanks @azade-c.
+- Anthropic: merge consecutive user turns (preserve newest metadata) before validation to avoid incorrect role errors.
-- Discord: keep reasoning italics intact when messages are chunked, so reasoning stays italic across multi-part sends.
+- Messaging: enforce context isolation for message tool sends; keep typing indicators alive during tool execution.
-- Models: keep explicit GitHub Copilot provider config and honor agent-dir auth profiles for auto-injection. (#705) — thanks @TAGOOZ.
+- Auto-reply: `/status` allowlist behavior, reasoning-tag enforcement on fallback, and system-event enqueueing for elevated/reasoning toggles.
-- Auto-reply: restore 300-char heartbeat ack limit and keep >300 char replies instead of dropping them; adjust long heartbeat test content accordingly.
+- Auto-reply: explain how to enable `/bash` when it’s disabled; add security notes + FAQ. (#722) — thanks @vrknetha.
-- Gateway: `agents.list` now honors explicit `agents.list` config without pulling stray agents from disk; GitHub Copilot CLI auth path uses the updated provider build.
+- Auto-reply: resolve ambiguous `/model` matches; fix streaming block reply media handling; keep >300 char heartbeat replies instead of dropping.
-- Google: apply patched pi-ai `google-gemini-cli` function call handling (strips ids) after upgrading to pi-ai 0.43.0.
+- Discord/Slack: centralize reply-thread planning; fix autoThread routing + add per-channel autoThread; avoid duplicate listeners; keep reasoning italics intact; allow clearing channel parents via message tool.
-- Tools: allow Gemini/Claude-style aliases (`file_path`, `old_string`, `new_string`) in read/write/edit schemas while still enforcing required params at runtime to avoid validation loops.
+- Telegram: preserve forum topic thread ids, persist polling offsets, respect account bindings in webhook mode, and show typing indicator in General topics.
-- Auto-reply: elevated/reasoning toggles now enqueue system events so the model sees the mode change immediately.
+- Slack: accept slash commands with or without leading `/` for custom command configs.
-- Tools: keep `image` available in sandbox and fail over when image models return empty output (fixes “(no text returned)”).
+- Cron: persist disabled jobs correctly; accept `jobId` aliases for update/run/remove params.
-- Discord: add per-channel `autoThread` to auto-create threads for top-level messages. (#800) — thanks @davidguttman.
+- Gateway/CLI: honor `CLAWDBOT_LAUNCHD_LABEL` / `CLAWDBOT_SYSTEMD_UNIT` overrides; `agents.list` respects explicit config; reduce noisy loopback WS logs during tests; run `clawdbot doctor --non-interactive` during updates.
-- Discord: fix autoThread routing so replies stay in the created thread and avoid reply references. (#807) — thanks @davidguttman.
+- Onboarding/Control UI: refuse invalid configs (run doctor first); quote Windows browser URLs for OAuth; keep chat scroll position unless the user is near the bottom.
-- Onboarding: TUI defaults to `deliver: false` to avoid cross-provider auto-delivery leaks; onboarding spawns the TUI with explicit `deliver: false`. (#791 — thanks @roshanasingh4)
+- Tools/UI: harden tool input schemas for strict providers; drop null-only union variants for Gemini schema cleanup; treat `maxChars: 0` as unlimited; keep TUI last streamed response instead of "(no output)".
+- Connections UI: polish multi-account account cards.
+
+### Maintenance
+- Dependencies: bump Pi packages to 0.45.3 and refresh patched pi-ai.
+- Testing: update Vitest + browser-playwright to 4.0.17.
+- Docs: add Amazon Bedrock provider notes and link from models/FAQ.
 
 ## 2026.1.11
 

README.md

@@ -247,7 +247,7 @@ Send these in WhatsApp/Telegram/Slack/Microsoft Teams/WebChat (group commands ar
 - `/status` — compact session status (model + tokens, cost when available)
 - `/new` or `/reset` — reset the session
 - `/compact` — compact session context (summary)
-- `/think <level>` — off|minimal|low|medium|high
+- `/think <level>` — off|minimal|low|medium|high|xhigh (GPT-5.2 + Codex models only)
 - `/verbose on|off`
 - `/cost on|off` — append per-response token/cost usage lines
 - `/restart` — restart the gateway (owner-only in groups)

docs/automation/cron-jobs.md

@@ -105,7 +105,7 @@ Isolation options (only for `session=isolated`):
 ### Model and thinking overrides
 Isolated jobs (`agentTurn`) can override the model and thinking level:
 - `model`: Provider/model string (e.g., `anthropic/claude-sonnet-4-20250514`) or alias (e.g., `opus`)
-- `thinking`: Thinking level (`off`, `minimal`, `low`, `medium`, `high`)
+- `thinking`: Thinking level (`off`, `minimal`, `low`, `medium`, `high`, `xhigh`; GPT-5.2 + Codex models only)
 
 Note: You can set `model` on main-session jobs too, but it changes the shared main
 session model. We recommend model overrides only for isolated jobs to avoid

docs/bedrock.md

@@ -0,0 +1,71 @@
+---
+summary: "Use Amazon Bedrock (Converse API) models with Clawdbot"
+read_when:
+  - You want to use Amazon Bedrock models with Clawdbot
+  - You need AWS credential/region setup for model calls
+---
+# Amazon Bedrock
+
+Clawdbot can use **Amazon Bedrock** models via pi‑ai’s **Bedrock Converse**
+streaming provider. Bedrock auth uses the **AWS SDK default credential chain**,
+not an API key.
+
+## What pi‑ai supports
+
+- Provider: `amazon-bedrock`
+- API: `bedrock-converse-stream`
+- Auth: AWS credentials (env vars, shared config, or instance role)
+- Region: `AWS_REGION` or `AWS_DEFAULT_REGION` (default: `us-east-1`)
+
+## Setup (manual)
+
+1) Ensure AWS credentials are available on the **gateway host**:
+
+```bash
+export AWS_ACCESS_KEY_ID="AKIA..."
+export AWS_SECRET_ACCESS_KEY="..."
+export AWS_REGION="us-east-1"
+# Optional:
+export AWS_SESSION_TOKEN="..."
+export AWS_PROFILE="your-profile"
+```
+
+2) Add a Bedrock provider and model to your config:
+
+```json5
+{
+  models: {
+    providers: {
+      "amazon-bedrock": {
+        baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
+        api: "bedrock-converse-stream",
+        models: [
+          {
+            id: "anthropic.claude-3-7-sonnet-20250219-v1:0",
+            name: "Claude 3.7 Sonnet (Bedrock)",
+            reasoning: true,
+            input: ["text"],
+            cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+            contextWindow: 200000,
+            maxTokens: 8192
+          }
+        ]
+      }
+    }
+  },
+  agents: {
+    defaults: {
+      model: { primary: "amazon-bedrock/anthropic.claude-3-7-sonnet-20250219-v1:0" }
+    }
+  }
+}
+```
+
+## Notes
+
+- Bedrock requires **model access** enabled in your AWS account/region.
+- If you use profiles, set `AWS_PROFILE` on the gateway host.
+- Reasoning support depends on the model; check the Bedrock model card for
+  current capabilities.
+- If you prefer a managed key flow, you can also place an OpenAI‑compatible
+  proxy in front of Bedrock and configure it as an OpenAI provider instead.

docs/cli/index.md

@@ -398,7 +398,7 @@ Required:
 Options:
 - `--to <dest>` (for session key and optional delivery)
 - `--session-id <id>`
-- `--thinking <off|minimal|low|medium|high>`
+- `--thinking <off|minimal|low|medium|high|xhigh>` (GPT-5.2 + Codex models only)
 - `--verbose <on|off>`
 - `--provider <whatsapp|telegram|discord|slack|signal|imessage>`
 - `--local`

docs/experiments/proposals/hooks-gmail-model.md

@@ -39,7 +39,7 @@ Add `hooks.gmail.model` config option to specify an optional cheaper model for G
 | Field | Type | Default | Description |
 |-------|------|---------|-------------|
 | `hooks.gmail.model` | `string` | (none) | Model to use for Gmail hook processing. Accepts `provider/model` refs or aliases from `agents.defaults.models`. |
-| `hooks.gmail.thinking` | `string` | (inherited) | Thinking level override (`off`, `minimal`, `low`, `medium`, `high`). If unset, inherits from `agents.defaults.thinkingDefault` or model's default. |
+| `hooks.gmail.thinking` | `string` | (inherited) | Thinking level override (`off`, `minimal`, `low`, `medium`, `high`, `xhigh`; GPT-5.2 + Codex models only). If unset, inherits from `agents.defaults.thinkingDefault` or model's default. |
 
