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

AGENTS.md

@@ -100,6 +100,7 @@
 - **Multi-agent safety:** focus reports on your edits; avoid guard-rail disclaimers unless truly blocked; when multiple agents touch the same file, continue if safe; end with a brief “other files present” note only if relevant.
 - Bug investigations: read source code of relevant npm dependencies and all related local code before concluding; aim for high-confidence root cause.
 - Code style: add brief comments for tricky logic; keep files under ~500 LOC when feasible (split/refactor as needed).
+- Tool schema guardrails (google-antigravity): avoid `Type.Union` in tool input schemas; no `anyOf`/`oneOf`/`allOf`. Use `stringEnum`/`optionalStringEnum` (Type.Unsafe enum) for string lists, and `Type.Optional(...)` instead of `... | null`. Keep top-level tool schema as `type: "object"` with `properties`.
 - When asked to open a “session” file, open the Pi session logs under `~/.clawdbot/agents/main/sessions/*.jsonl` (newest unless a specific ID is given), not the default `sessions.json`. If logs are needed from another machine, SSH via Tailscale and read the same path there.
 - Menubar dimming + restart flow mirrors Trimmy: use `scripts/restart-mac.sh` (kills all Clawdbot variants, runs `swift build`, packages, relaunches). Icon dimming depends on MenuBarExtraAccess wiring in AppMain; keep `appearsDisabled` updates intact when touching the status item.
 - Do not rebuild the macOS app over SSH; rebuilds must be run directly on the Mac.

CHANGELOG.md

@@ -1,116 +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)
-
-### Fixes
-- Gateway: honor `CLAWDBOT_LAUNCHD_LABEL` / `CLAWDBOT_SYSTEMD_UNIT` overrides when checking or restarting the daemon.
-
-## 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

@@ -569,7 +569,9 @@ Inbound messages are routed to an agent via bindings.
   - `name`: display name for the agent.
   - `workspace`: default `~/clawd-<agentId>` (for `main`, falls back to `agents.defaults.workspace`).
   - `agentDir`: default `~/.clawdbot/agents/<agentId>/agent`.
-  - `model`: per-agent default model (provider/model), overrides `agents.defaults.model` for that agent.
+  - `model`: per-agent default model, overrides `agents.defaults.model` for that agent.
+    - string form: `"provider/model"`, overrides only `agents.defaults.model.primary`
+    - object form: `{ primary, fallbacks }` (fallbacks override `agents.defaults.model.fallbacks`; `[]` disables global fallbacks for that agent)
   - `identity`: per-agent name/theme/emoji (used for mention patterns + ack reactions).
   - `groupChat`: per-agent mention-gating (`mentionPatterns`).
   - `sandbox`: per-agent sandbox config (overrides `agents.defaults.sandbox`).
@@ -583,6 +585,7 @@ Inbound messages are routed to an agent via bindings.
   - `subagents`: per-agent sub-agent defaults.
     - `allowAgents`: allowlist of agent ids for `sessions_spawn` from this agent (`["*"]` = allow any; default: only same agent)
   - `tools`: per-agent tool restrictions (applied before sandbox tool policy).
+    - `profile`: base tool profile (applied before allow/deny)
     - `allow`: array of allowed tool names
     - `deny`: array of denied tool names (deny wins)
 - `agents.defaults`: shared agent defaults (model, workspace, sandbox, etc.).
@@ -1507,6 +1510,34 @@ Legacy: `tools.bash` is still accepted as an alias.
 - `archiveAfterMinutes`: auto-archive sub-agent sessions after N minutes (default 60; set `0` to disable)
 - Per-subagent tool policy: `tools.subagents.tools.allow` / `tools.subagents.tools.deny` (deny wins)
 
+`tools.profile` sets a **base tool allowlist** before `tools.allow`/`tools.deny`:
+- `minimal`: `session_status` only
+- `coding`: `group:fs`, `group:runtime`, `group:sessions`, `group:memory`, `image`
+- `messaging`: `group:messaging`, `sessions_list`, `sessions_history`, `sessions_send`, `session_status`
+- `full`: no restriction (same as unset)
+
+Per-agent override: `agents.list[].tools.profile`.
+
+Example (messaging-only by default, allow Slack + Discord tools too):
+```json5
+{
+  tools: {
+    profile: "messaging",
+    allow: ["slack", "discord"]
+  }
+}
+```
+
+Example (coding profile, but deny exec/process everywhere):
+```json5
+{
+  tools: {
+    profile: "coding",
+    deny: ["group:runtime"]
+  }
+}
+```
+
 `tools.allow` / `tools.deny` configure a global tool allow/deny policy (deny wins).
 This is applied even when the Docker sandbox is **off**.
 
@@ -1517,6 +1548,17 @@ Example (disable browser/canvas everywhere):
 }
 ```
 
+Tool groups (shorthands) work in **global** and **per-agent** tool policies:
+- `group:runtime`: `exec`, `bash`, `process`
+- `group:fs`: `read`, `write`, `edit`, `apply_patch`
+- `group:sessions`: `sessions_list`, `sessions_history`, `sessions_send`, `sessions_spawn`, `session_status`
+- `group:memory`: `memory_search`, `memory_get`
+- `group:ui`: `browser`, `canvas`
+- `group:automation`: `cron`, `gateway`
+- `group:messaging`: `message`
+- `group:nodes`: `nodes`
+- `group:clawdbot`: all built-in Clawdbot tools (excludes provider plugins)
+
 `tools.elevated` controls elevated (host) exec access:
 - `enabled`: allow elevated mode (default true)
 - `allowFrom`: per-provider allowlists (empty = disabled)

docs/gateway/sandbox-vs-tool-policy-vs-elevated.md

@@ -50,6 +50,7 @@ See [Sandboxing](/gateway/sandboxing) for the full matrix (scope, workspace moun
 ## Tool policy: which tools exist/are callable
 
 Two layers matter:
+- **Tool profile**: `tools.profile` and `agents.list[].tools.profile` (base allowlist)
 - **Global/per-agent tool policy**: `tools.allow`/`tools.deny` and `agents.list[].tools.allow`/`agents.list[].tools.deny`
 - **Sandbox tool policy** (only applies when sandboxed): `tools.sandbox.tools.allow`/`tools.sandbox.tools.deny` and `agents.list[].tools.sandbox.tools.*`
 
@@ -59,7 +60,7 @@ Rules of thumb:
 
 ### Tool groups (shorthands)
 
-For sandbox tool policy, you can use `group:*` entries that expand to multiple tools:
+Tool policies (global, agent, sandbox) support `group:*` entries that expand to multiple tools:
 
 ```json5
 {
@@ -78,6 +79,11 @@ Available groups:
 - `group:fs`: `read`, `write`, `edit`, `apply_patch`
 - `group:sessions`: `sessions_list`, `sessions_history`, `sessions_send`, `sessions_spawn`, `session_status`
 - `group:memory`: `memory_search`, `memory_get`
+- `group:ui`: `browser`, `canvas`
+- `group:automation`: `cron`, `gateway`
+- `group:messaging`: `message`
+- `group:nodes`: `nodes`
+- `group:clawdbot`: all built-in Clawdbot tools (excludes provider plugins)
 
 ## Elevated: exec-only “run on host”
 

docs/multi-agent-sandbox-tools.md

@@ -106,6 +106,28 @@ For debugging “why is this blocked?”, see [Sandbox vs Tool Policy vs Elevate
 
 ---
 
+### Example 2b: Global coding profile + messaging-only agent
+
+```json
+{
+  "tools": { "profile": "coding" },
+  "agents": {
+    "list": [
+      {
+        "id": "support",
+        "tools": { "profile": "messaging", "allow": ["slack"] }
+      }
+    ]
+  }
+}
+```
+
+**Result:**
+- default agents get coding tools
+- `support` agent is messaging-only (+ Slack tool)
+
+---
+
 ### Example 3: Different Sandbox Modes per Agent
 
 ```json
@@ -165,22 +187,29 @@ agents.list[].sandbox.prune.* > agents.defaults.sandbox.prune.*
 
 ### Tool Restrictions
 The filtering order is:
-1. **Global tool policy** (`tools.allow` / `tools.deny`)
+1. **Tool profile** (`tools.profile` or `agents.list[].tools.profile`)
-2. **Agent-specific tool policy** (`agents.list[].tools`)
+2. **Global tool policy** (`tools.allow` / `tools.deny`)
-3. **Sandbox tool policy** (`tools.sandbox.tools` or `agents.list[].tools.sandbox.tools`)
+3. **Agent-specific tool policy** (`agents.list[].tools`)
-4. **Subagent tool policy** (`tools.subagents.tools`, if applicable)
+4. **Sandbox tool policy** (`tools.sandbox.tools` or `agents.list[].tools.sandbox.tools`)
+5. **Subagent tool policy** (`tools.subagents.tools`, if applicable)
 
 Each level can further restrict tools, but cannot grant back denied tools from earlier levels.
 If `agents.list[].tools.sandbox.tools` is set, it replaces `tools.sandbox.tools` for that agent.
+If `agents.list[].tools.profile` is set, it overrides `tools.profile` for that agent.
 
 ### Tool groups (shorthands)
 
-Sandbox tool policy supports `group:*` entries that expand to multiple concrete tools:
+Tool policies (global, agent, sandbox) support `group:*` entries that expand to multiple concrete tools:
 
 - `group:runtime`: `exec`, `bash`, `process`
 - `group:fs`: `read`, `write`, `edit`, `apply_patch`
 - `group:sessions`: `sessions_list`, `sessions_history`, `sessions_send`, `sessions_spawn`, `session_status`
 - `group:memory`: `memory_search`, `memory_get`
+- `group:ui`: `browser`, `canvas`
+- `group:automation`: `cron`, `gateway`
+- `group:messaging`: `message`
+- `group:nodes`: `nodes`
+- `group:clawdbot`: all built-in Clawdbot tools (excludes provider plugins)
 
 ### Elevated Mode
 `tools.elevated` is the global baseline (sender-based allowlist). `agents.list[].tools.elevated` can further restrict elevated for specific agents (both must allow).

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

@@ -12,6 +12,8 @@ read_when:
 - 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/index.md

@@ -22,6 +22,77 @@ You can globally allow/deny tools via `tools.allow` / `tools.deny` in `clawdbot.
 }
 ```
 