 ### Alias Support
 

docs/gateway/configuration.md

@@ -746,8 +746,10 @@ Controls how chat commands are enabled across connectors.
 ```json5
 {
   commands: {
-    native: false,          // register native commands when supported
+    native: "auto",         // register native commands when supported (auto)
     text: true,             // parse slash commands in chat messages
+    bash: false,            // allow ! (alias: /bash) (host-only; requires tools.elevated allowlists)
+    bashForegroundMs: 2000, // bash foreground window (0 backgrounds immediately)
     config: false,          // allow /config (writes to disk)
     debug: false,           // allow /debug (runtime-only overrides)
     restart: false,         // allow /restart + gateway restart tool
@@ -761,6 +763,8 @@ Notes:
 - `commands.text: false` disables parsing chat messages for commands.
 - `commands.native: "auto"` (default) turns on native commands for Discord/Telegram and leaves Slack off; unsupported providers stay text-only.
 - Set `commands.native: true|false` to force all, or override per provider with `discord.commands.native`, `telegram.commands.native`, `slack.commands.native` (bool or `"auto"`). `false` clears previously registered commands on Discord/Telegram at startup; Slack commands are managed in the Slack app.
+- `commands.bash: true` enables `! <cmd>` to run host shell commands (`/bash <cmd>` also works as an alias). Requires `tools.elevated.enabled` and allowlisting the sender in `tools.elevated.allowFrom.<provider>`.
+- `commands.bashForegroundMs` controls how long bash waits before backgrounding. While a bash job is running, new `! <cmd>` requests are rejected (one at a time).
 - `commands.config: true` enables `/config` (reads/writes `clawdbot.json`).
 - `commands.debug: true` enables `/debug` (runtime-only overrides).
 - `commands.restart: true` enables `/restart` and the gateway tool restart action.

docs/providers/models.md

@@ -31,6 +31,7 @@ model as `provider/model`.
 - [Z.AI](/providers/zai)
 - [GLM models](/providers/glm)
 - [MiniMax](/providers/minimax)
+- [Amazon Bedrock](/bedrock)
 
 For the full provider catalog (xAI, Groq, Mistral, etc.) and advanced configuration,
 see [Model providers](/concepts/model-providers).

docs/start/faq.md

@@ -5,6 +5,109 @@ summary: "Frequently asked questions about Clawdbot setup, configuration, and us
 
 Quick answers plus deeper troubleshooting for real-world setups (local dev, VPS, multi-agent, OAuth/API keys, model failover). For runtime diagnostics, see [Troubleshooting](/gateway/troubleshooting). For the full config reference, see [Configuration](/gateway/configuration).
 
+## Table of contents
+
+- [What is Clawdbot?](#what-is-clawdbot)
+  - [What is Clawdbot, in one paragraph?](#what-is-clawdbot-in-one-paragraph)
+- [Quick start and first-run setup](#quick-start-and-first-run-setup)
+  - [What’s the recommended way to install and set up Clawdbot?](#whats-the-recommended-way-to-install-and-set-up-clawdbot)
+  - [How do I open the dashboard after onboarding?](#how-do-i-open-the-dashboard-after-onboarding)
+  - [How do I authenticate the dashboard (token) on localhost vs remote?](#how-do-i-authenticate-the-dashboard-token-on-localhost-vs-remote)
+  - [What runtime do I need?](#what-runtime-do-i-need)
+  - [What does the onboarding wizard actually do?](#what-does-the-onboarding-wizard-actually-do)
+  - [How does Anthropic "setup-token" auth work?](#how-does-anthropic-setup-token-auth-work)
+  - [Do you support Claude subscription auth (Claude Code OAuth)?](#do-you-support-claude-subscription-auth-claude-code-oauth)
+  - [Is AWS Bedrock supported?](#is-aws-bedrock-supported)
+  - [How does Codex auth work?](#how-does-codex-auth-work)
+  - [Is a local model OK for casual chats?](#is-a-local-model-ok-for-casual-chats)
+  - [How do I keep hosted model traffic in a specific region?](#how-do-i-keep-hosted-model-traffic-in-a-specific-region)
+  - [Can I use Bun?](#can-i-use-bun)
+  - [Telegram: what goes in `allowFrom`?](#telegram-what-goes-in-allowfrom)
+  - [Can multiple people use one WhatsApp number with different Clawdbots?](#can-multiple-people-use-one-whatsapp-number-with-different-clawdbots)
+  - [Can I run a "fast chat" agent and an "Opus for coding" agent?](#can-i-run-a-fast-chat-agent-and-an-opus-for-coding-agent)
+  - [Does Homebrew work on Linux?](#does-homebrew-work-on-linux)
+  - [Can I switch between npm and git installs later?](#can-i-switch-between-npm-and-git-installs-later)
+- [Skills and automation](#skills-and-automation)
+  - [How do I customize skills without keeping the repo dirty?](#how-do-i-customize-skills-without-keeping-the-repo-dirty)
+  - [Can I load skills from a custom folder?](#can-i-load-skills-from-a-custom-folder)
+  - [How can I use different models for different tasks?](#how-can-i-use-different-models-for-different-tasks)
+  - [How do I install skills on Linux?](#how-do-i-install-skills-on-linux)
+- [Sandboxing and memory](#sandboxing-and-memory)
+  - [Is there a dedicated sandboxing doc?](#is-there-a-dedicated-sandboxing-doc)
+  - [How do I bind a host folder into the sandbox?](#how-do-i-bind-a-host-folder-into-the-sandbox)
+  - [How does memory work?](#how-does-memory-work)
+  - [Does semantic memory search require an OpenAI API key?](#does-semantic-memory-search-require-an-openai-api-key)
+- [Where things live on disk](#where-things-live-on-disk)
+  - [Where does Clawdbot store its data?](#where-does-clawdbot-store-its-data)
+  - [How do I completely uninstall Clawdbot?](#how-do-i-completely-uninstall-clawdbot)
+  - [Can agents work outside the workspace?](#can-agents-work-outside-the-workspace)
+  - [I’m in remote mode — where is the session store?](#im-in-remote-mode-where-is-the-session-store)
+- [Config basics](#config-basics)
+  - [What format is the config? Where is it?](#what-format-is-the-config-where-is-it)
+  - [I set `gateway.bind: "lan"` (or `"tailnet"`) and now nothing listens / the UI says unauthorized](#i-set-gatewaybind-lan-or-tailnet-and-now-nothing-listens-the-ui-says-unauthorized)
+  - [Why do I need a token on localhost now?](#why-do-i-need-a-token-on-localhost-now)
+  - [Do I have to restart after changing config?](#do-i-have-to-restart-after-changing-config)
+  - [How do I run a central Gateway with specialized workers across devices?](#how-do-i-run-a-central-gateway-with-specialized-workers-across-devices)
+  - [Can the Clawdbot browser run headless?](#can-the-clawdbot-browser-run-headless)
+- [Remote gateways + nodes](#remote-gateways-nodes)
+  - [How do commands propagate between Telegram, the gateway, and nodes?](#how-do-commands-propagate-between-telegram-the-gateway-and-nodes)
+  - [Do nodes run a gateway daemon?](#do-nodes-run-a-gateway-daemon)
+  - [Is there an API / RPC way to apply config?](#is-there-an-api-rpc-way-to-apply-config)
+  - [What’s a minimal “sane” config for a first install?](#whats-a-minimal-sane-config-for-a-first-install)
+- [Env vars and .env loading](#env-vars-and-env-loading)
+  - [How does Clawdbot load environment variables?](#how-does-clawdbot-load-environment-variables)
+  - [“I started the Gateway via a daemon and my env vars disappeared.” What now?](#i-started-the-gateway-via-a-daemon-and-my-env-vars-disappeared-what-now)
+- [Sessions & multiple chats](#sessions-multiple-chats)
+  - [How do I start a fresh conversation?](#how-do-i-start-a-fresh-conversation)
+  - [How do I completely reset Clawdbot (but keep it installed)?](#how-do-i-completely-reset-clawdbot-but-keep-it-installed)
+  - [Do I need to add a “bot account” to a WhatsApp group?](#do-i-need-to-add-a-bot-account-to-a-whatsapp-group)
+  - [Why doesn’t Clawdbot reply in a group?](#why-doesnt-clawdbot-reply-in-a-group)
+  - [Do groups/threads share context with DMs?](#do-groupsthreads-share-context-with-dms)
+- [Models: defaults, selection, aliases, switching](#models-defaults-selection-aliases-switching)
+  - [What is the “default model”?](#what-is-the-default-model)
+  - [How do I switch models on the fly (without restarting)?](#how-do-i-switch-models-on-the-fly-without-restarting)
+  - [Why do I see “Model … is not allowed” and then no reply?](#why-do-i-see-model-is-not-allowed-and-then-no-reply)
+  - [Are opus / sonnet / gpt built‑in shortcuts?](#are-opus-sonnet-gpt-builtin-shortcuts)
+  - [How do I define/override model shortcuts (aliases)?](#how-do-i-defineoverride-model-shortcuts-aliases)
+  - [How do I add models from other providers like OpenRouter or Z.AI?](#how-do-i-add-models-from-other-providers-like-openrouter-or-zai)
+- [Model failover and “All models failed”](#model-failover-and-all-models-failed)
+  - [How does failover work?](#how-does-failover-work)
+  - [What does this error mean?](#what-does-this-error-mean)
+  - [Fix checklist for `No credentials found for profile "anthropic:default"`](#fix-checklist-for-no-credentials-found-for-profile-anthropicdefault)
+  - [Why did it also try Google Gemini and fail?](#why-did-it-also-try-google-gemini-and-fail)
+- [Auth profiles: what they are and how to manage them](#auth-profiles-what-they-are-and-how-to-manage-them)
+  - [What is an auth profile?](#what-is-an-auth-profile)
+  - [What are typical profile IDs?](#what-are-typical-profile-ids)
+  - [Can I control which auth profile is tried first?](#can-i-control-which-auth-profile-is-tried-first)
+  - [OAuth vs API key: what’s the difference?](#oauth-vs-api-key-whats-the-difference)
+- [Gateway: ports, “already running”, and remote mode](#gateway-ports-already-running-and-remote-mode)
+  - [What port does the Gateway use?](#what-port-does-the-gateway-use)
+  - [Why does `clawdbot daemon status` say `Runtime: running` but `RPC probe: failed`?](#why-does-clawdbot-daemon-status-say-runtime-running-but-rpc-probe-failed)
+  - [Why does `clawdbot daemon status` show `Config (cli)` and `Config (daemon)` different?](#why-does-clawdbot-daemon-status-show-config-cli-and-config-daemon-different)
+  - [What does “another gateway instance is already listening” mean?](#what-does-another-gateway-instance-is-already-listening-mean)
+  - [How do I run Clawdbot in remote mode (client connects to a Gateway elsewhere)?](#how-do-i-run-clawdbot-in-remote-mode-client-connects-to-a-gateway-elsewhere)
+  - [The Control UI says “unauthorized” (or keeps reconnecting). What now?](#the-control-ui-says-unauthorized-or-keeps-reconnecting-what-now)
+  - [I set `gateway.bind: "tailnet"` but it can’t bind / nothing listens](#i-set-gatewaybind-tailnet-but-it-cant-bind-nothing-listens)
+  - [Can I run multiple Gateways on the same host?](#can-i-run-multiple-gateways-on-the-same-host)
+- [Logging and debugging](#logging-and-debugging)
+  - [Where are logs?](#where-are-logs)
+  - [How do I start/stop/restart the Gateway daemon?](#how-do-i-startstoprestart-the-gateway-daemon)
+  - [What’s the fastest way to get more details when something fails?](#whats-the-fastest-way-to-get-more-details-when-something-fails)
+- [Media & attachments](#media-attachments)
+  - [My skill generated an image/PDF, but nothing was sent](#my-skill-generated-an-imagepdf-but-nothing-was-sent)
+- [Security and access control](#security-and-access-control)
+  - [Is it safe to expose Clawdbot to inbound DMs?](#is-it-safe-to-expose-clawdbot-to-inbound-dms)
+  - [WhatsApp: will it message my contacts? How does pairing work?](#whatsapp-will-it-message-my-contacts-how-does-pairing-work)
+- [Chat commands, aborting tasks, and “it won’t stop”](#chat-commands-aborting-tasks-and-it-wont-stop)
+- [How do I stop/cancel a running task?](#how-do-i-stopcancel-a-running-task)
+- [Why does `/bash` say it’s disabled?](#why-does-bash-say-its-disabled)
+- [Why does it feel like the bot “ignores” rapid‑fire messages?](#why-does-it-feel-like-the-bot-ignores-rapidfire-messages)
+- [Common troubleshooting](#common-troubleshooting)
+  - [“All models failed” — what should I check first?](#all-models-failed-what-should-i-check-first)
+  - [I’m running on my personal WhatsApp number — why is self-chat weird?](#im-running-on-my-personal-whatsapp-number-why-is-self-chat-weird)
+  - [WhatsApp logged me out. How do I re‑auth?](#whatsapp-logged-me-out-how-do-i-reauth)