+## Tool profiles (base allowlist)
+
+`tools.profile` sets a **base tool allowlist** before `tools.allow`/`tools.deny`.
+Per-agent override: `agents.list[].tools.profile`.
+
+Profiles:
+- `minimal`: `session_status` only
+- `coding`: `group:fs`, `group:runtime`, `group:sessions`, `group:memory`, `image`
+- `messaging`: `group:messaging`, `sessions_list`, `sessions_history`, `sessions_send`, `session_status`
+- `full`: no restriction (same as unset)
+
+Example (messaging-only by default, allow Slack + Discord tools too):
+```json5
+{
+  tools: {
+    profile: "messaging",
+    allow: ["slack", "discord"]
+  }
+}
+```
+
+Example (coding profile, but deny exec/process everywhere):
+```json5
+{
+  tools: {
+    profile: "coding",
+    deny: ["group:runtime"]
+  }
+}
+```
+
+Example (global coding profile, messaging-only support agent):
+```json5
+{
+  tools: { profile: "coding" },
+  agents: {
+    list: [
+      {
+        id: "support",
+        tools: { profile: "messaging", allow: ["slack"] }
+      }
+    ]
+  }
+}
+```
+
+## Tool groups (shorthands)
+
+Tool policies (global, agent, sandbox) support `group:*` entries that expand to multiple tools.
+Use these in `tools.allow` / `tools.deny`.
+
+Available groups:
+- `group:runtime`: `exec`, `bash`, `process`
+- `group:fs`: `read`, `write`, `edit`, `apply_patch`
+- `group:sessions`: `sessions_list`, `sessions_history`, `sessions_send`, `sessions_spawn`, `session_status`
+- `group:memory`: `memory_search`, `memory_get`
+- `group:ui`: `browser`, `canvas`
+- `group:automation`: `cron`, `gateway`
+- `group:messaging`: `message`
+- `group:nodes`: `nodes`
+- `group:clawdbot`: all built-in Clawdbot tools (excludes provider plugins)
+
+Example (allow only file tools + browser):
+```json5
+{
+  tools: {
+    allow: ["group:fs", "browser"]
+  }
+}
+```
+
 ## Plugins + tools
 
 Plugins can register **additional tools** (and CLI commands) beyond the core set.

docs/tools/slash-commands.md

@@ -44,6 +44,7 @@ They run immediately, are stripped before the model sees the message, and the re
   - 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).
@@ -65,7 +66,7 @@ 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`)
@@ -90,6 +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/agent-scope.test.ts

@@ -1,6 +1,10 @@
 import { describe, expect, it } from "vitest";
 import type { ClawdbotConfig } from "../config/config.js";
-import { resolveAgentConfig } from "./agent-scope.js";
+import {
+  resolveAgentConfig,
+  resolveAgentModelFallbacksOverride,
+  resolveAgentModelPrimary,
+} from "./agent-scope.js";
 
 describe("resolveAgentConfig", () => {
   it("should return undefined when no agents config exists", () => {
@@ -47,6 +51,68 @@ describe("resolveAgentConfig", () => {
     });
   });
 
+  it("supports per-agent model primary+fallbacks", () => {
+    const cfg: ClawdbotConfig = {
+      agents: {
+        defaults: {
+          model: {
+            primary: "anthropic/claude-sonnet-4",
+            fallbacks: ["openai/gpt-4.1"],
+          },
+        },
+        list: [
+          {
+            id: "linus",
+            model: {
+              primary: "anthropic/claude-opus-4",
+              fallbacks: ["openai/gpt-5.2"],
+            },
+          },
+        ],
+      },
+    };
+
+    expect(resolveAgentModelPrimary(cfg, "linus")).toBe(
+      "anthropic/claude-opus-4",
+    );
+    expect(resolveAgentModelFallbacksOverride(cfg, "linus")).toEqual([
+      "openai/gpt-5.2",
+    ]);
+
+    // If fallbacks isn't present, we don't override the global fallbacks.
+    const cfgNoOverride: ClawdbotConfig = {
+      agents: {
+        list: [
+          {
+            id: "linus",
+            model: {
+              primary: "anthropic/claude-opus-4",
+            },
+          },
+        ],
+      },
+    };
+    expect(resolveAgentModelFallbacksOverride(cfgNoOverride, "linus")).toBe(
+      undefined,
+    );
+
+    // Explicit empty list disables global fallbacks for that agent.
+    const cfgDisable: ClawdbotConfig = {
+      agents: {
+        list: [
+          {
+            id: "linus",
+            model: {
+              primary: "anthropic/claude-opus-4",
+              fallbacks: [],
+            },
+          },
+        ],
+      },
+    };
+    expect(resolveAgentModelFallbacksOverride(cfgDisable, "linus")).toEqual([]);
+  });
+
   it("should return agent-specific sandbox config", () => {
     const cfg: ClawdbotConfig = {
       agents: {

src/agents/agent-scope.ts

@@ -21,7 +21,7 @@ type ResolvedAgentConfig = {
   name?: string;
   workspace?: string;
   agentDir?: string;
-  model?: string;
+  model?: AgentEntry["model"];
   memorySearch?: AgentEntry["memorySearch"];
   humanDelay?: AgentEntry["humanDelay"];
   identity?: AgentEntry["identity"];
@@ -95,7 +95,11 @@ export function resolveAgentConfig(
     workspace:
       typeof entry.workspace === "string" ? entry.workspace : undefined,
     agentDir: typeof entry.agentDir === "string" ? entry.agentDir : undefined,
-    model: typeof entry.model === "string" ? entry.model : undefined,
+    model:
+      typeof entry.model === "string" ||
+      (entry.model && typeof entry.model === "object")
+        ? entry.model
+        : undefined,
     memorySearch: entry.memorySearch,
     humanDelay: entry.humanDelay,
     identity: entry.identity,
@@ -109,6 +113,28 @@ export function resolveAgentConfig(
   };
 }
 
+export function resolveAgentModelPrimary(
+  cfg: ClawdbotConfig,
+  agentId: string,
+): string | undefined {
+  const raw = resolveAgentConfig(cfg, agentId)?.model;
+  if (!raw) return undefined;
+  if (typeof raw === "string") return raw.trim() || undefined;
+  const primary = raw.primary?.trim();
+  return primary || undefined;
+}
+
+export function resolveAgentModelFallbacksOverride(
+  cfg: ClawdbotConfig,
+  agentId: string,
+): string[] | undefined {
+  const raw = resolveAgentConfig(cfg, agentId)?.model;
+  if (!raw || typeof raw === "string") return undefined;
+  // Important: treat an explicitly provided empty array as an override to disable global fallbacks.
+  if (!Object.hasOwn(raw, "fallbacks")) return undefined;
+  return Array.isArray(raw.fallbacks) ? raw.fallbacks : undefined;
+}
+
 export function resolveAgentWorkspaceDir(cfg: ClawdbotConfig, agentId: string) {
   const id = normalizeAgentId(agentId);
   const configured = resolveAgentConfig(cfg, id)?.workspace?.trim();

src/agents/bash-tools.ts

@@ -40,20 +40,6 @@ const DEFAULT_PATH =
   process.env.PATH ??
   "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin";
 
-// NOTE: Using Type.Unsafe with enum instead of Type.Union([Type.Literal(...)])
-// because Claude API on Vertex AI rejects nested anyOf schemas as invalid JSON Schema.
-// Type.Union of literals compiles to { anyOf: [{enum:["a"]}, {enum:["b"]}, ...] }
-// which is valid but not accepted. A flat enum { type: "string", enum: [...] } works.
-const _stringEnum = <T extends readonly string[]>(
-  values: T,
-  options?: { description?: string },
-) =>
-  Type.Unsafe<T[number]>({
-    type: "string",
-    enum: values as unknown as string[],
-    ...options,
-  });
-
 export type ExecToolDefaults = {
   backgroundMs?: number;
   timeoutSec?: number;

src/agents/model-fallback.test.ts

@@ -124,6 +124,106 @@ describe("runWithModelFallback", () => {
     expect(run.mock.calls[1]?.[1]).toBe("claude-haiku-3-5");
   });
 