+  - [Build errors on `main` — what’s the standard fix path?](#build-errors-on-main-whats-the-standard-fix-path)
+
 ## First 60 seconds if something's broken
 
 1) **Quick status (first check)**
@@ -60,7 +163,7 @@ Quick answers plus deeper troubleshooting for real-world setups (local dev, VPS,
 
 Clawdbot is a personal AI assistant you run on your own devices. It replies on the messaging surfaces you already use (WhatsApp, Telegram, Slack, Discord, Signal, iMessage, WebChat) and can also do voice + a live Canvas on supported platforms. The **Gateway** is the always‑on control plane; the assistant is the product.
 
-## Quick start and first‑run setup
+## Quick start and first-run setup
 
 ### What’s the recommended way to install and set up Clawdbot?
 
@@ -128,7 +231,7 @@ Yes. Clawdbot can **reuse Claude Code CLI credentials** (OAuth) and also support
 
 ### Is AWS Bedrock supported?
 
-Not currently. Clawdbot doesn’t ship a Bedrock provider today. If you must use Bedrock, the common workaround is an OpenAI‑compatible proxy in front of Bedrock, then point Clawdbot at that endpoint. See [Model providers](/providers/models) and [Model providers (full list)](/concepts/model-providers).
+Yes — via pi‑ai’s **Amazon Bedrock (Converse)** provider with **manual config**. You must supply AWS credentials/region on the gateway host and add a Bedrock provider entry in your models config. See [Amazon Bedrock](/bedrock) and [Model providers](/providers/models). If you prefer a managed key flow, an OpenAI‑compatible proxy in front of Bedrock is still a valid option.
 
 ### How does Codex auth work?
 
@@ -196,6 +299,8 @@ clawdbot daemon restart
 
 Doctor detects a gateway service entrypoint mismatch and offers to rewrite the service config to match the current install (use `--repair` in automation).
 
+## Skills and automation
+
 ### How do I customize skills without keeping the repo dirty?
 
 Use managed overrides instead of editing the repo copy. Put your changes in `~/.clawdbot/skills/<name>/SKILL.md` (or add a folder via `skills.load.extraDirs` in `~/.clawdbot/clawdbot.json`). Precedence is `<workspace>/skills` > `~/.clawdbot/skills` > bundled, so managed overrides win without touching git. Only upstream-worthy edits should live in the repo and go out as PRs.
@@ -240,6 +345,8 @@ clawdhub update --all
 
 ClawdHub installs into `./skills` under your current directory (or falls back to your configured Clawdbot workspace); Clawdbot treats that as `<workspace>/skills` on the next session. For shared skills across agents, place them in `~/.clawdbot/skills/<name>/SKILL.md`. Some skills expect binaries installed via Homebrew; on Linux that means Linuxbrew (see the Homebrew Linux FAQ entry above). See [Skills](/tools/skills) and [ClawdHub](/tools/clawdhub).
 
+## Sandboxing and memory
+
 ### Is there a dedicated sandboxing doc?
 
 Yes. See [Sandboxing](/gateway/sandboxing). For Docker-specific setup (full gateway in Docker or sandbox images), see [Docker](/install/docker).
@@ -957,6 +1064,26 @@ Slash commands overview: see [Slash commands](/tools/slash-commands).
 
 Most commands must be sent as a **standalone** message that starts with `/`, but a few shortcuts (like `/status`) also work inline for allowlisted senders.
 
+### Why does `/bash` say it’s disabled?
+
+The host shell command (`! <cmd>` or `/bash <cmd>`) is **disabled by default** because it runs directly on the gateway host. To enable it, update your config and restart the Gateway:
+
+```json5
+{
+  commands: { bash: true },
+  tools: {
+    elevated: {
+      enabled: true,
+      allowFrom: {
+        "<provider>": ["<sender-id>"]
+      }
+    }
+  }
+}
+```
+
+Keep the allowlist tight and avoid enabling it in public rooms. See [Slash commands](/tools/slash-commands) and [Security](/gateway/security).
+
 ### Why does it feel like the bot “ignores” rapid‑fire messages?
 
 Queue mode controls how new messages interact with an in‑flight run. Use `/queue` to change modes:

docs/tools/agent-send.md

@@ -38,7 +38,7 @@ clawdbot agent --to +15555550123 --message "Summon reply" --deliver
 - `--local`: run locally (requires provider keys in your shell)
 - `--deliver`: send the reply to the chosen provider (requires `--to`)
 - `--provider`: `whatsapp|telegram|discord|slack|signal|imessage` (default: `whatsapp`)
-- `--thinking <off|minimal|low|medium|high>`: persist thinking level
+- `--thinking <off|minimal|low|medium|high|xhigh>`: persist thinking level (GPT-5.2 + Codex models only)
 - `--verbose <on|off>`: persist verbose level
 - `--timeout <seconds>`: override agent timeout
 - `--json`: output structured JSON

docs/tools/elevated.md

@@ -7,10 +7,13 @@ read_when:
 
 ## What it does
 - Elevated mode allows the exec tool to run with elevated privileges when the feature is available and the sender is approved.
+- The bash chat command (`!`; `/bash` alias) uses the same `tools.elevated` allowlists because it always runs on the host.
 - **Optional for sandboxed agents**: elevated only changes behavior when the agent is running in a sandbox. If the agent already runs unsandboxed, elevated is effectively a no-op.
 - Directive forms: `/elevated on`, `/elevated off`, `/elev on`, `/elev off`.
 - Only `on|off` are accepted; anything else returns a hint and does not change state.
 
+Security note: enabling elevated access or `/bash` effectively grants host shell access. Keep allowlists tight and avoid enabling it in public rooms.
+
 ## What it controls (and what it doesn’t)
 - **Availability gates**: `tools.elevated` is the global baseline. `agents.list[].tools.elevated` can further restrict elevated per agent (both must allow).
 - **Per-session state**: `/elevated on|off` sets the elevated level for the current session key.

docs/tools/slash-commands.md

@@ -7,6 +7,7 @@ read_when:
 # Slash commands
 
 Commands are handled by the Gateway. Most commands must be sent as a **standalone** message that starts with `/`.
+The host-only bash chat command uses `! <cmd>` (with `/bash <cmd>` as an alias).
 
 There are two related systems:
 
@@ -26,6 +27,8 @@ They run immediately, are stripped before the model sees the message, and the re
   commands: {
     native: "auto",
     text: true,
+    bash: false,
+    bashForegroundMs: 2000,
     config: false,
     debug: false,
     restart: false,
@@ -40,6 +43,9 @@ They run immediately, are stripped before the model sees the message, and the re
   - Auto: on for Discord/Telegram; off for Slack (until you add slash commands); ignored for providers without native support.
   - Set `discord.commands.native`, `telegram.commands.native`, or `slack.commands.native` to override per provider (bool or `"auto"`).
   - `false` clears previously registered commands on Discord/Telegram at startup. Slack commands are managed in the Slack app and are not removed automatically.
+- `commands.bash` (default `false`) enables `! <cmd>` to run host shell commands (`/bash <cmd>` is an alias; requires `tools.elevated` allowlists).
+- When `commands.bash` is `false`, `/bash` replies with a short enablement hint (config keys + allowlist).
+- `commands.bashForegroundMs` (default `2000`) controls how long bash waits before switching to background mode (`0` backgrounds immediately).
 - `commands.config` (default `false`) enables `/config` (reads/writes `clawdbot.json`).
 - `commands.debug` (default `false`) enables `/debug` (runtime-only overrides).
 - `commands.useAccessGroups` (default `true`) enforces allowlists/policies for commands.
@@ -60,15 +66,19 @@ Text + native (when enabled):
 - `/activation mention|always` (groups only)
 - `/send on|off|inherit` (owner-only)
 - `/reset` or `/new`
-- `/think <level>` (aliases: `/thinking`, `/t`)
+- `/think <off|minimal|low|medium|high|xhigh>` (GPT-5.2 + Codex models only; aliases: `/thinking`, `/t`)
 - `/verbose on|off` (alias: `/v`)
 - `/reasoning on|off|stream` (alias: `/reason`; when on, sends a separate message prefixed `Reasoning:`; `stream` = Telegram draft only)
 - `/elevated on|off` (alias: `/elev`)
 - `/model <name>` (alias: `/models`; or `/<alias>` from `agents.defaults.models.*.alias`)
 - `/queue <mode>` (plus options like `debounce:2s cap:25 drop:summarize`; send `/queue` to see current settings)
+- `/bash <command>` (host-only; alias for `! <command>`; requires `commands.bash: true` + `tools.elevated` allowlists)
 
 Text-only:
 - `/compact [instructions]` (see [/concepts/compaction](/concepts/compaction))
+- `! <command>` (host-only; one at a time; use `!poll` + `!stop` for long-running jobs)
+- `!poll` (check output / status; accepts optional `sessionId`; `/bash poll` also works)
+- `!stop` (stop the running bash job; accepts optional `sessionId`; `/bash stop` also works)
 
 Notes:
 - Commands accept an optional `:` between the command and args (e.g. `/think: high`, `/send: on`, `/help:`).
@@ -81,6 +91,10 @@ Notes:
 - **Inline shortcuts (allowlisted senders only):** `/help`, `/commands`, `/status` (`/usage`), `/whoami` (`/id`) also work when embedded in text.
 - Unauthorized command-only messages are silently ignored, and inline `/...` tokens are treated as plain text.
 
+## Security note: `!` / `/bash`
+
+`! <cmd>` and `/bash <cmd>` run **directly on the gateway host** as the gateway user. Keep this disabled unless you fully trust the sender and have locked down allowlists. Treat it like giving someone SSH access to the host.
+
 ## Usage vs cost (what shows where)
 
 - **Provider usage/quota** (example: “Claude 80% left”) shows up in `/status` when provider usage tracking is enabled.

docs/tools/thinking.md

@@ -7,11 +7,12 @@ read_when:
 
 ## What it does
 - Inline directive in any inbound body: `/t <level>`, `/think:<level>`, or `/thinking <level>`.
-- Levels (aliases): `off | minimal | low | medium | high`
+- Levels (aliases): `off | minimal | low | medium | high | xhigh` (GPT-5.2 + Codex models only)
   - minimal → “think”
   - low → “think hard”
   - medium → “think harder”
   - high → “ultrathink” (max budget)
+  - xhigh → “ultrathink+” (GPT-5.2 + Codex models only)
   - `highest`, `max` map to `high`.
 
 ## Resolution order

docs/web/tui.md

@@ -50,7 +50,7 @@ Use SSH tunneling or Tailscale to reach the Gateway WS.
 - `/agent <id>` (or `/agents`)
 - `/session <key>` (or `/sessions`)
 - `/model <provider/model>` (or `/model list`, `/models`)
-- `/think <off|minimal|low|medium|high>`
+- `/think <off|minimal|low|medium|high|xhigh>` (GPT-5.2 + Codex models only)
 - `/verbose <on|off>`
 - `/reasoning <on|off|stream>` (stream = Telegram draft only)
 - `/cost <on|off>`

package.json

@@ -132,10 +132,10 @@
     "@grammyjs/runner": "^2.0.3",
     "@grammyjs/transformer-throttler": "^1.2.1",
     "@homebridge/ciao": "^1.3.4",
-    "@mariozechner/pi-agent-core": "^0.43.0",
+    "@mariozechner/pi-agent-core": "^0.45.3",
-    "@mariozechner/pi-ai": "^0.43.0",
+    "@mariozechner/pi-ai": "^0.45.3",
-    "@mariozechner/pi-coding-agent": "^0.43.0",
+    "@mariozechner/pi-coding-agent": "^0.45.3",
-    "@mariozechner/pi-tui": "^0.43.0",
+    "@mariozechner/pi-tui": "^0.45.3",
     "@microsoft/agents-hosting": "^1.1.1",
     "@microsoft/agents-hosting-express": "^1.1.1",
     "@microsoft/agents-hosting-extensions-teams": "^1.1.1",
@@ -207,7 +207,7 @@
       "@sinclair/typebox": "0.34.47"
     },
     "patchedDependencies": {
-      "@mariozechner/pi-ai@0.43.0": "patches/@mariozechner__pi-ai@0.43.0.patch"
+      "@mariozechner/pi-ai@0.45.3": "patches/@mariozechner__pi-ai@0.45.3.patch"
     }
   },
   "vitest": {

patches/@mariozechner__pi-ai@0.45.3.patch

@@ -1,8 +1,8 @@
 diff --git a/dist/providers/google-gemini-cli.js b/dist/providers/google-gemini-cli.js
-index 12540bb1069087a0d0a2967f792008627b9f79d9..f30b525620e6d8e45146b439ec3733e4053c9d2a 100644
+index cc9e0cb..2b18ec4 100644
 --- a/dist/providers/google-gemini-cli.js
 +++ b/dist/providers/google-gemini-cli.js
-@@ -248,6 +248,11 @@ export const streamGoogleGeminiCli = (model, context, options) => {
+@@ -329,6 +329,11 @@ export const streamGoogleGeminiCli = (model, context, options) => {
                          break; // Success, exit retry loop
                      }
                      const errorText = await response.text();