+  it("does not append configured primary when fallbacksOverride is set", async () => {
+    const cfg = makeCfg({
+      agents: {
+        defaults: {
+          model: {
+            primary: "openai/gpt-4.1-mini",
+          },
+        },
+      },
+    });
+    const run = vi
+      .fn()
+      .mockImplementation(() =>
+        Promise.reject(Object.assign(new Error("nope"), { status: 401 })),
+      );
+
+    await expect(
+      runWithModelFallback({
+        cfg,
+        provider: "anthropic",
+        model: "claude-opus-4-5",
+        fallbacksOverride: ["anthropic/claude-haiku-3-5"],
+        run,
+      }),
+    ).rejects.toThrow("All models failed");
+
+    expect(run.mock.calls).toEqual([
+      ["anthropic", "claude-opus-4-5"],
+      ["anthropic", "claude-haiku-3-5"],
+    ]);
+  });
+
+  it("uses fallbacksOverride instead of agents.defaults.model.fallbacks", async () => {
+    const cfg = {
+      agents: {
+        defaults: {
+          model: {
+            fallbacks: ["openai/gpt-5.2"],
+          },
+        },
+      },
+    } as ClawdbotConfig;
+
+    const calls: Array<{ provider: string; model: string }> = [];
+
+    const res = await runWithModelFallback({
+      cfg,
+      provider: "anthropic",
+      model: "claude-opus-4-5",
+      fallbacksOverride: ["openai/gpt-4.1"],
+      run: async (provider, model) => {
+        calls.push({ provider, model });
+        if (provider === "anthropic") {
+          throw Object.assign(new Error("nope"), { status: 401 });
+        }
+        if (provider === "openai" && model === "gpt-4.1") {
+          return "ok";
+        }
+        throw new Error(`unexpected candidate: ${provider}/${model}`);
+      },
+    });
+
+    expect(res.result).toBe("ok");
+    expect(calls).toEqual([
+      { provider: "anthropic", model: "claude-opus-4-5" },
+      { provider: "openai", model: "gpt-4.1" },
+    ]);
+  });
+
+  it("treats an empty fallbacksOverride as disabling global fallbacks", async () => {
+    const cfg = {
+      agents: {
+        defaults: {
+          model: {
+            fallbacks: ["openai/gpt-5.2"],
+          },
+        },
+      },
+    } as ClawdbotConfig;
+
+    const calls: Array<{ provider: string; model: string }> = [];
+
+    await expect(
+      runWithModelFallback({
+        cfg,
+        provider: "anthropic",
+        model: "claude-opus-4-5",
+        fallbacksOverride: [],
+        run: async (provider, model) => {
+          calls.push({ provider, model });
+          throw new Error("primary failed");
+        },
+      }),
+    ).rejects.toThrow("primary failed");
+
+    expect(calls).toEqual([
+      { provider: "anthropic", model: "claude-opus-4-5" },
+    ]);
+  });
+
   it("falls back on missing API key errors", async () => {
     const cfg = makeCfg();
     const run = vi

src/agents/model-fallback.ts

@@ -126,6 +126,8 @@ function resolveFallbackCandidates(params: {
   cfg: ClawdbotConfig | undefined;
   provider: string;
   model: string;
+  /** Optional explicit fallbacks list; when provided (even empty), replaces agents.defaults.model.fallbacks. */
+  fallbacksOverride?: string[];
 }): ModelCandidate[] {
   const provider = params.provider.trim() || DEFAULT_PROVIDER;
   const model = params.model.trim() || DEFAULT_MODEL;
@@ -159,6 +161,7 @@ function resolveFallbackCandidates(params: {
   addCandidate({ provider, model }, false);
 
   const modelFallbacks = (() => {
+    if (params.fallbacksOverride !== undefined) return params.fallbacksOverride;
     const model = params.cfg?.agents?.defaults?.model as
       | { fallbacks?: string[] }
       | string
@@ -177,7 +180,11 @@ function resolveFallbackCandidates(params: {
     addCandidate(resolved.ref, true);
   }
 
-  if (primary?.provider && primary.model) {
+  if (
+    params.fallbacksOverride === undefined &&
+    primary?.provider &&
+    primary.model
+  ) {
     addCandidate({ provider: primary.provider, model: primary.model }, false);
   }
 
@@ -188,6 +195,8 @@ export async function runWithModelFallback<T>(params: {
   cfg: ClawdbotConfig | undefined;
   provider: string;
   model: string;
+  /** Optional explicit fallbacks list; when provided (even empty), replaces agents.defaults.model.fallbacks. */
+  fallbacksOverride?: string[];
   run: (provider: string, model: string) => Promise<T>;
   onError?: (attempt: {
     provider: string;
@@ -202,7 +211,12 @@ export async function runWithModelFallback<T>(params: {
   model: string;
   attempts: FallbackAttempt[];
 }> {
-  const candidates = resolveFallbackCandidates(params);
+  const candidates = resolveFallbackCandidates({
+    cfg: params.cfg,
+    provider: params.provider,
+    model: params.model,
+    fallbacksOverride: params.fallbacksOverride,
+  });
   const attempts: FallbackAttempt[] = [];
   let lastError: unknown;
 

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/agents/pi-tools.test.ts

@@ -5,6 +5,7 @@ import type { AgentTool } from "@mariozechner/pi-agent-core";
 import sharp from "sharp";
 import { describe, expect, it, vi } from "vitest";
 import type { ClawdbotConfig } from "../config/config.js";
+import { createClawdbotTools } from "./clawdbot-tools.js";
 import { __testing, createClawdbotCodingTools } from "./pi-tools.js";
 import { createBrowserTool } from "./tools/browser-tool.js";
 
@@ -16,7 +17,7 @@ describe("createClawdbotCodingTools", () => {
     expect(schema.anyOf).toBeUndefined();
   });
 
-  it("merges properties for union tool schemas", () => {
+  it("keeps browser tool schema properties after normalization", () => {
     const tools = createClawdbotCodingTools();
     const browser = tools.find((tool) => tool.name === "browser");
     expect(browser).toBeDefined();
@@ -266,6 +267,95 @@ describe("createClawdbotCodingTools", () => {
     expect(offenders).toEqual([]);
   });
 
+  it("avoids anyOf/oneOf/allOf in tool schemas", () => {
+    const tools = createClawdbotCodingTools();
+    const offenders: Array<{
+      name: string;
+      keyword: string;
+      path: string;
+    }> = [];
+    const keywords = new Set(["anyOf", "oneOf", "allOf"]);
+
+    const walk = (value: unknown, path: string, name: string): void => {
+      if (!value) return;
+      if (Array.isArray(value)) {
+        for (const [index, entry] of value.entries()) {
+          walk(entry, `${path}[${index}]`, name);
+        }
+        return;
+      }
+      if (typeof value !== "object") return;
+
+      const record = value as Record<string, unknown>;
+      for (const [key, entry] of Object.entries(record)) {
+        const nextPath = path ? `${path}.${key}` : key;
+        if (keywords.has(key)) {
+          offenders.push({ name, keyword: key, path: nextPath });
+        }
+        walk(entry, nextPath, name);
+      }
+    };
+
+    for (const tool of tools) {
+      walk(tool.parameters, "", tool.name);
+    }
+
+    expect(offenders).toEqual([]);
+  });
+
+  it("keeps raw core tool schemas union-free", () => {
+    const tools = createClawdbotTools();
+    const coreTools = new Set([
+      "browser",
+      "canvas",
+      "nodes",
+      "cron",
+      "message",
+      "gateway",
+      "agents_list",
+      "sessions_list",
+      "sessions_history",
+      "sessions_send",
+      "sessions_spawn",
+      "session_status",
+      "memory_search",
+      "memory_get",
+      "image",
+    ]);
+    const offenders: Array<{
+      name: string;
+      keyword: string;
+      path: string;
+    }> = [];
+    const keywords = new Set(["anyOf", "oneOf", "allOf"]);
+
+    const walk = (value: unknown, path: string, name: string): void => {
+      if (!value) return;
+      if (Array.isArray(value)) {
+        for (const [index, entry] of value.entries()) {
+          walk(entry, `${path}[${index}]`, name);
+        }
+        return;
+      }
+      if (typeof value !== "object") return;
+      const record = value as Record<string, unknown>;
+      for (const [key, entry] of Object.entries(record)) {
+        const nextPath = path ? `${path}.${key}` : key;
+        if (keywords.has(key)) {
+          offenders.push({ name, keyword: key, path: nextPath });
+        }
+        walk(entry, nextPath, name);
+      }
+    };
+
+    for (const tool of tools) {
+      if (!coreTools.has(tool.name)) continue;
+      walk(tool.parameters, "", tool.name);
+    }
+
+    expect(offenders).toEqual([]);
+  });
+
   it("does not expose provider-specific message tools", () => {
     const tools = createClawdbotCodingTools({ messageProvider: "discord" });
     const names = new Set(tools.map((tool) => tool.name));
@@ -517,6 +607,57 @@ describe("createClawdbotCodingTools", () => {
     expect(tools.some((tool) => tool.name === "browser")).toBe(false);
   });
 