@@ -14,41 +14,8 @@ index 12540bb1069087a0d0a2967f792008627b9f79d9..f30b525620e6d8e45146b439ec3733e4
                      // Check if retryable
                      if (attempt < MAX_RETRIES && isRetryableError(response.status, errorText)) {
                          // Use server-provided delay or exponential backoff
-diff --git a/dist/providers/google-shared.js b/dist/providers/google-shared.js
-index ae4710b0f134ac4a48f5b7053f454d1068bee71f..b1b5bd94586f68461ccc44e4a9cdf3acb4e0d084 100644
---- a/dist/providers/google-shared.js
-+++ b/dist/providers/google-shared.js
-@@ -42,6 +42,8 @@ export function retainThoughtSignature(existing, incoming) {
- export function convertMessages(model, context) {
-     const contents = [];
-     const transformedMessages = transformMessages(context.messages, model);
-+    const shouldStripFunctionId = typeof model.provider === "string" &&
-+        model.provider.startsWith("google");
-     for (const msg of transformedMessages) {
-         if (msg.role === "user") {
-             if (typeof msg.content === "string") {
-@@ -113,6 +115,9 @@ export function convertMessages(model, context) {
-                             args: block.arguments,
-                         },
-                     };
-+                    if (shouldStripFunctionId && part?.functionCall?.id) {
-+                        delete part.functionCall.id; // Google Gemini/Vertex do not support 'id' in functionCall
-+                    }
-                     if (block.thoughtSignature) {
-                         part.thoughtSignature = block.thoughtSignature;
-                     }
-@@ -155,6 +160,9 @@ export function convertMessages(model, context) {
-                     ...(hasImages && supportsMultimodalFunctionResponse && { parts: imageParts }),
-                 },
-             };
-+            if (shouldStripFunctionId && functionResponsePart.functionResponse?.id) {
-+                delete functionResponsePart.functionResponse.id; // Google Gemini/Vertex do not support 'id' in functionResponse
-+            }
-             // Cloud Code Assist API requires all function responses to be in a single user turn.
-             // Check if the last content is already a user turn with function responses and merge.
-             const lastContent = contents[contents.length - 1];
 diff --git a/dist/providers/openai-codex-responses.js b/dist/providers/openai-codex-responses.js
-index ad0a2aabbe10382cee4e463b68a02864dd235e57..8c001acfd0b4e0743181c246f1bedcf8cd2ffb02 100644
+index 7488c79..4c34587 100644
 --- a/dist/providers/openai-codex-responses.js
 +++ b/dist/providers/openai-codex-responses.js
 @@ -517,7 +517,7 @@ function convertTools(tools) {
@@ -61,10 +28,10 @@ index ad0a2aabbe10382cee4e463b68a02864dd235e57..8c001acfd0b4e0743181c246f1bedcf8
  }
  function mapStopReason(status) {
 diff --git a/dist/providers/openai-responses.js b/dist/providers/openai-responses.js
-index f07085c64390b211340d6a826b28ea9c2e77302f..7f758532246cc7b062df48e9cec4e6c904b76a99 100644
+index c4714f4..4d1e6b0 100644
 --- a/dist/providers/openai-responses.js
 +++ b/dist/providers/openai-responses.js
-@@ -396,10 +396,16 @@ function convertMessages(model, context) {
+@@ -400,10 +400,16 @@ function convertMessages(model, context) {
          }
          else if (msg.role === "assistant") {
              const output = [];
@@ -81,7 +48,7 @@ index f07085c64390b211340d6a826b28ea9c2e77302f..7f758532246cc7b062df48e9cec4e6c9
                          const reasoningItem = JSON.parse(block.thinkingSignature);
                          output.push(reasoningItem);
                      }
-@@ -434,6 +440,16 @@ function convertMessages(model, context) {
+@@ -438,6 +444,16 @@ function convertMessages(model, context) {
                      });
                  }
              }

pnpm-workspace.yaml

@@ -10,4 +10,4 @@ onlyBuiltDependencies:
   - sharp
 
 patchedDependencies:
-  '@mariozechner/pi-ai@0.43.0': patches/@mariozechner__pi-ai@0.43.0.patch
+  '@mariozechner/pi-ai@0.45.3': patches/@mariozechner__pi-ai@0.45.3.patch

src/agents/model-selection.ts

@@ -7,7 +7,13 @@ export type ModelRef = {
   model: string;
 };
 
-export type ThinkLevel = "off" | "minimal" | "low" | "medium" | "high";
+export type ThinkLevel =
+  | "off"
+  | "minimal"
+  | "low"
+  | "medium"
+  | "high"
+  | "xhigh";
 
 export type ModelAliasIndex = {
   byAlias: Map<string, { alias: string; ref: ModelRef }>;

src/agents/pi-embedded-runner.ts

@@ -1046,7 +1046,7 @@ export function resolveEmbeddedSessionLane(key: string) {
 }
 
 function mapThinkingLevel(level?: ThinkLevel): ThinkingLevel {
-  // pi-agent-core supports "xhigh" too; Clawdbot doesn't surface it for now.
+  // pi-agent-core supports "xhigh"; Clawdbot enables it for specific models.
   if (!level) return "off";
   return level;
 }

src/auto-reply/commands-registry.ts

@@ -251,6 +251,13 @@ export const CHAT_COMMANDS: ChatCommandDefinition[] = (() => {
       textAlias: "/queue",
       acceptsArgs: true,
     }),
+    defineChatCommand({
+      key: "bash",
+      description: "Run host shell commands (host-only).",
+      textAlias: "/bash",
+      scope: "text",
+      acceptsArgs: true,
+    }),
   ];
 
   registerAlias(commands, "status", "/usage");
@@ -314,6 +321,7 @@ export function isCommandEnabled(
 ): boolean {
   if (commandKey === "config") return cfg.commands?.config === true;
   if (commandKey === "debug") return cfg.commands?.debug === true;
+  if (commandKey === "bash") return cfg.commands?.bash === true;
   return true;
 }
 

src/auto-reply/reply.directive.test.ts

@@ -69,6 +69,97 @@ describe("directive behavior", () => {
     vi.restoreAllMocks();
   });
 
+  it("accepts /thinking xhigh for codex models", async () => {
+    await withTempHome(async (home) => {
+      const storePath = path.join(home, "sessions.json");
+
+      const res = await getReplyFromConfig(
+        {
+          Body: "/thinking xhigh",
+          From: "+1004",
+          To: "+2000",
+        },
+        {},
+        {
+          agents: {
+            defaults: {
+              model: "openai-codex/gpt-5.2-codex",
+              workspace: path.join(home, "clawd"),
+            },
+          },
+          whatsapp: { allowFrom: ["*"] },
+          session: { store: storePath },
+        },
+      );
+
+      const texts = (Array.isArray(res) ? res : [res])
+        .map((entry) => entry?.text)
+        .filter(Boolean);
+      expect(texts).toContain("Thinking level set to xhigh.");
+    });
+  });
+
+  it("accepts /thinking xhigh for openai gpt-5.2", async () => {
+    await withTempHome(async (home) => {
+      const storePath = path.join(home, "sessions.json");
+
+      const res = await getReplyFromConfig(
+        {
+          Body: "/thinking xhigh",
+          From: "+1004",
+          To: "+2000",
+        },
+        {},
+        {
+          agents: {
+            defaults: {
+              model: "openai/gpt-5.2",
+              workspace: path.join(home, "clawd"),
+            },
+          },
+          whatsapp: { allowFrom: ["*"] },
+          session: { store: storePath },
+        },
+      );
+
+      const texts = (Array.isArray(res) ? res : [res])
+        .map((entry) => entry?.text)
+        .filter(Boolean);
+      expect(texts).toContain("Thinking level set to xhigh.");
+    });
+  });
+
+  it("rejects /thinking xhigh for non-codex models", async () => {
+    await withTempHome(async (home) => {
+      const storePath = path.join(home, "sessions.json");
+
+      const res = await getReplyFromConfig(
+        {
+          Body: "/thinking xhigh",
+          From: "+1004",
+          To: "+2000",
+        },
+        {},
+        {
+          agents: {
+            defaults: {
+              model: "openai/gpt-4.1-mini",
+              workspace: path.join(home, "clawd"),
+            },
+          },
+          whatsapp: { allowFrom: ["*"] },
+          session: { store: storePath },
+        },
+      );
+
+      const texts = (Array.isArray(res) ? res : [res])
+        .map((entry) => entry?.text)
+        .filter(Boolean);
+      expect(texts).toContain(
+        'Thinking level "xhigh" is only supported for openai/gpt-5.2, openai-codex/gpt-5.2-codex or openai-codex/gpt-5.1-codex.',
+      );
+    });
+  });
   it("keeps reserved command aliases from matching after trimming", async () => {
     await withTempHome(async (home) => {
       vi.mocked(runEmbeddedPiAgent).mockReset();

src/auto-reply/reply.ts

@@ -29,7 +29,10 @@ import {
   type ClawdbotConfig,
   loadConfig,
 } from "../config/config.js";
-import { resolveSessionFilePath } from "../config/sessions.js";
+import {
+  resolveSessionFilePath,
+  saveSessionStore,
+} from "../config/sessions.js";
 import { logVerbose } from "../globals.js";
 import { clearCommandLane, getQueueSize } from "../process/command-queue.js";
 import { getProviderDock } from "../providers/dock.js";
@@ -94,8 +97,10 @@ import {
 import type { MsgContext, TemplateContext } from "./templating.js";
 import {
   type ElevatedLevel,
+  formatXHighModelHint,
   normalizeThinkLevel,
   type ReasoningLevel,
+  supportsXHighThinking,
   type ThinkLevel,
   type VerboseLevel,
 } from "./thinking.js";
@@ -976,6 +981,11 @@ export async function getReplyFromConfig(
       command: inlineCommandContext,
       agentId,
       directives,
+      elevated: {
+        enabled: elevatedEnabled,
+        allowed: elevatedAllowed,
+        failures: elevatedFailures,
+      },
       sessionEntry,
       sessionStore,
       sessionKey,
@@ -1033,6 +1043,11 @@ export async function getReplyFromConfig(
     command,
     agentId,
     directives,
+    elevated: {
+      enabled: elevatedEnabled,
+      allowed: elevatedAllowed,
+      failures: elevatedFailures,
+    },
     sessionEntry,
     sessionStore,
     sessionKey,
@@ -1187,7 +1202,10 @@ export async function getReplyFromConfig(
   if (!resolvedThinkLevel && prefixedCommandBody) {
     const parts = prefixedCommandBody.split(/\s+/);
     const maybeLevel = normalizeThinkLevel(parts[0]);
-    if (maybeLevel) {
+    if (
+      maybeLevel &&
+      (maybeLevel !== "xhigh" || supportsXHighThinking(provider, model))
+    ) {
       resolvedThinkLevel = maybeLevel;
       prefixedCommandBody = parts.slice(1).join(" ").trim();
     }
@@ -1195,6 +1213,33 @@ export async function getReplyFromConfig(
   if (!resolvedThinkLevel) {
     resolvedThinkLevel = await modelState.resolveDefaultThinkingLevel();
   }
+  if (
+    resolvedThinkLevel === "xhigh" &&
+    !supportsXHighThinking(provider, model)
+  ) {
+    const explicitThink =
+      directives.hasThinkDirective && directives.thinkLevel !== undefined;
+    if (explicitThink) {
+      typing.cleanup();
+      return {
+        text: `Thinking level "xhigh" is only supported for ${formatXHighModelHint()}. Use /think high or switch to one of those models.`,
+      };
+    }
+    resolvedThinkLevel = "high";
+    if (
+      sessionEntry &&
+      sessionStore &&
+      sessionKey &&
+      sessionEntry.thinkingLevel === "xhigh"
+    ) {
+      sessionEntry.thinkingLevel = "high";
+      sessionEntry.updatedAt = Date.now();
+      sessionStore[sessionKey] = sessionEntry;
+      if (storePath) {
+        await saveSessionStore(storePath, sessionStore);
+      }
+    }
+  }
   const sessionIdFinal = sessionId ?? crypto.randomUUID();
   const sessionFile = resolveSessionFilePath(sessionIdFinal, sessionEntry);
   const queueBodyBase = transcribedText

src/auto-reply/reply/bash-command.ts

@@ -0,0 +1,444 @@
+import { resolveSessionAgentId } from "../../agents/agent-scope.js";
+import {
+  getFinishedSession,
+  getSession,
+  markExited,
+} from "../../agents/bash-process-registry.js";
+import { createExecTool } from "../../agents/bash-tools.js";
+import { resolveSandboxRuntimeStatus } from "../../agents/sandbox.js";
+import { killProcessTree } from "../../agents/shell-utils.js";
+import type { ClawdbotConfig } from "../../config/config.js";
+import { logVerbose } from "../../globals.js";
+import type { MsgContext } from "../templating.js";
+import type { ReplyPayload } from "../types.js";