+  it("applies tool profiles before allow/deny policies", () => {
+    const tools = createClawdbotCodingTools({
+      config: { tools: { profile: "messaging" } },
+    });
+    const names = new Set(tools.map((tool) => tool.name));
+    expect(names.has("message")).toBe(true);
+    expect(names.has("sessions_send")).toBe(true);
+    expect(names.has("sessions_spawn")).toBe(false);
+    expect(names.has("exec")).toBe(false);
+    expect(names.has("browser")).toBe(false);
+  });
+
+  it("expands group shorthands in global tool policy", () => {
+    const tools = createClawdbotCodingTools({
+      config: { tools: { allow: ["group:fs"] } },
+    });
+    const names = new Set(tools.map((tool) => tool.name));
+    expect(names.has("read")).toBe(true);
+    expect(names.has("write")).toBe(true);
+    expect(names.has("edit")).toBe(true);
+    expect(names.has("exec")).toBe(false);
+    expect(names.has("browser")).toBe(false);
+  });
+
+  it("expands group shorthands in global tool deny policy", () => {
+    const tools = createClawdbotCodingTools({
+      config: { tools: { deny: ["group:fs"] } },
+    });
+    const names = new Set(tools.map((tool) => tool.name));
+    expect(names.has("read")).toBe(false);
+    expect(names.has("write")).toBe(false);
+    expect(names.has("edit")).toBe(false);
+    expect(names.has("exec")).toBe(true);
+  });
+
+  it("lets agent profiles override global profiles", () => {
+    const tools = createClawdbotCodingTools({
+      sessionKey: "agent:work:main",
+      config: {
+        tools: { profile: "coding" },
+        agents: {
+          list: [{ id: "work", tools: { profile: "messaging" } }],
+        },
+      },
+    });
+    const names = new Set(tools.map((tool) => tool.name));
+    expect(names.has("message")).toBe(true);
+    expect(names.has("exec")).toBe(false);
+    expect(names.has("read")).toBe(false);
+  });
+
   it("removes unsupported JSON Schema keywords for Cloud Code Assist API compatibility", () => {
     const tools = createClawdbotCodingTools();
 

src/agents/pi-tools.ts

@@ -28,6 +28,11 @@ import type { SandboxContext, SandboxToolPolicy } from "./sandbox.js";
 import { assertSandboxPath } from "./sandbox-paths.js";
 import { cleanSchemaForGemini } from "./schema/clean-for-gemini.js";
 import { sanitizeToolResultImages } from "./tool-images.js";
+import {
+  expandToolGroups,
+  normalizeToolName,
+  resolveToolProfilePolicy,
+} from "./tool-policy.js";
 
 // NOTE(steipete): Upstream read now does file-magic MIME detection; we keep the wrapper
 // to normalize payloads and sanitize oversized images before they hit providers.
@@ -291,21 +296,6 @@ function cleanToolSchemaForGemini(schema: Record<string, unknown>): unknown {
   return cleanSchemaForGemini(schema);
 }
 
-const TOOL_NAME_ALIASES: Record<string, string> = {
-  bash: "exec",
-  "apply-patch": "apply_patch",
-};
-
-function normalizeToolName(name: string) {
-  const normalized = name.trim().toLowerCase();
-  return TOOL_NAME_ALIASES[normalized] ?? normalized;
-}
-
-function normalizeToolNames(list?: string[]) {
-  if (!list) return [];
-  return list.map(normalizeToolName).filter(Boolean);
-}
-
 function isOpenAIProvider(provider?: string) {
   const normalized = provider?.trim().toLowerCase();
   return normalized === "openai" || normalized === "openai-codex";
@@ -357,8 +347,8 @@ function isToolAllowedByPolicyName(
   policy?: SandboxToolPolicy,
 ): boolean {
   if (!policy) return true;
-  const deny = new Set(normalizeToolNames(policy.deny));
+  const deny = new Set(expandToolGroups(policy.deny));
-  const allowRaw = normalizeToolNames(policy.allow);
+  const allowRaw = expandToolGroups(policy.allow);
   const allow = allowRaw.length > 0 ? new Set(allowRaw) : null;
   const normalized = normalizeToolName(name);
   if (deny.has(normalized)) return false;
@@ -391,11 +381,15 @@ function resolveEffectiveToolPolicy(params: {
       : undefined;
   const agentTools = agentConfig?.tools;
   const hasAgentToolPolicy =
-    Array.isArray(agentTools?.allow) || Array.isArray(agentTools?.deny);
+    Array.isArray(agentTools?.allow) ||
+    Array.isArray(agentTools?.deny) ||
+    typeof agentTools?.profile === "string";
   const globalTools = params.config?.tools;
+  const profile = agentTools?.profile ?? globalTools?.profile;
   return {
     agentId,
     policy: hasAgentToolPolicy ? agentTools : globalTools,
+    profile,
   };
 }
 
@@ -703,10 +697,15 @@ export function createClawdbotCodingTools(options?: {
 }): AnyAgentTool[] {
   const execToolName = "exec";
   const sandbox = options?.sandbox?.enabled ? options.sandbox : undefined;
-  const { agentId, policy: effectiveToolsPolicy } = resolveEffectiveToolPolicy({
+  const {
+    agentId,
+    policy: effectiveToolsPolicy,
+    profile,
+  } = resolveEffectiveToolPolicy({
     config: options?.config,
     sessionKey: options?.sessionKey,
   });
+  const profilePolicy = resolveToolProfilePolicy(profile);
   const scopeKey =
     options?.exec?.scopeKey ?? (agentId ? `agent:${agentId}` : undefined);
   const subagentPolicy =
@@ -714,6 +713,7 @@ export function createClawdbotCodingTools(options?: {
       ? resolveSubagentToolPolicy(options.config)
       : undefined;
   const allowBackground = isToolAllowedByPolicies("process", [
+    profilePolicy,
     effectiveToolsPolicy,
     sandbox?.tools,
     subagentPolicy,
@@ -829,12 +829,15 @@ export function createClawdbotCodingTools(options?: {
       hasRepliedRef: options?.hasRepliedRef,
     }),
   ];
-  const toolsFiltered = effectiveToolsPolicy
+  const toolsFiltered = profilePolicy
-    ? filterToolsByPolicy(tools, effectiveToolsPolicy)
+    ? filterToolsByPolicy(tools, profilePolicy)
     : tools;
-  const sandboxed = sandbox
+  const policyFiltered = effectiveToolsPolicy
-    ? filterToolsByPolicy(toolsFiltered, sandbox.tools)
+    ? filterToolsByPolicy(toolsFiltered, effectiveToolsPolicy)
     : toolsFiltered;
+  const sandboxed = sandbox
+    ? filterToolsByPolicy(policyFiltered, sandbox.tools)
+    : policyFiltered;
   const subagentFiltered = subagentPolicy
     ? filterToolsByPolicy(sandboxed, subagentPolicy)
     : sandboxed;

src/agents/sandbox.ts

@@ -33,6 +33,7 @@ import {
   resolveSessionAgentId,
 } from "./agent-scope.js";
 import { syncSkillsToWorkspace } from "./skills.js";
+import { expandToolGroups } from "./tool-policy.js";
 import {
   DEFAULT_AGENT_WORKSPACE_DIR,
   DEFAULT_AGENTS_FILENAME,
@@ -239,58 +240,10 @@ const BROWSER_BRIDGES = new Map<
   { bridge: BrowserBridge; containerName: string }
 >();
 
-function normalizeToolList(values?: string[]) {
-  if (!values) return [];
-  return values
-    .map((value) => value.trim())
-    .filter(Boolean)
-    .map((value) => value.toLowerCase());
-}
-
-const TOOL_GROUPS: Record<string, string[]> = {
-  // NOTE: Keep canonical (lowercase) tool names here.
-  "group:memory": ["memory_search", "memory_get"],
-  // Basic workspace/file tools
-  "group:fs": ["read", "write", "edit", "apply_patch"],
-  // Session management tools
-  "group:sessions": [
-    "sessions_list",
-    "sessions_history",
-    "sessions_send",
-    "sessions_spawn",
-    "session_status",
-  ],
-  // Host/runtime execution tools
-  "group:runtime": ["exec", "bash", "process"],
-};
-
-function expandToolGroupEntry(entry: string): string[] {
-  const raw = entry.trim();
-  if (!raw) return [];
-  const lower = raw.toLowerCase();
-
-  const group = TOOL_GROUPS[lower];
-  if (group) return group;
-  return [raw];
-}
-
-function expandToolGroups(values?: string[]): string[] {
-  if (!values) return [];
-  const out: string[] = [];
-  for (const value of values) {
-    for (const expanded of expandToolGroupEntry(value)) {
-      const trimmed = expanded.trim();
-      if (!trimmed) continue;
-      out.push(trimmed);
-    }
-  }
-  return out;
-}
-
 function isToolAllowed(policy: SandboxToolPolicy, name: string) {
-  const deny = new Set(normalizeToolList(expandToolGroups(policy.deny)));
+  const deny = new Set(expandToolGroups(policy.deny));
   if (deny.has(name.toLowerCase())) return false;
-  const allow = normalizeToolList(expandToolGroups(policy.allow));
+  const allow = expandToolGroups(policy.allow);
   if (allow.length === 0) return true;
   return allow.includes(name.toLowerCase());
 }
@@ -687,8 +640,8 @@ export function formatSandboxToolPolicyBlockedMessage(params: {
   });
   if (!runtime.sandboxed) return undefined;
 
-  const deny = new Set(normalizeToolList(runtime.toolPolicy.deny));
+  const deny = new Set(expandToolGroups(runtime.toolPolicy.deny));
-  const allow = normalizeToolList(runtime.toolPolicy.allow);
+  const allow = expandToolGroups(runtime.toolPolicy.allow);
   const allowSet = allow.length > 0 ? new Set(allow) : null;
   const blockedByDeny = deny.has(tool);
   const blockedByAllow = allowSet ? !allowSet.has(tool) : false;

src/agents/schema/typebox.ts

@@ -0,0 +1,27 @@
+import { Type } from "@sinclair/typebox";
+
+type StringEnumOptions<T extends readonly string[]> = {
+  description?: string;
+  title?: string;
+  default?: T[number];
+};
+
+// NOTE: Avoid Type.Union([Type.Literal(...)]) which compiles to anyOf.
+// Some providers reject anyOf in tool schemas; a flat string enum is safer.
+export function stringEnum<T extends readonly string[]>(
+  values: T,
+  options: StringEnumOptions<T> = {},
+) {
+  return Type.Unsafe<T[number]>({
+    type: "string",
+    enum: [...values],
+    ...options,
+  });
+}
+
+export function optionalStringEnum<T extends readonly string[]>(
+  values: T,
+  options: StringEnumOptions<T> = {},
+) {
+  return Type.Optional(stringEnum(values, options));
+}

src/agents/tool-policy.test.ts

@@ -0,0 +1,38 @@
+import { describe, expect, it } from "vitest";
+import {
+  expandToolGroups,
+  resolveToolProfilePolicy,
+  TOOL_GROUPS,
+} from "./tool-policy.js";
+
+describe("tool-policy", () => {
+  it("expands groups and normalizes aliases", () => {
+    const expanded = expandToolGroups([
+      "group:runtime",
+      "BASH",
+      "apply-patch",
+      "group:fs",
+    ]);
+    const set = new Set(expanded);
+    expect(set.has("exec")).toBe(true);
+    expect(set.has("bash")).toBe(true);
+    expect(set.has("process")).toBe(true);
+    expect(set.has("apply_patch")).toBe(true);
+    expect(set.has("read")).toBe(true);
+    expect(set.has("write")).toBe(true);
+    expect(set.has("edit")).toBe(true);
+  });
+
+  it("resolves known profiles and ignores unknown ones", () => {
+    const coding = resolveToolProfilePolicy("coding");
+    expect(coding?.allow).toContain("group:fs");
+    expect(resolveToolProfilePolicy("nope")).toBeUndefined();
+  });
+
+  it("includes core tool groups in group:clawdbot", () => {
+    const group = TOOL_GROUPS["group:clawdbot"];
+    expect(group).toContain("browser");
+    expect(group).toContain("message");
+    expect(group).toContain("session_status");
+  });
+});

src/agents/tool-policy.ts

@@ -0,0 +1,116 @@
+export type ToolProfileId = "minimal" | "coding" | "messaging" | "full";
+
+type ToolProfilePolicy = {
+  allow?: string[];
+  deny?: string[];
+};
+
+const TOOL_NAME_ALIASES: Record<string, string> = {
+  bash: "exec",
+  "apply-patch": "apply_patch",
+};
+
+export const TOOL_GROUPS: Record<string, string[]> = {
+  // NOTE: Keep canonical (lowercase) tool names here.
+  "group:memory": ["memory_search", "memory_get"],
+  // Basic workspace/file tools
+  "group:fs": ["read", "write", "edit", "apply_patch"],
+  // Host/runtime execution tools
+  "group:runtime": ["exec", "bash", "process"],
+  // Session management tools
+  "group:sessions": [
+    "sessions_list",
+    "sessions_history",
+    "sessions_send",
+    "sessions_spawn",
+    "session_status",
+  ],
+  // UI helpers
+  "group:ui": ["browser", "canvas"],
+  // Automation + infra
+  "group:automation": ["cron", "gateway"],
+  // Messaging surface
+  "group:messaging": ["message"],
+  // Nodes + device tools
+  "group:nodes": ["nodes"],
+  // All Clawdbot native tools (excludes provider plugins).
+  "group:clawdbot": [
+    "browser",
+    "canvas",
+    "nodes",
+    "cron",
+    "message",
+    "gateway",
+    "agents_list",
+    "sessions_list",
+    "sessions_history",
+    "sessions_send",
+    "sessions_spawn",
+    "session_status",
+    "memory_search",
+    "memory_get",
+    "image",
+  ],
+};
+
+const TOOL_PROFILES: Record<ToolProfileId, ToolProfilePolicy> = {
+  minimal: {
+    allow: ["session_status"],
+  },
+  coding: {
+    allow: [
+      "group:fs",
+      "group:runtime",
+      "group:sessions",
+      "group:memory",
+      "image",
+    ],
+  },
+  messaging: {
+    allow: [
+      "group:messaging",
+      "sessions_list",
+      "sessions_history",
+      "sessions_send",
+      "session_status",
+    ],
+  },
+  full: {},
+};
+
+export function normalizeToolName(name: string) {
+  const normalized = name.trim().toLowerCase();
+  return TOOL_NAME_ALIASES[normalized] ?? normalized;
+}
+
+export function normalizeToolList(list?: string[]) {
+  if (!list) return [];
+  return list.map(normalizeToolName).filter(Boolean);
+}
+
+export function expandToolGroups(list?: string[]) {
+  const normalized = normalizeToolList(list);
+  const expanded: string[] = [];
+  for (const value of normalized) {
+    const group = TOOL_GROUPS[value];
+    if (group) {
+      expanded.push(...group);
+      continue;
+    }
+    expanded.push(value);
+  }
+  return Array.from(new Set(expanded));
+}
+
+export function resolveToolProfilePolicy(
+  profile?: string,
+): ToolProfilePolicy | undefined {
+  if (!profile) return undefined;
+  const resolved = TOOL_PROFILES[profile as ToolProfileId];
+  if (!resolved) return undefined;
+  if (!resolved.allow && !resolved.deny) return undefined;
+  return {
+    allow: resolved.allow ? [...resolved.allow] : undefined,
+    deny: resolved.deny ? [...resolved.deny] : undefined,
+  };
+}

src/agents/tools/browser-tool.ts

@@ -22,6 +22,7 @@ import {
 import { resolveBrowserConfig } from "../../browser/config.js";
 import { DEFAULT_AI_SNAPSHOT_MAX_CHARS } from "../../browser/constants.js";
 import { loadConfig } from "../../config/config.js";
+import { optionalStringEnum, stringEnum } from "../schema/typebox.js";
 import {
   type AnyAgentTool,
   imageResultFromFile,
@@ -43,16 +44,35 @@ const BROWSER_ACT_KINDS = [
   "close",
 ] as const;
 
-type BrowserActKind = (typeof BROWSER_ACT_KINDS)[number];
+const BROWSER_TOOL_ACTIONS = [
+  "status",
+  "start",
+  "stop",
+  "tabs",
+  "open",
+  "focus",
+  "close",
+  "snapshot",
+  "screenshot",
+  "navigate",
+  "console",
+  "pdf",
+  "upload",
+  "dialog",
+  "act",
+] as const;
+
+const BROWSER_TARGETS = ["sandbox", "host", "custom"] as const;
+
+const BROWSER_SNAPSHOT_FORMATS = ["aria", "ai"] as const;
+
+const BROWSER_IMAGE_TYPES = ["png", "jpeg"] as const;
 
 // NOTE: Using a flattened object schema instead of Type.Union([Type.Object(...), ...])
 // because Claude API on Vertex AI rejects nested anyOf schemas as invalid JSON Schema.
 // The discriminator (kind) determines which properties are relevant; runtime validates.
 const BrowserActSchema = Type.Object({
-  kind: Type.Unsafe<BrowserActKind>({
+  kind: stringEnum(BROWSER_ACT_KINDS),
-    type: "string",
-    enum: [...BROWSER_ACT_KINDS],
-  }),
   // Common fields
   targetId: Type.Optional(Type.String()),
   ref: Type.Optional(Type.String()),
@@ -89,37 +109,15 @@ const BrowserActSchema = Type.Object({
 // A root-level `Type.Union([...])` compiles to `{ anyOf: [...] }` (no `type`),
 // which OpenAI rejects ("Invalid schema ... type: None"). Keep this schema an object.
 const BrowserToolSchema = Type.Object({
-  action: Type.Union([
+  action: stringEnum(BROWSER_TOOL_ACTIONS),
-    Type.Literal("status"),
+  target: optionalStringEnum(BROWSER_TARGETS),
-    Type.Literal("start"),
-    Type.Literal("stop"),
-    Type.Literal("tabs"),
-    Type.Literal("open"),
-    Type.Literal("focus"),
-    Type.Literal("close"),
-    Type.Literal("snapshot"),
-    Type.Literal("screenshot"),
-    Type.Literal("navigate"),
-    Type.Literal("console"),
-    Type.Literal("pdf"),
-    Type.Literal("upload"),
-    Type.Literal("dialog"),
-    Type.Literal("act"),
-  ]),
-  target: Type.Optional(
-    Type.Union([
-      Type.Literal("sandbox"),
-      Type.Literal("host"),
-      Type.Literal("custom"),
-    ]),
-  ),
   profile: Type.Optional(Type.String()),
   controlUrl: Type.Optional(Type.String()),
   targetUrl: Type.Optional(Type.String()),
   targetId: Type.Optional(Type.String()),
   limit: Type.Optional(Type.Number()),
   maxChars: Type.Optional(Type.Number()),
-  format: Type.Optional(Type.Union([Type.Literal("aria"), Type.Literal("ai")])),
+  format: optionalStringEnum(BROWSER_SNAPSHOT_FORMATS),
   interactive: Type.Optional(Type.Boolean()),
   compact: Type.Optional(Type.Boolean()),
   depth: Type.Optional(Type.Number()),
@@ -128,7 +126,7 @@ const BrowserToolSchema = Type.Object({
   fullPage: Type.Optional(Type.Boolean()),
   ref: Type.Optional(Type.String()),
   element: Type.Optional(Type.String()),
-  type: Type.Optional(Type.Union([Type.Literal("png"), Type.Literal("jpeg")])),
+  type: optionalStringEnum(BROWSER_IMAGE_TYPES),
   level: Type.Optional(Type.String()),
   paths: Type.Optional(Type.Array(Type.String())),
   inputRef: Type.Optional(Type.String()),