+import { stripMentions, stripStructuralPrefixes } from "./mentions.js";
+
+const CHAT_BASH_SCOPE_KEY = "chat:bash";
+const DEFAULT_FOREGROUND_MS = 2000;
+const MAX_FOREGROUND_MS = 30_000;
+
+type BashRequest =
+  | { action: "help" }
+  | { action: "run"; command: string }
+  | { action: "poll"; sessionId?: string }
+  | { action: "stop"; sessionId?: string };
+
+type ActiveBashJob =
+  | { state: "starting"; startedAt: number; command: string }
+  | {
+      state: "running";
+      sessionId: string;
+      startedAt: number;
+      command: string;
+      watcherAttached: boolean;
+    };
+
+let activeJob: ActiveBashJob | null = null;
+
+function clampNumber(value: number, min: number, max: number) {
+  return Math.min(Math.max(value, min), max);
+}
+
+function resolveForegroundMs(cfg: ClawdbotConfig): number {
+  const raw = cfg.commands?.bashForegroundMs;
+  if (typeof raw !== "number" || Number.isNaN(raw))
+    return DEFAULT_FOREGROUND_MS;
+  return clampNumber(Math.floor(raw), 0, MAX_FOREGROUND_MS);
+}
+
+function formatSessionSnippet(sessionId: string) {
+  const trimmed = sessionId.trim();
+  if (trimmed.length <= 12) return trimmed;
+  return `${trimmed.slice(0, 8)}…`;
+}
+
+function formatOutputBlock(text: string) {
+  const trimmed = text.trim();
+  if (!trimmed) return "(no output)";
+  return `\`\`\`txt\n${trimmed}\n\`\`\``;
+}
+
+function parseBashRequest(raw: string): BashRequest | null {
+  const trimmed = raw.trimStart();
+  let restSource = "";
+  if (trimmed.toLowerCase().startsWith("/bash")) {
+    const match = trimmed.match(/^\/bash(?:\s*:\s*|\s+|$)([\s\S]*)$/i);
+    if (!match) return null;
+    restSource = match[1] ?? "";
+  } else if (trimmed.startsWith("!")) {
+    restSource = trimmed.slice(1);
+    if (restSource.trimStart().startsWith(":")) {
+      restSource = restSource.trimStart().slice(1);
+    }
+  } else {
+    return null;
+  }
+
+  const rest = restSource.trimStart();
+  if (!rest) return { action: "help" };
+  const tokenMatch = rest.match(/^(\S+)(?:\s+([\s\S]+))?$/);
+  const token = tokenMatch?.[1]?.trim() ?? "";
+  const remainder = tokenMatch?.[2]?.trim() ?? "";
+  const lowered = token.toLowerCase();
+  if (lowered === "poll") {
+    return { action: "poll", sessionId: remainder || undefined };
+  }
+  if (lowered === "stop") {
+    return { action: "stop", sessionId: remainder || undefined };
+  }
+  if (lowered === "help") {
+    return { action: "help" };
+  }
+  return { action: "run", command: rest };
+}
+
+function resolveRawCommandBody(params: {
+  ctx: MsgContext;
+  cfg: ClawdbotConfig;
+  agentId?: string;
+  isGroup: boolean;
+}) {
+  const source =
+    params.ctx.CommandBody ?? params.ctx.RawBody ?? params.ctx.Body ?? "";
+  const stripped = stripStructuralPrefixes(source);
+  return params.isGroup
+    ? stripMentions(stripped, params.ctx, params.cfg, params.agentId)
+    : stripped;
+}
+
+function getScopedSession(sessionId: string) {
+  const running = getSession(sessionId);
+  if (running && running.scopeKey === CHAT_BASH_SCOPE_KEY) return { running };
+  const finished = getFinishedSession(sessionId);
+  if (finished && finished.scopeKey === CHAT_BASH_SCOPE_KEY)
+    return { finished };
+  return {};
+}
+
+function ensureActiveJobState() {
+  if (!activeJob) return null;
+  if (activeJob.state === "starting") return activeJob;
+  const { running, finished } = getScopedSession(activeJob.sessionId);
+  if (running) return activeJob;
+  if (finished) {
+    activeJob = null;
+    return null;
+  }
+  activeJob = null;
+  return null;
+}
+
+function attachActiveWatcher(sessionId: string) {
+  if (!activeJob || activeJob.state !== "running") return;
+  if (activeJob.sessionId !== sessionId) return;
+  if (activeJob.watcherAttached) return;
+  const { running } = getScopedSession(sessionId);
+  const child = running?.child;
+  if (!child) return;
+  activeJob.watcherAttached = true;
+  child.once("close", () => {
+    if (activeJob?.state === "running" && activeJob.sessionId === sessionId) {
+      activeJob = null;
+    }
+  });
+}
+
+function buildUsageReply(): ReplyPayload {
+  return {
+    text: [
+      "⚙️ Usage:",
+      "- ! <command>",
+      "- !poll | ! poll",
+      "- !stop | ! stop",
+      "- /bash ... (alias; same subcommands as !)",
+    ].join("\n"),
+  };
+}
+
+function buildBashDisabledReply(provider?: string): ReplyPayload {
+  const providerKey = provider?.trim().toLowerCase() || "<provider>";
+  return {
+    text: [
+      "⚠️ /bash is disabled (default).",
+      "Enable it in your config (host-only; trusted senders only):",
+      "```json5",
+      "{",
+      "  commands: {",
+      "    bash: true",
+      "  },",
+      "  tools: {",
+      "    elevated: {",
+      "      enabled: true,",
+      `      allowFrom: { "${providerKey}": ["<sender-id>"] }`,
+      "    }",
+      "  }",
+      "}",
+      "```",
+      "Restart the Gateway after updating config.",
+    ].join("\n"),
+  };
+}
+
+function formatElevatedUnavailableMessage(params: {
+  runtimeSandboxed: boolean;
+  failures: Array<{ gate: string; key: string }>;
+  sessionKey?: string;
+}): string {
+  const lines: string[] = [];
+  lines.push(
+    `elevated is not available right now (runtime=${params.runtimeSandboxed ? "sandboxed" : "direct"}).`,
+  );
+  if (params.failures.length > 0) {
+    lines.push(
+      `Failing gates: ${params.failures
+        .map((f) => `${f.gate} (${f.key})`)
+        .join(", ")}`,
+    );
+  } else {
+    lines.push(
+      "Failing gates: enabled (tools.elevated.enabled / agents.list[].tools.elevated.enabled), allowFrom (tools.elevated.allowFrom.<provider>).",
+    );
+  }
+  lines.push("Fix-it keys:");
+  lines.push("- tools.elevated.enabled");
+  lines.push("- tools.elevated.allowFrom.<provider>");
+  lines.push("- agents.list[].tools.elevated.enabled");
+  lines.push("- agents.list[].tools.elevated.allowFrom.<provider>");
+  if (params.sessionKey) {
+    lines.push(`See: clawdbot sandbox explain --session ${params.sessionKey}`);
+  }
+  return lines.join("\n");
+}
+
+export async function handleBashChatCommand(params: {
+  ctx: MsgContext;
+  cfg: ClawdbotConfig;
+  agentId?: string;
+  sessionKey: string;
+  isGroup: boolean;
+  elevated: {
+    enabled: boolean;
+    allowed: boolean;
+    failures: Array<{ gate: string; key: string }>;
+  };
+}): Promise<ReplyPayload> {
+  if (params.cfg.commands?.bash !== true) {
+    logVerbose("Blocked /bash: commands.bash is disabled.");
+    const provider =
+      params.ctx.Provider ??
+      params.ctx.Surface ??
+      params.ctx.OriginatingChannel ??
+      undefined;
+    return buildBashDisabledReply(provider ?? undefined);
+  }
+
+  const agentId =
+    params.agentId ??
+    resolveSessionAgentId({
+      sessionKey: params.sessionKey,
+      config: params.cfg,
+    });
+
+  if (!params.elevated.enabled || !params.elevated.allowed) {
+    const runtimeSandboxed = resolveSandboxRuntimeStatus({
+      cfg: params.cfg,
+      sessionKey: params.ctx.SessionKey,
+    }).sandboxed;
+    return {
+      text: formatElevatedUnavailableMessage({
+        runtimeSandboxed,
+        failures: params.elevated.failures,
+        sessionKey: params.ctx.SessionKey,
+      }),
+    };
+  }
+
+  const rawBody = resolveRawCommandBody({
+    ctx: params.ctx,
+    cfg: params.cfg,
+    agentId,
+    isGroup: params.isGroup,
+  }).trim();
+  const request = parseBashRequest(rawBody);
+  if (!request) {
+    return { text: "⚠️ Unrecognized bash request." };
+  }
+
+  const liveJob = ensureActiveJobState();
+
+  if (request.action === "help") {
+    return buildUsageReply();
+  }
+
+  if (request.action === "poll") {
+    const sessionId =
+      request.sessionId?.trim() ||
+      (liveJob?.state === "running" ? liveJob.sessionId : "");
+    if (!sessionId) {
+      return { text: "⚙️ No active bash job." };
+    }
+    const { running, finished } = getScopedSession(sessionId);
+    if (running) {
+      attachActiveWatcher(sessionId);
+      const runtimeSec = Math.max(
+        0,
+        Math.floor((Date.now() - running.startedAt) / 1000),
+      );
+      const tail = running.tail || "(no output yet)";
+      return {
+        text: [
+          `⚙️ bash still running (session ${formatSessionSnippet(sessionId)}, ${runtimeSec}s).`,
+          formatOutputBlock(tail),
+          "Hint: !stop (or /bash stop)",
+        ].join("\n"),
+      };
+    }
+    if (finished) {
+      if (activeJob?.state === "running" && activeJob.sessionId === sessionId) {
+        activeJob = null;
+      }
+      const exitLabel = finished.exitSignal
+        ? `signal ${String(finished.exitSignal)}`
+        : `code ${String(finished.exitCode ?? 0)}`;
+      const prefix = finished.status === "completed" ? "⚙️" : "⚠️";
+      return {
+        text: [
+          `${prefix} bash finished (session ${formatSessionSnippet(sessionId)}).`,
+          `Exit: ${exitLabel}`,
+          formatOutputBlock(finished.aggregated || finished.tail),
+        ].join("\n"),
+      };
+    }
+    if (activeJob?.state === "running" && activeJob.sessionId === sessionId) {
+      activeJob = null;
+    }
+    return {
+      text: `⚙️ No bash session found for ${formatSessionSnippet(sessionId)}.`,
+    };
+  }
+
+  if (request.action === "stop") {
+    const sessionId =
+      request.sessionId?.trim() ||
+      (liveJob?.state === "running" ? liveJob.sessionId : "");
+    if (!sessionId) {
+      return { text: "⚙️ No active bash job." };
+    }
+    const { running } = getScopedSession(sessionId);
+    if (!running) {
+      if (activeJob?.state === "running" && activeJob.sessionId === sessionId) {
+        activeJob = null;
+      }
+      return {
+        text: `⚙️ No running bash job found for ${formatSessionSnippet(sessionId)}.`,
+      };
+    }
+    if (!running.backgrounded) {
+      return {
+        text: `⚠️ Session ${formatSessionSnippet(sessionId)} is not backgrounded.`,
+      };
+    }
+    const pid = running.pid ?? running.child?.pid;
+    if (pid) {
+      killProcessTree(pid);
+    }
+    markExited(running, null, "SIGKILL", "failed");
+    if (activeJob?.state === "running" && activeJob.sessionId === sessionId) {
+      activeJob = null;
+    }
+    return {
+      text: `⚙️ bash stopped (session ${formatSessionSnippet(sessionId)}).`,
+    };
+  }
+
+  // request.action === "run"
+  if (liveJob) {
+    const label =
+      liveJob.state === "running"
+        ? formatSessionSnippet(liveJob.sessionId)
+        : "starting";
+    return {
+      text: `⚠️ A bash job is already running (${label}). Use !poll / !stop (or /bash poll / /bash stop).`,
+    };
+  }
+
+  const commandText = request.command.trim();
+  if (!commandText) return buildUsageReply();
+
+  activeJob = {
+    state: "starting",
+    startedAt: Date.now(),
+    command: commandText,
+  };
+
+  try {
+    const foregroundMs = resolveForegroundMs(params.cfg);
+    const shouldBackgroundImmediately = foregroundMs <= 0;
+    const timeoutSec =
+      params.cfg.tools?.exec?.timeoutSec ?? params.cfg.tools?.bash?.timeoutSec;
+    const execTool = createExecTool({
+      scopeKey: CHAT_BASH_SCOPE_KEY,
+      allowBackground: true,
+      timeoutSec,
+      elevated: {
+        enabled: params.elevated.enabled,
+        allowed: params.elevated.allowed,
+        defaultLevel: "on",
+      },
+    });
+    const result = await execTool.execute("chat-bash", {
+      command: commandText,
+      background: shouldBackgroundImmediately,
+      yieldMs: shouldBackgroundImmediately ? undefined : foregroundMs,
+      timeout: timeoutSec,
+      elevated: true,
+    });
+
+    if (result.details?.status === "running") {
+      const sessionId = result.details.sessionId;
+      activeJob = {
+        state: "running",
+        sessionId,
+        startedAt: result.details.startedAt,
+        command: commandText,
+        watcherAttached: false,
+      };
+      attachActiveWatcher(sessionId);
+      const snippet = formatSessionSnippet(sessionId);
+      logVerbose(`Started bash session ${snippet}: ${commandText}`);
+      return {
+        text: `⚙️ bash started (session ${sessionId}). Still running; use !poll / !stop (or /bash poll / /bash stop).`,