src/agents/tools/canvas-tool.ts

@@ -8,6 +8,7 @@ import {
   parseCanvasSnapshotPayload,
 } from "../../cli/nodes-canvas.js";
 import { imageMimeFromFormat } from "../../media/mime.js";
+import { optionalStringEnum, stringEnum } from "../schema/typebox.js";
 import {
   type AnyAgentTool,
   imageResult,
@@ -17,76 +18,44 @@ import {
 import { callGatewayTool, type GatewayCallOptions } from "./gateway.js";
 import { resolveNodeId } from "./nodes-utils.js";
 
-const CanvasToolSchema = Type.Union([
+const CANVAS_ACTIONS = [
-  Type.Object({
+  "present",
-    action: Type.Literal("present"),
+  "hide",
+  "navigate",
+  "eval",
+  "snapshot",
+  "a2ui_push",
+  "a2ui_reset",
+] as const;
+
+const CANVAS_SNAPSHOT_FORMATS = ["png", "jpg", "jpeg"] as const;
+
+// Flattened schema: runtime validates per-action requirements.
+const CanvasToolSchema = Type.Object({
+  action: stringEnum(CANVAS_ACTIONS),
   gatewayUrl: Type.Optional(Type.String()),
   gatewayToken: Type.Optional(Type.String()),
   timeoutMs: Type.Optional(Type.Number()),
   node: Type.Optional(Type.String()),
+  // present
   target: Type.Optional(Type.String()),
   x: Type.Optional(Type.Number()),
   y: Type.Optional(Type.Number()),
   width: Type.Optional(Type.Number()),
   height: Type.Optional(Type.Number()),
-  }),
+  // navigate
-  Type.Object({
+  url: Type.Optional(Type.String()),
-    action: Type.Literal("hide"),
+  // eval
-    gatewayUrl: Type.Optional(Type.String()),
+  javaScript: Type.Optional(Type.String()),
-    gatewayToken: Type.Optional(Type.String()),
+  // snapshot
-    timeoutMs: Type.Optional(Type.Number()),
+  format: optionalStringEnum(CANVAS_SNAPSHOT_FORMATS),
-    node: Type.Optional(Type.String()),
-  }),
-  Type.Object({
-    action: Type.Literal("navigate"),
-    gatewayUrl: Type.Optional(Type.String()),
-    gatewayToken: Type.Optional(Type.String()),
-    timeoutMs: Type.Optional(Type.Number()),
-    node: Type.Optional(Type.String()),
-    url: Type.String(),
-  }),
-  Type.Object({
-    action: Type.Literal("eval"),
-    gatewayUrl: Type.Optional(Type.String()),
-    gatewayToken: Type.Optional(Type.String()),
-    timeoutMs: Type.Optional(Type.Number()),
-    node: Type.Optional(Type.String()),
-    javaScript: Type.String(),
-  }),
-  Type.Object({
-    action: Type.Literal("snapshot"),
-    gatewayUrl: Type.Optional(Type.String()),
-    gatewayToken: Type.Optional(Type.String()),
-    timeoutMs: Type.Optional(Type.Number()),
-    node: Type.Optional(Type.String()),
-    format: Type.Optional(
-      Type.Union([
-        Type.Literal("png"),
-        Type.Literal("jpg"),
-        Type.Literal("jpeg"),
-      ]),
-    ),
   maxWidth: Type.Optional(Type.Number()),
   quality: Type.Optional(Type.Number()),
   delayMs: Type.Optional(Type.Number()),
-  }),
+  // a2ui_push
-  Type.Object({
-    action: Type.Literal("a2ui_push"),
-    gatewayUrl: Type.Optional(Type.String()),
-    gatewayToken: Type.Optional(Type.String()),
-    timeoutMs: Type.Optional(Type.Number()),
-    node: Type.Optional(Type.String()),
   jsonl: Type.Optional(Type.String()),
   jsonlPath: Type.Optional(Type.String()),
-  }),
+});
-  Type.Object({
-    action: Type.Literal("a2ui_reset"),
-    gatewayUrl: Type.Optional(Type.String()),
-    gatewayToken: Type.Optional(Type.String()),
-    timeoutMs: Type.Optional(Type.Number()),
-    node: Type.Optional(Type.String()),
-  }),
-]);
 
 export function createCanvasTool(): AnyAgentTool {
   return {

src/agents/tools/cron-tool.ts

@@ -3,89 +3,42 @@ import {
   normalizeCronJobCreate,
   normalizeCronJobPatch,
 } from "../../cron/normalize.js";
+import { optionalStringEnum, stringEnum } from "../schema/typebox.js";
 import { type AnyAgentTool, jsonResult, readStringParam } from "./common.js";
 import { callGatewayTool, type GatewayCallOptions } from "./gateway.js";
 
 // NOTE: We use Type.Object({}, { additionalProperties: true }) for job/patch
-// instead of CronAddParamsSchema/CronJobPatchSchema because:
+// instead of CronAddParamsSchema/CronJobPatchSchema because the gateway schemas
-//
+// contain nested unions. Tool schemas need to stay provider-friendly, so we
-// 1. CronAddParamsSchema contains nested Type.Union (for schedule, payload, etc.)
+// accept "any object" here and validate at runtime.
-// 2. TypeBox compiles Type.Union to JSON Schema `anyOf`
-// 3. pi-ai's sanitizeSchemaForGoogle() strips `anyOf` from nested properties
-// 4. This leaves empty schemas `{}` which Claude rejects as invalid
-//
-// The actual validation happens at runtime via normalizeCronJobCreate/Patch
-// and the gateway's validateCronAddParams. This schema just needs to accept
-// any object so the AI can pass through the job definition.
-//
-// See: https://github.com/anthropics/anthropic-cookbook/blob/main/misc/tool_use_best_practices.md
-// Claude requires valid JSON Schema 2020-12 with explicit types.
 
-const CronToolSchema = Type.Union([
+const CRON_ACTIONS = [
-  Type.Object({
+  "status",
-    action: Type.Literal("status"),
+  "list",
-    gatewayUrl: Type.Optional(Type.String()),
+  "add",
-    gatewayToken: Type.Optional(Type.String()),
+  "update",
-    timeoutMs: Type.Optional(Type.Number()),
+  "remove",
-  }),
+  "run",
-  Type.Object({
+  "runs",
-    action: Type.Literal("list"),
+  "wake",
+] as const;
+
+const CRON_WAKE_MODES = ["now", "next-heartbeat"] as const;
+
+// Flattened schema: runtime validates per-action requirements.
+const CronToolSchema = Type.Object({
+  action: stringEnum(CRON_ACTIONS),
   gatewayUrl: Type.Optional(Type.String()),
   gatewayToken: Type.Optional(Type.String()),
   timeoutMs: Type.Optional(Type.Number()),
   includeDisabled: Type.Optional(Type.Boolean()),
-  }),
+  job: Type.Optional(Type.Object({}, { additionalProperties: true })),
-  Type.Object({
-    action: Type.Literal("add"),
-    gatewayUrl: Type.Optional(Type.String()),
-    gatewayToken: Type.Optional(Type.String()),
-    timeoutMs: Type.Optional(Type.Number()),
-    job: Type.Object({}, { additionalProperties: true }),
-  }),
-  Type.Object({
-    action: Type.Literal("update"),
-    gatewayUrl: Type.Optional(Type.String()),
-    gatewayToken: Type.Optional(Type.String()),
-    timeoutMs: Type.Optional(Type.Number()),
   jobId: Type.Optional(Type.String()),
   id: Type.Optional(Type.String()),
-    patch: Type.Object({}, { additionalProperties: true }),
+  patch: Type.Optional(Type.Object({}, { additionalProperties: true })),
-  }),
+  text: Type.Optional(Type.String()),
-  Type.Object({
+  mode: optionalStringEnum(CRON_WAKE_MODES),
-    action: Type.Literal("remove"),
+});
-    gatewayUrl: Type.Optional(Type.String()),
-    gatewayToken: Type.Optional(Type.String()),
-    timeoutMs: Type.Optional(Type.Number()),
-    jobId: Type.Optional(Type.String()),
-    id: Type.Optional(Type.String()),
-  }),
-  Type.Object({
-    action: Type.Literal("run"),
-    gatewayUrl: Type.Optional(Type.String()),
-    gatewayToken: Type.Optional(Type.String()),
-    timeoutMs: Type.Optional(Type.Number()),
-    jobId: Type.Optional(Type.String()),
-    id: Type.Optional(Type.String()),
-  }),
-  Type.Object({
-    action: Type.Literal("runs"),
-    gatewayUrl: Type.Optional(Type.String()),
-    gatewayToken: Type.Optional(Type.String()),
-    timeoutMs: Type.Optional(Type.Number()),
-    jobId: Type.Optional(Type.String()),
-    id: Type.Optional(Type.String()),
-  }),
-  Type.Object({
-    action: Type.Literal("wake"),
-    gatewayUrl: Type.Optional(Type.String()),
-    gatewayToken: Type.Optional(Type.String()),
-    timeoutMs: Type.Optional(Type.Number()),
-    text: Type.String(),
-    mode: Type.Optional(
-      Type.Union([Type.Literal("now"), Type.Literal("next-heartbeat")]),
-    ),
-  }),
-]);
 
 export function createCronTool(): AnyAgentTool {
   return {

src/agents/tools/discord-actions-guild.ts

@@ -31,6 +31,7 @@ import {
 function readParentIdParam(
   params: Record<string, unknown>,
 ): string | null | undefined {
+  if (params.clearParent === true) return null;
   if (params.parentId === null) return null;
   return readStringParam(params, "parentId");
 }

src/agents/tools/discord-actions.test.ts

@@ -218,6 +218,26 @@ describe("handleDiscordGuildAction - channel management", () => {
     });
   });
 
+  it("clears the channel parent when clearParent is true", async () => {
+    await handleDiscordGuildAction(
+      "channelEdit",
+      {
+        channelId: "C1",
+        clearParent: true,
+      },
+      channelsEnabled,
+    );
+    expect(editChannelDiscord).toHaveBeenCalledWith({
+      channelId: "C1",
+      name: undefined,
+      topic: undefined,
+      position: undefined,
+      parentId: null,
+      nsfw: undefined,
+      rateLimitPerUser: undefined,
+    });
+  });
+
   it("deletes a channel", async () => {
     await handleDiscordGuildAction(
       "channelDelete",
@@ -264,6 +284,24 @@ describe("handleDiscordGuildAction - channel management", () => {
     });
   });
 
+  it("clears the channel parent on move when clearParent is true", async () => {
+    await handleDiscordGuildAction(
+      "channelMove",
+      {
+        guildId: "G1",
+        channelId: "C1",
+        clearParent: true,
+      },
+      channelsEnabled,
+    );
+    expect(moveChannelDiscord).toHaveBeenCalledWith({
+      guildId: "G1",
+      channelId: "C1",
+      parentId: null,
+      position: undefined,
+    });
+  });
+
   it("creates a category with type=4", async () => {
     await handleDiscordGuildAction(
       "categoryCreate",

src/agents/tools/gateway-tool.ts

@@ -7,6 +7,7 @@ import {
   type RestartSentinelPayload,
   writeRestartSentinel,
 } from "../../infra/restart-sentinel.js";
+import { stringEnum } from "../schema/typebox.js";
 import { type AnyAgentTool, jsonResult, readStringParam } from "./common.js";
 import { callGatewayTool } from "./gateway.js";
 
@@ -18,16 +19,11 @@ const GATEWAY_ACTIONS = [
   "update.run",
 ] as const;
 
-type GatewayAction = (typeof GATEWAY_ACTIONS)[number];
-
 // NOTE: Using a flattened object schema instead of Type.Union([Type.Object(...), ...])
 // because Claude API on Vertex AI rejects nested anyOf schemas as invalid JSON Schema.
 // The discriminator (action) determines which properties are relevant; runtime validates.
 const GatewayToolSchema = Type.Object({
-  action: Type.Unsafe<GatewayAction>({
+  action: stringEnum(GATEWAY_ACTIONS),
-    type: "string",
-    enum: [...GATEWAY_ACTIONS],
-  }),
   // restart
   delayMs: Type.Optional(Type.Number()),
   reason: Type.Optional(Type.String()),

src/agents/tools/message-tool.ts

@@ -16,6 +16,7 @@ import {
   type ProviderMessageActionName,
 } from "../../providers/plugins/types.js";
 import { normalizeAccountId } from "../../routing/session-key.js";
+import { stringEnum } from "../schema/typebox.js";
 import type { AnyAgentTool } from "./common.js";
 import { jsonResult, readNumberParam, readStringParam } from "./common.js";
 
@@ -90,12 +91,17 @@ const MessageToolCommonSchema = {
   timeoutMs: Type.Optional(Type.Number()),
   name: Type.Optional(Type.String()),
   type: Type.Optional(Type.Number()),
-  parentId: Type.Optional(Type.Union([Type.String(), Type.Null()])),
+  parentId: Type.Optional(Type.String()),
   topic: Type.Optional(Type.String()),
   position: Type.Optional(Type.Number()),
   nsfw: Type.Optional(Type.Boolean()),
   rateLimitPerUser: Type.Optional(Type.Number()),
   categoryId: Type.Optional(Type.String()),
+  clearParent: Type.Optional(
+    Type.Boolean({
+      description: "Clear the parent/category when supported by the provider.",
+    }),
+  ),
 };
 
 function buildMessageToolSchemaFromActions(
@@ -105,31 +111,10 @@ function buildMessageToolSchemaFromActions(
   const props: Record<string, unknown> = { ...MessageToolCommonSchema };
   if (!options.includeButtons) delete props.buttons;
 
-  const schemas: Array<ReturnType<typeof Type.Object>> = [];
+  return Type.Object({
-  if (actions.includes("send")) {
+    action: stringEnum(actions),
-    schemas.push(
-      Type.Object({
-        action: Type.Literal("send"),
-        to: Type.String(),
-        message: Type.String(),
     ...props,
-      }),
+  });
-    );
-  }
-
-  const nonSendActions = actions.filter((action) => action !== "send");
-  if (nonSendActions.length > 0) {
-    schemas.push(
-      Type.Object({
-        action: Type.Union(
-          nonSendActions.map((action) => Type.Literal(action)),
-        ),
-        ...props,
-      }),
-    );
-  }
-
-  return schemas.length === 1 ? schemas[0] : Type.Union(schemas);
 }
 
 const MessageToolSchema = buildMessageToolSchemaFromActions(AllMessageActions, {

src/agents/tools/nodes-tool.ts

@@ -18,151 +18,73 @@ import {
 } from "../../cli/nodes-screen.js";
 import { parseDurationMs } from "../../cli/parse-duration.js";
 import { imageMimeFromFormat } from "../../media/mime.js";
+import { optionalStringEnum, stringEnum } from "../schema/typebox.js";
 import { sanitizeToolResultImages } from "../tool-images.js";
 import { type AnyAgentTool, jsonResult, readStringParam } from "./common.js";
 import { callGatewayTool, type GatewayCallOptions } from "./gateway.js";
 import { resolveNodeId } from "./nodes-utils.js";
 