+      };
+    }
+
+    // Completed in foreground.
+    activeJob = null;
+    const exitCode =
+      result.details?.status === "completed" ? result.details.exitCode : 0;
+    const output =
+      result.details?.status === "completed"
+        ? result.details.aggregated
+        : result.content
+            .map((chunk) => (chunk.type === "text" ? chunk.text : ""))
+            .join("\n");
+    return {
+      text: [
+        `⚙️ bash: ${commandText}`,
+        `Exit: ${exitCode}`,
+        formatOutputBlock(output || "(no output)"),
+      ].join("\n"),
+    };
+  } catch (err) {
+    activeJob = null;
+    const message = err instanceof Error ? err.message : String(err);
+    return {
+      text: [`⚠️ bash failed: ${commandText}`, formatOutputBlock(message)].join(
+        "\n",
+      ),
+    };
+  }
+}
+
+export function resetBashChatCommandForTests() {
+  activeJob = null;
+}

src/auto-reply/reply/commands.test.ts

@@ -2,6 +2,7 @@ import { describe, expect, it } from "vitest";
 
 import type { ClawdbotConfig } from "../../config/config.js";
 import type { MsgContext } from "../templating.js";
+import { resetBashChatCommandForTests } from "./bash-command.js";
 import { buildCommandContext, handleCommands } from "./commands.js";
 import { parseInlineDirectives } from "./directive-handling.js";
 
@@ -33,6 +34,7 @@ function buildParams(
     cfg,
     command,
     directives: parseInlineDirectives(commandBody),
+    elevated: { enabled: true, allowed: true, failures: [] },
     sessionKey: "agent:main:main",
     workspaceDir: "/tmp",
     defaultGroupActivation: () => "mention",
@@ -47,6 +49,37 @@ function buildParams(
 }
 
 describe("handleCommands gating", () => {
+  it("blocks /bash when disabled", async () => {
+    resetBashChatCommandForTests();
+    const cfg = {
+      commands: { bash: false, text: true },
+      whatsapp: { allowFrom: ["*"] },
+    } as ClawdbotConfig;
+    const params = buildParams("/bash echo hi", cfg);
+    const result = await handleCommands(params);
+    expect(result.shouldContinue).toBe(false);
+    expect(result.reply?.text).toContain("bash is disabled");
+  });
+
+  it("blocks /bash when elevated is not allowlisted", async () => {
+    resetBashChatCommandForTests();
+    const cfg = {
+      commands: { bash: true, text: true },
+      whatsapp: { allowFrom: ["*"] },
+    } as ClawdbotConfig;
+    const params = buildParams("/bash echo hi", cfg);
+    params.elevated = {
+      enabled: true,
+      allowed: false,
+      failures: [
+        { gate: "allowFrom", key: "tools.elevated.allowFrom.whatsapp" },
+      ],
+    };
+    const result = await handleCommands(params);
+    expect(result.shouldContinue).toBe(false);
+    expect(result.reply?.text).toContain("elevated is not available");
+  });
+
   it("blocks /config when disabled", async () => {
     const cfg = {
       commands: { config: false, debug: false, text: true },
@@ -70,6 +103,32 @@ describe("handleCommands gating", () => {
   });
 });
 
+describe("handleCommands bash alias", () => {
+  it("routes !poll through the /bash handler", async () => {
+    resetBashChatCommandForTests();
+    const cfg = {
+      commands: { bash: true, text: true },
+      whatsapp: { allowFrom: ["*"] },
+    } as ClawdbotConfig;
+    const params = buildParams("!poll", cfg);
+    const result = await handleCommands(params);
+    expect(result.shouldContinue).toBe(false);
+    expect(result.reply?.text).toContain("No active bash job");
+  });
+
+  it("routes !stop through the /bash handler", async () => {
+    resetBashChatCommandForTests();
+    const cfg = {
+      commands: { bash: true, text: true },
+      whatsapp: { allowFrom: ["*"] },
+    } as ClawdbotConfig;
+    const params = buildParams("!stop", cfg);
+    const result = await handleCommands(params);
+    expect(result.shouldContinue).toBe(false);
+    expect(result.reply?.text).toContain("No active bash job");
+  });
+});
+
 describe("handleCommands identity", () => {
   it("returns sender details for /whoami", async () => {
     const cfg = {

src/auto-reply/reply/commands.ts

@@ -83,6 +83,7 @@ import type {
 } from "../thinking.js";
 import type { ReplyPayload } from "../types.js";
 import { isAbortTrigger, setAbortMemory } from "./abort.js";
+import { handleBashChatCommand } from "./bash-command.js";
 import { parseConfigCommand } from "./config-commands.js";
 import { parseDebugCommand } from "./debug-commands.js";
 import type { InlineDirectives } from "./directive-handling.js";
@@ -400,6 +401,11 @@ export async function handleCommands(params: {
   command: CommandContext;
   agentId?: string;
   directives: InlineDirectives;
+  elevated: {
+    enabled: boolean;
+    allowed: boolean;
+    failures: Array<{ gate: string; key: string }>;
+  };
   sessionEntry?: SessionEntry;
   sessionStore?: Record<string, SessionEntry>;
   sessionKey: string;
@@ -425,6 +431,7 @@ export async function handleCommands(params: {
     cfg,
     command,
     directives,
+    elevated,
     sessionEntry,
     sessionStore,
     sessionKey,
@@ -465,6 +472,30 @@ export async function handleCommands(params: {
     commandSource: ctx.CommandSource,
   });
 
+  const bashSlashRequested =
+    allowTextCommands &&
+    (command.commandBodyNormalized === "/bash" ||
+      command.commandBodyNormalized.startsWith("/bash "));
+  const bashBangRequested =
+    allowTextCommands && command.commandBodyNormalized.startsWith("!");
+  if (bashSlashRequested || (bashBangRequested && command.isAuthorizedSender)) {
+    if (!command.isAuthorizedSender) {
+      logVerbose(
+        `Ignoring /bash from unauthorized sender: ${command.senderId || "<unknown>"}`,
+      );
+      return { shouldContinue: false };
+    }
+    const reply = await handleBashChatCommand({
+      ctx,
+      cfg,
+      agentId: params.agentId,
+      sessionKey,
+      isGroup,
+      elevated,
+    });
+    return { shouldContinue: false, reply };
+  }
+
   if (allowTextCommands && activationCommand.hasCommand) {
     if (!isGroup) {
       return {

src/auto-reply/reply/directive-handling.ts

@@ -37,6 +37,11 @@ import { applyVerboseOverride } from "../../sessions/level-overrides.js";
 import { shortenHomePath } from "../../utils.js";
 import { extractModelDirective } from "../model.js";
 import type { MsgContext } from "../templating.js";
+import {
+  formatThinkingLevels,
+  formatXHighModelHint,
+  supportsXHighThinking,
+} from "../thinking.js";
 import type { ReplyPayload } from "../types.js";
 import {
   type ElevatedLevel,
@@ -778,6 +783,7 @@ export async function handleDirectiveOnly(params: {
     allowedModelCatalog,
     resetModelOverride,
     provider,
+    model,
     initialModelLabel,
     formatModelSwitchEvent,
     currentThinkLevel,
@@ -943,6 +949,117 @@ export async function handleDirectiveOnly(params: {
     }
   }
 
+  let modelSelection: ModelDirectiveSelection | undefined;
+  let profileOverride: string | undefined;
+  if (directives.hasModelDirective && directives.rawModelDirective) {
+    const raw = directives.rawModelDirective.trim();
+    if (/^[0-9]+$/.test(raw)) {
+      const resolvedDefault = resolveConfiguredModelRef({
+        cfg: params.cfg,
+        defaultProvider,
+        defaultModel,
+      });
+      const pickerCatalog: ModelPickerCatalogEntry[] = (() => {
+        const keys = new Set<string>();
+        const out: ModelPickerCatalogEntry[] = [];
+
+        const push = (entry: ModelPickerCatalogEntry) => {
+          const provider = normalizeProviderId(entry.provider);
+          const id = String(entry.id ?? "").trim();
+          if (!provider || !id) return;
+          const key = modelKey(provider, id);
+          if (keys.has(key)) return;
+          keys.add(key);
+          out.push({ provider, id, name: entry.name });
+        };
+
+        for (const entry of allowedModelCatalog) push(entry);
+
+        for (const rawKey of Object.keys(
+          params.cfg.agents?.defaults?.models ?? {},
+        )) {
+          const resolved = resolveModelRefFromString({
+            raw: String(rawKey),
+            defaultProvider,
+            aliasIndex,
+          });
+          if (!resolved) continue;
+          push({
+            provider: resolved.ref.provider,
+            id: resolved.ref.model,
+            name: resolved.ref.model,
+          });
+        }
+        if (resolvedDefault.model) {
+          push({
+            provider: resolvedDefault.provider,
+            id: resolvedDefault.model,
+            name: resolvedDefault.model,
+          });
+        }
+        return out;
+      })();
+
+      const items = buildModelPickerItems(pickerCatalog);
+      const index = Number.parseInt(raw, 10) - 1;
+      const item = Number.isFinite(index) ? items[index] : undefined;
+      if (!item) {
+        return {
+          text: `Invalid model selection "${raw}". Use /model to list.`,
+        };
+      }
+      const picked = pickProviderForModel({
+        item,
+        preferredProvider: params.provider,
+      });
+      if (!picked) {
+        return {
+          text: `Invalid model selection "${raw}". Use /model to list.`,
+        };
+      }
+      const key = `${picked.provider}/${picked.model}`;
+      const aliases = aliasIndex.byKey.get(key);
+      const alias = aliases && aliases.length > 0 ? aliases[0] : undefined;
+      modelSelection = {
+        provider: picked.provider,
+        model: picked.model,
+        isDefault:
+          picked.provider === defaultProvider && picked.model === defaultModel,
+        ...(alias ? { alias } : {}),
+      };
+    } else {
+      const resolved = resolveModelDirectiveSelection({
+        raw,
+        defaultProvider,
+        defaultModel,
+        aliasIndex,
+        allowedModelKeys,
+      });
+      if (resolved.error) {
+        return { text: resolved.error };
+      }
+      modelSelection = resolved.selection;
+    }
+    if (modelSelection && directives.rawModelProfile) {
+      const profileResolved = resolveProfileOverride({
+        rawProfile: directives.rawModelProfile,
+        provider: modelSelection.provider,
+        cfg: params.cfg,
+        agentDir,
+      });
+      if (profileResolved.error) {
+        return { text: profileResolved.error };
+      }
+      profileOverride = profileResolved.profileId;
+    }
+  }
+  if (directives.rawModelProfile && !modelSelection) {
+    return { text: "Auth profile override requires a model selection." };
+  }
+
+  const resolvedProvider = modelSelection?.provider ?? provider;
+  const resolvedModel = modelSelection?.model ?? model;
+
   if (directives.hasThinkDirective && !directives.thinkLevel) {
     // If no argument was provided, show the current level
     if (!directives.rawThinkLevel) {
@@ -950,12 +1067,12 @@ export async function handleDirectiveOnly(params: {
       return {
         text: withOptions(
           `Current thinking level: ${level}.`,
-          "off, minimal, low, medium, high",
+          formatThinkingLevels(resolvedProvider, resolvedModel),
         ),
       };
     }
     return {
-      text: `Unrecognized thinking level "${directives.rawThinkLevel}". Valid levels: off, minimal, low, medium, high.`,
+      text: `Unrecognized thinking level "${directives.rawThinkLevel}". Valid levels: ${formatThinkingLevels(resolvedProvider, resolvedModel)}.`,
     };
   }
   if (directives.hasVerboseDirective && !directives.verboseLevel) {
@@ -1098,126 +1215,25 @@ export async function handleDirectiveOnly(params: {
     return { text: errors.join(" ") };
   }
 
-  let modelSelection: ModelDirectiveSelection | undefined;
+  if (
-  let profileOverride: string | undefined;
+    directives.hasThinkDirective &&
-  if (directives.hasModelDirective && directives.rawModelDirective) {
+    directives.thinkLevel === "xhigh" &&
-    const raw = directives.rawModelDirective.trim();
+    !supportsXHighThinking(resolvedProvider, resolvedModel)
-    if (/^[0-9]+$/.test(raw)) {
+  ) {
-      const resolvedDefault = resolveConfiguredModelRef({
+    return {
-        cfg: params.cfg,
+      text: `Thinking level "xhigh" is only supported for ${formatXHighModelHint()}.`,
-        defaultProvider,
+    };
-        defaultModel,
-      });
-      const pickerCatalog: ModelPickerCatalogEntry[] = (() => {
-        const keys = new Set<string>();
-        const out: ModelPickerCatalogEntry[] = [];
-
-        const push = (entry: ModelPickerCatalogEntry) => {
-          const provider = normalizeProviderId(entry.provider);
-          const id = String(entry.id ?? "").trim();
-          if (!provider || !id) return;
-          const key = modelKey(provider, id);
-          if (keys.has(key)) return;
-          keys.add(key);
-          out.push({ provider, id, name: entry.name });
-        };
-
-        for (const entry of allowedModelCatalog) push(entry);
-
-        for (const rawKey of Object.keys(
-          params.cfg.agents?.defaults?.models ?? {},
-        )) {
-          const resolved = resolveModelRefFromString({
-            raw: String(rawKey),
-            defaultProvider,
-            aliasIndex,
-          });
-          if (!resolved) continue;
-          push({
-            provider: resolved.ref.provider,
-            id: resolved.ref.model,
-            name: resolved.ref.model,
-          });
-        }
-        if (resolvedDefault.model) {
-          push({
-            provider: resolvedDefault.provider,
-            id: resolvedDefault.model,
-            name: resolvedDefault.model,
-          });
-        }
-        return out;
-      })();
-
-      const items = buildModelPickerItems(pickerCatalog);
-      const index = Number.parseInt(raw, 10) - 1;
-      const item = Number.isFinite(index) ? items[index] : undefined;
-      if (!item) {
-        return {
-          text: `Invalid model selection "${raw}". Use /model to list.`,
-        };
-      }
-      const picked = pickProviderForModel({
-        item,
-        preferredProvider: params.provider,
-      });
-      if (!picked) {
-        return {
-          text: `Invalid model selection "${raw}". Use /model to list.`,
-        };
-      }
-      const key = `${picked.provider}/${picked.model}`;
-      const aliases = aliasIndex.byKey.get(key);
-      const alias = aliases && aliases.length > 0 ? aliases[0] : undefined;
-      modelSelection = {
-        provider: picked.provider,
-        model: picked.model,
-        isDefault:
-          picked.provider === defaultProvider && picked.model === defaultModel,
-        ...(alias ? { alias } : {}),
-      };
-    } else {
-      const resolved = resolveModelDirectiveSelection({
-        raw,
-        defaultProvider,
-        defaultModel,
-        aliasIndex,
-        allowedModelKeys,
-      });
-      if (resolved.error) {
-        return { text: resolved.error };
-      }
-      modelSelection = resolved.selection;
-    }
-    if (modelSelection) {
-      if (directives.rawModelProfile) {
-        const profileResolved = resolveProfileOverride({
-          rawProfile: directives.rawModelProfile,
-          provider: modelSelection.provider,
-          cfg: params.cfg,
-          agentDir,
-        });
-        if (profileResolved.error) {
-          return { text: profileResolved.error };
-        }
-        profileOverride = profileResolved.profileId;
-      }
-      const nextLabel = `${modelSelection.provider}/${modelSelection.model}`;
-      if (nextLabel !== initialModelLabel) {
-        enqueueSystemEvent(
-          formatModelSwitchEvent(nextLabel, modelSelection.alias),
-          {
-            sessionKey,
-            contextKey: `model:${nextLabel}`,
-          },
-        );
-      }
-    }
-  }
-  if (directives.rawModelProfile && !modelSelection) {
-    return { text: "Auth profile override requires a model selection." };
   }
 
+  const nextThinkLevel = directives.hasThinkDirective
+    ? directives.thinkLevel
+    : ((sessionEntry?.thinkingLevel as ThinkLevel | undefined) ??
+      currentThinkLevel);
+  const shouldDowngradeXHigh =
+    !directives.hasThinkDirective &&
+    nextThinkLevel === "xhigh" &&
+    !supportsXHighThinking(resolvedProvider, resolvedModel);
+
   if (sessionEntry && sessionStore && sessionKey) {
     const prevElevatedLevel =
       currentElevatedLevel ??
@@ -1239,6 +1255,9 @@ export async function handleDirectiveOnly(params: {
       if (directives.thinkLevel === "off") delete sessionEntry.thinkingLevel;
       else sessionEntry.thinkingLevel = directives.thinkLevel;
     }
+    if (shouldDowngradeXHigh) {
+      sessionEntry.thinkingLevel = "high";
+    }
     if (directives.hasVerboseDirective && directives.verboseLevel) {
       applyVerboseOverride(sessionEntry, directives.verboseLevel);
     }
@@ -1295,6 +1314,18 @@ export async function handleDirectiveOnly(params: {
     if (storePath) {
       await saveSessionStore(storePath, sessionStore);
     }
+    if (modelSelection) {
+      const nextLabel = `${modelSelection.provider}/${modelSelection.model}`;
+      if (nextLabel !== initialModelLabel) {
+        enqueueSystemEvent(
+          formatModelSwitchEvent(nextLabel, modelSelection.alias),
+          {
+            sessionKey,
+            contextKey: `model:${nextLabel}`,
+          },
+        );
+      }
+    }
     if (elevatedChanged) {
       const nextElevated = (sessionEntry.elevatedLevel ??
         "off") as ElevatedLevel;
@@ -1345,6 +1376,11 @@ export async function handleDirectiveOnly(params: {
     );
     if (shouldHintDirectRuntime) parts.push(formatElevatedRuntimeHint());
   }
+  if (shouldDowngradeXHigh) {
+    parts.push(
+      `Thinking level set to high (xhigh not supported for ${resolvedProvider}/${resolvedModel}).`,
+    );
+  }
   if (modelSelection) {
     const label = `${modelSelection.provider}/${modelSelection.model}`;
     const labelWithAlias = modelSelection.alias

src/auto-reply/thinking.test.ts

@@ -1,10 +1,34 @@
 import { describe, expect, it } from "vitest";
-import { normalizeReasoningLevel, normalizeThinkLevel } from "./thinking.js";
+import {
+  listThinkingLevels,
+  normalizeReasoningLevel,
+  normalizeThinkLevel,
+} from "./thinking.js";
 
 describe("normalizeThinkLevel", () => {
   it("accepts mid as medium", () => {
     expect(normalizeThinkLevel("mid")).toBe("medium");
   });
+
+  it("accepts xhigh", () => {
+    expect(normalizeThinkLevel("xhigh")).toBe("xhigh");
+  });
+});
+
+describe("listThinkingLevels", () => {
+  it("includes xhigh for codex models", () => {
+    expect(listThinkingLevels(undefined, "gpt-5.2-codex")).toContain("xhigh");
+  });
+
+  it("includes xhigh for openai gpt-5.2", () => {
+    expect(listThinkingLevels("openai", "gpt-5.2")).toContain("xhigh");
+  });
+
+  it("excludes xhigh for non-codex models", () => {
+    expect(listThinkingLevels(undefined, "gpt-4.1-mini")).not.toContain(
+      "xhigh",
+    );
+  });
 });
 
 describe("normalizeReasoningLevel", () => {

src/auto-reply/thinking.ts

@@ -1,9 +1,30 @@
-export type ThinkLevel = "off" | "minimal" | "low" | "medium" | "high";
+export type ThinkLevel =
+  | "off"
+  | "minimal"
+  | "low"
+  | "medium"
+  | "high"
+  | "xhigh";
 export type VerboseLevel = "off" | "on";
 export type ElevatedLevel = "off" | "on";
 export type ReasoningLevel = "off" | "on" | "stream";
 export type UsageDisplayLevel = "off" | "on";
 
+export const XHIGH_MODEL_REFS = [
+  "openai/gpt-5.2",
+  "openai-codex/gpt-5.2-codex",
+  "openai-codex/gpt-5.1-codex",
+] as const;
+
+const XHIGH_MODEL_SET = new Set(
+  XHIGH_MODEL_REFS.map((entry) => entry.toLowerCase()),
+);
+const XHIGH_MODEL_IDS = new Set(
+  XHIGH_MODEL_REFS.map((entry) => entry.split("/")[1]?.toLowerCase()).filter(
+    (entry): entry is string => Boolean(entry),
+  ),
+);
+
 // Normalize user-provided thinking level strings to the canonical enum.
 export function normalizeThinkLevel(
   raw?: string | null,
@@ -32,10 +53,49 @@ export function normalizeThinkLevel(
     ].includes(key)
   )
     return "high";
+  if (["xhigh", "x-high", "x_high"].includes(key)) return "xhigh";
   if (["think"].includes(key)) return "minimal";
   return undefined;
 }
 
+export function supportsXHighThinking(
+  provider?: string | null,
+  model?: string | null,
+): boolean {
+  const modelKey = model?.trim().toLowerCase();
+  if (!modelKey) return false;
+  const providerKey = provider?.trim().toLowerCase();
+  if (providerKey) {
+    return XHIGH_MODEL_SET.has(`${providerKey}/${modelKey}`);
+  }
+  return XHIGH_MODEL_IDS.has(modelKey);
+}
+
+export function listThinkingLevels(
+  provider?: string | null,
+  model?: string | null,
+): ThinkLevel[] {
+  const levels: ThinkLevel[] = ["off", "minimal", "low", "medium", "high"];
+  if (supportsXHighThinking(provider, model)) levels.push("xhigh");
+  return levels;
+}
+
+export function formatThinkingLevels(
+  provider?: string | null,
+  model?: string | null,
+  separator = ", ",
+): string {
+  return listThinkingLevels(provider, model).join(separator);
+}
+
+export function formatXHighModelHint(): string {
+  const refs = [...XHIGH_MODEL_REFS] as string[];
+  if (refs.length === 0) return "unknown model";
+  if (refs.length === 1) return refs[0];
+  if (refs.length === 2) return `${refs[0]} or ${refs[1]}`;
+  return `${refs.slice(0, -1).join(", ")} or ${refs[refs.length - 1]}`;
+}
+
 // Normalize verbose flags used to toggle agent verbosity.
 export function normalizeVerboseLevel(
   raw?: string | null,

src/commands/agent.ts

@@ -30,8 +30,11 @@ import { hasNonzeroUsage } from "../agents/usage.js";
 import { ensureAgentWorkspace } from "../agents/workspace.js";
 import type { MsgContext } from "../auto-reply/templating.js";
 import {
+  formatThinkingLevels,
+  formatXHighModelHint,
   normalizeThinkLevel,
   normalizeVerboseLevel,
+  supportsXHighThinking,
   type ThinkLevel,
   type VerboseLevel,
 } from "../auto-reply/thinking.js";
@@ -216,17 +219,26 @@ export async function agentCommand(
     ensureBootstrapFiles: !agentCfg?.skipBootstrap,
   });
   const workspaceDir = workspace.dir;
+  const configuredModel = resolveConfiguredModelRef({
+    cfg,
+    defaultProvider: DEFAULT_PROVIDER,
+    defaultModel: DEFAULT_MODEL,
+  });
+  const thinkingLevelsHint = formatThinkingLevels(
+    configuredModel.provider,
+    configuredModel.model,
+  );
 
   const thinkOverride = normalizeThinkLevel(opts.thinking);
   const thinkOnce = normalizeThinkLevel(opts.thinkingOnce);
   if (opts.thinking && !thinkOverride) {
     throw new Error(
-      "Invalid thinking level. Use one of: off, minimal, low, medium, high.",
+      `Invalid thinking level. Use one of: ${thinkingLevelsHint}.`,
     );
   }
   if (opts.thinkingOnce && !thinkOnce) {
     throw new Error(
-      "Invalid one-shot thinking level. Use one of: off, minimal, low, medium, high.",
+      `Invalid one-shot thinking level. Use one of: ${thinkingLevelsHint}.`,
     );
   }
 
@@ -444,6 +456,29 @@ export async function agentCommand(
       catalog: catalogForThinking,
     });
   }
+  if (
+    resolvedThinkLevel === "xhigh" &&
+    !supportsXHighThinking(provider, model)
+  ) {
+    const explicitThink = Boolean(thinkOnce || thinkOverride);
+    if (explicitThink) {
+      throw new Error(
+        `Thinking level "xhigh" is only supported for ${formatXHighModelHint()}.`,
+      );
+    }
+    resolvedThinkLevel = "high";
+    if (
+      sessionEntry &&
+      sessionStore &&
+      sessionKey &&
+      sessionEntry.thinkingLevel === "xhigh"
+    ) {
+      sessionEntry.thinkingLevel = "high";
+      sessionEntry.updatedAt = Date.now();
+      sessionStore[sessionKey] = sessionEntry;
+      await saveSessionStore(storePath, sessionStore);
+    }
+  }
   const sessionFile = resolveSessionFilePath(sessionId, sessionEntry, {
     agentId: sessionAgentId,
   });

src/config/schema.ts

@@ -155,6 +155,8 @@ const FIELD_LABELS: Record<string, string> = {
   "agents.defaults.cliBackends": "CLI Backends",
   "commands.native": "Native Commands",
   "commands.text": "Text Commands",
+  "commands.bash": "Allow Bash Chat Command",
+  "commands.bashForegroundMs": "Bash Foreground Window (ms)",
   "commands.config": "Allow /config",
   "commands.debug": "Allow /debug",
   "commands.restart": "Allow Restart",
@@ -289,6 +291,10 @@ const FIELD_HELP: Record<string, string> = {
   "commands.native":
     "Register native commands with connectors that support it (Discord/Slack/Telegram).",
   "commands.text": "Allow text command parsing (slash commands only).",
+  "commands.bash":
+    "Allow bash chat command (`!`; `/bash` alias) to run host shell commands (default: false; requires tools.elevated).",
+  "commands.bashForegroundMs":
+    "How long bash waits before backgrounding (default: 2000; 0 backgrounds immediately).",
   "commands.config":
     "Allow /config chat command to read/write config on disk (default: false).",
   "commands.debug":

src/config/types.ts

@@ -1244,6 +1244,10 @@ export type CommandsConfig = {
   native?: NativeCommandsSetting;
   /** Enable text command parsing (default: true). */
   text?: boolean;
+  /** Allow bash chat command (`!`; `/bash` alias) (default: false). */
+  bash?: boolean;
+  /** How long bash waits before backgrounding (default: 2000; 0 backgrounds immediately). */
+  bashForegroundMs?: number;
   /** Allow /config command (default: false). */
   config?: boolean;
   /** Allow /debug command (default: false). */
@@ -1629,7 +1633,7 @@ export type AgentDefaultsConfig = {
   /** Vector memory search configuration (per-agent overrides supported). */
   memorySearch?: MemorySearchConfig;
   /** Default thinking level when no /think directive is present. */
-  thinkingDefault?: "off" | "minimal" | "low" | "medium" | "high";
+  thinkingDefault?: "off" | "minimal" | "low" | "medium" | "high" | "xhigh";
   /** Default verbose level when no /verbose directive is present. */
   verboseDefault?: "off" | "on";
   /** Default elevated level when no /elevated directive is present. */

src/config/zod-schema.ts

@@ -724,6 +724,8 @@ const CommandsSchema = z
   .object({
     native: NativeCommandsSettingSchema.optional().default("auto"),
     text: z.boolean().optional(),
+    bash: z.boolean().optional(),
+    bashForegroundMs: z.number().int().min(0).max(30_000).optional(),
     config: z.boolean().optional(),
     debug: z.boolean().optional(),
     restart: z.boolean().optional(),
@@ -1248,6 +1250,7 @@ const AgentDefaultsSchema = z
         z.literal("low"),
         z.literal("medium"),
         z.literal("high"),
+        z.literal("xhigh"),
       ])
       .optional(),
     verboseDefault: z.union([z.literal("off"), z.literal("on")]).optional(),

src/cron/isolated-agent.ts

@@ -32,7 +32,11 @@ import {
   DEFAULT_HEARTBEAT_ACK_MAX_CHARS,
   stripHeartbeatToken,
 } from "../auto-reply/heartbeat.js";
-import { normalizeThinkLevel } from "../auto-reply/thinking.js";
+import {
+  formatXHighModelHint,
+  normalizeThinkLevel,
+  supportsXHighThinking,
+} from "../auto-reply/thinking.js";
 import type { CliDeps } from "../cli/deps.js";
 import type { ClawdbotConfig } from "../config/config.js";
 import {
@@ -367,6 +371,11 @@ export async function runCronIsolatedAgentTurn(params: {
       catalog: await loadCatalog(),
     });
   }
+  if (thinkLevel === "xhigh" && !supportsXHighThinking(provider, model)) {
+    throw new Error(
+      `Thinking level "xhigh" is only supported for ${formatXHighModelHint()}.`,
+    );
+  }
 
   const timeoutMs = resolveAgentTimeoutMs({
     cfg: cfgWithAgentDefaults,

src/gateway/sessions-patch.ts

@@ -8,10 +8,13 @@ import {
 } from "../agents/model-selection.js";
 import { normalizeGroupActivation } from "../auto-reply/group-activation.js";
 import {
+  formatThinkingLevels,
+  formatXHighModelHint,
   normalizeElevatedLevel,
   normalizeReasoningLevel,
   normalizeThinkLevel,
   normalizeUsageDisplay,
+  supportsXHighThinking,
 } from "../auto-reply/thinking.js";
 import type { ClawdbotConfig } from "../config/config.js";
 import type { SessionEntry } from "../config/sessions.js";
@@ -95,8 +98,17 @@ export async function applySessionsPatchToStore(params: {
     } else if (raw !== undefined) {
       const normalized = normalizeThinkLevel(String(raw));
       if (!normalized) {
+        const resolvedDefault = resolveConfiguredModelRef({
+          cfg,
+          defaultProvider: DEFAULT_PROVIDER,
+          defaultModel: DEFAULT_MODEL,
+        });
+        const hintProvider =
+          existing?.providerOverride?.trim() || resolvedDefault.provider;
+        const hintModel =
+          existing?.modelOverride?.trim() || resolvedDefault.model;
         return invalid(
-          "invalid thinkingLevel (use off|minimal|low|medium|high)",
+          `invalid thinkingLevel (use ${formatThinkingLevels(hintProvider, hintModel, "|")})`,
         );
       }
       if (normalized === "off") delete next.thinkingLevel;
@@ -196,6 +208,24 @@ export async function applySessionsPatchToStore(params: {
     }
   }
 
+  if (next.thinkingLevel === "xhigh") {
+    const resolvedDefault = resolveConfiguredModelRef({
+      cfg,
+      defaultProvider: DEFAULT_PROVIDER,
+      defaultModel: DEFAULT_MODEL,
+    });
+    const effectiveProvider = next.providerOverride ?? resolvedDefault.provider;
+    const effectiveModel = next.modelOverride ?? resolvedDefault.model;
+    if (!supportsXHighThinking(effectiveProvider, effectiveModel)) {
+      if ("thinkingLevel" in patch) {
+        return invalid(
+          `thinkingLevel "xhigh" is only supported for ${formatXHighModelHint()}`,
+        );
+      }
+      next.thinkingLevel = "high";
+    }
+  }
+
   if ("sendPolicy" in patch) {
     const raw = patch.sendPolicy;
     if (raw === null) {

src/providers/google-shared.test.ts

@@ -171,7 +171,7 @@ describe("google-shared convertMessages", () => {
             {
               type: "thinking",
               thinking: "hidden",
-              thinkingSignature: "sig",
+              thinkingSignature: "c2ln",
             },
           ],
           api: "google-generative-ai",
@@ -202,7 +202,7 @@ describe("google-shared convertMessages", () => {
     expect(contents[0].role).toBe("model");
     expect(contents[0].parts?.[0]).toMatchObject({
       thought: true,
-      thoughtSignature: "sig",
+      thoughtSignature: "c2ln",
     });
   });
 
@@ -216,7 +216,7 @@ describe("google-shared convertMessages", () => {
             {
               type: "thinking",
               thinking: "structured",
-              thinkingSignature: "sig",
+              thinkingSignature: "c2ln",
             },
           ],
           api: "google-generative-ai",
@@ -247,7 +247,7 @@ describe("google-shared convertMessages", () => {
     expect(parts).toHaveLength(1);
     expect(parts[0]).toMatchObject({
       thought: true,
-      thoughtSignature: "sig",
+      thoughtSignature: "c2ln",
     });
   });
 

src/tui/commands.ts

@@ -1,6 +1,9 @@
 import type { SlashCommand } from "@mariozechner/pi-tui";
+import {
+  formatThinkingLevels,
+  listThinkingLevels,
+} from "../auto-reply/thinking.js";
 
-const THINK_LEVELS = ["off", "minimal", "low", "medium", "high"];
 const VERBOSE_LEVELS = ["on", "off"];
 const REASONING_LEVELS = ["on", "off"];
 const ELEVATED_LEVELS = ["on", "off"];
@@ -12,6 +15,11 @@ export type ParsedCommand = {
   args: string;
 };
 
+export type SlashCommandOptions = {
+  provider?: string;
+  model?: string;
+};
+
 const COMMAND_ALIASES: Record<string, string> = {
   elev: "elevated",
 };
@@ -27,7 +35,10 @@ export function parseCommand(input: string): ParsedCommand {
   };
 }
 
-export function getSlashCommands(): SlashCommand[] {
+export function getSlashCommands(
+  options: SlashCommandOptions = {},
+): SlashCommand[] {
+  const thinkLevels = listThinkingLevels(options.provider, options.model);
   return [
     { name: "help", description: "Show slash command help" },
     { name: "status", description: "Show gateway status summary" },
@@ -44,9 +55,9 @@ export function getSlashCommands(): SlashCommand[] {
       name: "think",
       description: "Set thinking level",
       getArgumentCompletions: (prefix) =>
-        THINK_LEVELS.filter((v) => v.startsWith(prefix.toLowerCase())).map(
+        thinkLevels
-          (value) => ({ value, label: value }),
+          .filter((v) => v.startsWith(prefix.toLowerCase()))
-        ),
+          .map((value) => ({ value, label: value })),
     },
     {
       name: "verbose",
@@ -105,7 +116,12 @@ export function getSlashCommands(): SlashCommand[] {
   ];
 }
 
-export function helpText(): string {
+export function helpText(options: SlashCommandOptions = {}): string {
+  const thinkLevels = formatThinkingLevels(
+    options.provider,
+    options.model,
+    "|",
+  );
   return [
     "Slash commands:",
     "/help",
@@ -113,7 +129,7 @@ export function helpText(): string {
     "/agent <id> (or /agents)",
     "/session <key> (or /sessions)",
     "/model <provider/model> (or /models)",
-    "/think <off|minimal|low|medium|high>",
+    `/think <${thinkLevels}>`,
     "/verbose <on|off>",
     "/reasoning <on|off>",
     "/cost <on|off>",

src/tui/tui.ts

@@ -7,7 +7,10 @@ import {
   TUI,
 } from "@mariozechner/pi-tui";
 import { resolveDefaultAgentId } from "../agents/agent-scope.js";
-import { normalizeUsageDisplay } from "../auto-reply/thinking.js";
+import {
+  formatThinkingLevels,
+  normalizeUsageDisplay,
+} from "../auto-reply/thinking.js";
 import { loadConfig } from "../config/config.js";
 import { formatAge } from "../infra/provider-summary.js";
 import {
@@ -239,6 +242,18 @@ export async function runTui(opts: TuiOptions) {
   root.addChild(footer);
   root.addChild(editor);
 
+  const updateAutocompleteProvider = () => {
+    editor.setAutocompleteProvider(
+      new CombinedAutocompleteProvider(
+        getSlashCommands({
+          provider: sessionInfo.modelProvider,
+          model: sessionInfo.model,
+        }),
+        process.cwd(),
+      ),
+    );
+  };
+
   const tui = new TUI(new ProcessTerminal());
   tui.addChild(root);
   tui.setFocus(editor);
@@ -524,6 +539,7 @@ export async function runTui(opts: TuiOptions) {
     } catch (err) {
       chatLog.addSystem(`sessions list failed: ${String(err)}`);
     }
+    updateAutocompleteProvider();
     updateFooter();
     tui.requestRender();
   };
@@ -861,7 +877,12 @@ export async function runTui(opts: TuiOptions) {
     if (!name) return;
     switch (name) {
       case "help":
-        chatLog.addSystem(helpText());
+        chatLog.addSystem(
+          helpText({
+            provider: sessionInfo.modelProvider,
+            model: sessionInfo.model,
+          }),
+        );
         break;
       case "status":
         try {
@@ -921,7 +942,12 @@ export async function runTui(opts: TuiOptions) {
         break;
       case "think":
         if (!args) {
-          chatLog.addSystem("usage: /think <off|minimal|low|medium|high>");
+          const levels = formatThinkingLevels(
+            sessionInfo.modelProvider,
+            sessionInfo.model,
+            "|",
+          );
+          chatLog.addSystem(`usage: /think <${levels}>`);
           break;
         }
         try {
@@ -1071,9 +1097,7 @@ export async function runTui(opts: TuiOptions) {
     tui.requestRender();
   };
 
-  editor.setAutocompleteProvider(
+  updateAutocompleteProvider();
-    new CombinedAutocompleteProvider(getSlashCommands(), process.cwd()),
-  );
   editor.onSubmit = (text) => {
     const value = text.trim();
     editor.setText("");

test/gateway.multi.e2e.test.ts

@@ -122,9 +122,9 @@ const spawnGatewayInstance = async (name: string): Promise<GatewayInstance> => {
 
   try {
     child = spawn(
-      "bun",
+      "node",
       [
-        "src/index.ts",
+        "dist/index.js",
         "gateway",
         "--port",
         String(port),
@@ -222,7 +222,7 @@ const runCliJson = async (
 ): Promise<unknown> => {
   const stdout: string[] = [];
   const stderr: string[] = [];
-  const child = spawn("bun", ["src/index.ts", ...args], {
+  const child = spawn("node", ["dist/index.js", ...args], {
     cwd: process.cwd(),
     env: { ...process.env, ...env },
     stdio: ["ignore", "pipe", "pipe"],

ui/package.json

@@ -15,10 +15,10 @@
     "vite": "7.3.1"
   },
   "devDependencies": {
-    "@vitest/browser-playwright": "4.0.16",
+    "@vitest/browser-playwright": "4.0.17",
     "playwright": "^1.57.0",
     "typescript": "^5.9.3",
-    "vitest": "4.0.16"
+    "vitest": "4.0.17"
   },
   "pnpm": {
     "minimumReleaseAge": 2880