-const NodesToolSchema = Type.Union([
+const NODES_TOOL_ACTIONS = [
-  Type.Object({
+  "status",
-    action: Type.Literal("status"),
+  "describe",
+  "pending",
+  "approve",
+  "reject",
+  "notify",
+  "camera_snap",
+  "camera_list",
+  "camera_clip",
+  "screen_record",
+  "location_get",
+  "run",
+] as const;
+
+const NOTIFY_PRIORITIES = ["passive", "active", "timeSensitive"] as const;
+const NOTIFY_DELIVERIES = ["system", "overlay", "auto"] as const;
+const CAMERA_FACING = ["front", "back", "both"] as const;
+const LOCATION_ACCURACY = ["coarse", "balanced", "precise"] as const;
+
+// Flattened schema: runtime validates per-action requirements.
+const NodesToolSchema = Type.Object({
+  action: stringEnum(NODES_TOOL_ACTIONS),
   gatewayUrl: Type.Optional(Type.String()),
   gatewayToken: Type.Optional(Type.String()),
   timeoutMs: Type.Optional(Type.Number()),
-  }),
+  node: Type.Optional(Type.String()),
-  Type.Object({
+  requestId: Type.Optional(Type.String()),
-    action: Type.Literal("describe"),
+  // notify
-    gatewayUrl: Type.Optional(Type.String()),
-    gatewayToken: Type.Optional(Type.String()),
-    timeoutMs: Type.Optional(Type.Number()),
-    node: Type.String(),
-  }),
-  Type.Object({
-    action: Type.Literal("pending"),
-    gatewayUrl: Type.Optional(Type.String()),
-    gatewayToken: Type.Optional(Type.String()),
-    timeoutMs: Type.Optional(Type.Number()),
-  }),
-  Type.Object({
-    action: Type.Literal("approve"),
-    gatewayUrl: Type.Optional(Type.String()),
-    gatewayToken: Type.Optional(Type.String()),
-    timeoutMs: Type.Optional(Type.Number()),
-    requestId: Type.String(),
-  }),
-  Type.Object({
-    action: Type.Literal("reject"),
-    gatewayUrl: Type.Optional(Type.String()),
-    gatewayToken: Type.Optional(Type.String()),
-    timeoutMs: Type.Optional(Type.Number()),
-    requestId: Type.String(),
-  }),
-  Type.Object({
-    action: Type.Literal("notify"),
-    gatewayUrl: Type.Optional(Type.String()),
-    gatewayToken: Type.Optional(Type.String()),
-    timeoutMs: Type.Optional(Type.Number()),
-    node: Type.String(),
   title: Type.Optional(Type.String()),
   body: Type.Optional(Type.String()),
   sound: Type.Optional(Type.String()),
-    priority: Type.Optional(
+  priority: optionalStringEnum(NOTIFY_PRIORITIES),
-      Type.Union([
+  delivery: optionalStringEnum(NOTIFY_DELIVERIES),
-        Type.Literal("passive"),
+  // camera_snap / camera_clip
-        Type.Literal("active"),
+  facing: optionalStringEnum(CAMERA_FACING, {
-        Type.Literal("timeSensitive"),
+    description: "camera_snap: front/back/both; camera_clip: front/back only.",
-      ]),
-    ),
-    delivery: Type.Optional(
-      Type.Union([
-        Type.Literal("system"),
-        Type.Literal("overlay"),
-        Type.Literal("auto"),
-      ]),
-    ),
   }),
-  Type.Object({
-    action: Type.Literal("camera_snap"),
-    gatewayUrl: Type.Optional(Type.String()),
-    gatewayToken: Type.Optional(Type.String()),
-    timeoutMs: Type.Optional(Type.Number()),
-    node: Type.String(),
-    facing: Type.Optional(
-      Type.Union([
-        Type.Literal("front"),
-        Type.Literal("back"),
-        Type.Literal("both"),
-      ]),
-    ),
   maxWidth: Type.Optional(Type.Number()),
   quality: Type.Optional(Type.Number()),
   delayMs: Type.Optional(Type.Number()),
   deviceId: Type.Optional(Type.String()),
-  }),
-  Type.Object({
-    action: Type.Literal("camera_list"),
-    gatewayUrl: Type.Optional(Type.String()),
-    gatewayToken: Type.Optional(Type.String()),
-    timeoutMs: Type.Optional(Type.Number()),
-    node: Type.String(),
-  }),
-  Type.Object({
-    action: Type.Literal("camera_clip"),
-    gatewayUrl: Type.Optional(Type.String()),
-    gatewayToken: Type.Optional(Type.String()),
-    timeoutMs: Type.Optional(Type.Number()),
-    node: Type.String(),
-    facing: Type.Optional(
-      Type.Union([Type.Literal("front"), Type.Literal("back")]),
-    ),
   duration: Type.Optional(Type.String()),
   durationMs: Type.Optional(Type.Number()),
   includeAudio: Type.Optional(Type.Boolean()),
-    deviceId: Type.Optional(Type.String()),
+  // screen_record
-  }),
-  Type.Object({
-    action: Type.Literal("screen_record"),
-    gatewayUrl: Type.Optional(Type.String()),
-    gatewayToken: Type.Optional(Type.String()),
-    timeoutMs: Type.Optional(Type.Number()),
-    node: Type.String(),
-    duration: Type.Optional(Type.String()),
-    durationMs: Type.Optional(Type.Number()),
   fps: Type.Optional(Type.Number()),
   screenIndex: Type.Optional(Type.Number()),
-    includeAudio: Type.Optional(Type.Boolean()),
   outPath: Type.Optional(Type.String()),
-  }),
+  // location_get
-  Type.Object({
-    action: Type.Literal("location_get"),
-    gatewayUrl: Type.Optional(Type.String()),
-    gatewayToken: Type.Optional(Type.String()),
-    timeoutMs: Type.Optional(Type.Number()),
-    node: Type.String(),
   maxAgeMs: Type.Optional(Type.Number()),
   locationTimeoutMs: Type.Optional(Type.Number()),
-    desiredAccuracy: Type.Optional(
+  desiredAccuracy: optionalStringEnum(LOCATION_ACCURACY),
-      Type.Union([
+  // run
-        Type.Literal("coarse"),
+  command: Type.Optional(Type.Array(Type.String())),
-        Type.Literal("balanced"),
-        Type.Literal("precise"),
-      ]),
-    ),
-  }),
-  Type.Object({
-    action: Type.Literal("run"),
-    gatewayUrl: Type.Optional(Type.String()),
-    gatewayToken: Type.Optional(Type.String()),
-    timeoutMs: Type.Optional(Type.Number()),
-    node: Type.String(),
-    command: Type.Array(Type.String()),
   cwd: Type.Optional(Type.String()),
   env: Type.Optional(Type.Array(Type.String())),
   commandTimeoutMs: Type.Optional(Type.Number()),
   invokeTimeoutMs: Type.Optional(Type.Number()),
   needsScreenRecording: Type.Optional(Type.Boolean()),
-  }),
+});
-]);
 
 export function createNodesTool(): AnyAgentTool {
   return {

src/agents/tools/sessions-spawn-tool.ts

@@ -12,6 +12,7 @@ import {
 import type { GatewayMessageProvider } from "../../utils/message-provider.js";
 import { resolveAgentConfig } from "../agent-scope.js";
 import { AGENT_LANE_SUBAGENT } from "../lanes.js";
+import { optionalStringEnum } from "../schema/typebox.js";
 import { buildSubagentSystemPrompt } from "../subagent-announce.js";
 import { registerSubagentRun } from "../subagent-registry.js";
 import type { AnyAgentTool } from "./common.js";
@@ -30,9 +31,7 @@ const SessionsSpawnToolSchema = Type.Object({
   runTimeoutSeconds: Type.Optional(Type.Number({ minimum: 0 })),
   // Back-compat alias. Prefer runTimeoutSeconds.
   timeoutSeconds: Type.Optional(Type.Number({ minimum: 0 })),
-  cleanup: Type.Optional(
+  cleanup: optionalStringEnum(["delete", "keep"] as const),
-    Type.Union([Type.Literal("delete"), Type.Literal("keep")]),
-  ),
 });
 
 function normalizeModelSelection(value: unknown): string | undefined {

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";
@@ -1197,7 +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();
     }
@@ -1205,6 +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/agent-runner.ts

@@ -1,5 +1,6 @@
 import crypto from "node:crypto";
 import fs from "node:fs";
+import { resolveAgentModelFallbacksOverride } from "../../agents/agent-scope.js";
 import { runCliAgent } from "../../agents/cli-runner.js";
 import { getCliSessionId, setCliSessionId } from "../../agents/cli-session.js";
 import { lookupContextTokens } from "../../agents/context.js";
@@ -394,6 +395,10 @@ export async function runReplyAgent(params: {
         cfg: followupRun.run.config,
         provider: followupRun.run.provider,
         model: followupRun.run.model,
+        fallbacksOverride: resolveAgentModelFallbacksOverride(
+          followupRun.run.config,
+          resolveAgentIdFromSessionKey(followupRun.run.sessionKey),
+        ),
         run: (provider, model) =>
           runEmbeddedPiAgent({
             sessionId: followupRun.run.sessionId,
@@ -586,6 +591,10 @@ export async function runReplyAgent(params: {
           cfg: followupRun.run.config,
           provider: followupRun.run.provider,
           model: followupRun.run.model,
+          fallbacksOverride: resolveAgentModelFallbacksOverride(
+            followupRun.run.config,
+            resolveAgentIdFromSessionKey(followupRun.run.sessionKey),
+          ),
           run: (provider, model) => {
             if (isCliProvider(provider, followupRun.run.config)) {
               const startedAt = Date.now();

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

@@ -155,6 +155,30 @@ function buildUsageReply(): ReplyPayload {
   };
 }
 
+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 }>;
@@ -199,9 +223,13 @@ export async function handleBashChatCommand(params: {
   };
 }): Promise<ReplyPayload> {
   if (params.cfg.commands?.bash !== true) {
-    return {
+    logVerbose("Blocked /bash: commands.bash is disabled.");
-      text: "⚠️ bash is disabled. Set commands.bash=true to enable.",
+    const provider =
-    };
+      params.ctx.Provider ??
+      params.ctx.Surface ??
+      params.ctx.OriginatingChannel ??
+      undefined;
+    return buildBashDisabledReply(provider ?? undefined);
   }
 
   const agentId =

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

@@ -1,6 +1,6 @@
 import {
-  resolveAgentConfig,
   resolveAgentDir,
+  resolveAgentModelPrimary,
   resolveDefaultAgentId,
   resolveSessionAgentId,
 } from "../../agents/agent-scope.js";
@@ -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) {
@@ -1098,125 +1215,24 @@ 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]+$/.test(raw)) {
+  ) {
-      const resolvedDefault = resolveConfiguredModelRef({
-        cfg: params.cfg,
-        defaultProvider,
-        defaultModel,
-      });
-      const pickerCatalog: ModelPickerCatalogEntry[] = (() => {
-        const keys = new Set<string>();
-        const out: ModelPickerCatalogEntry[] = [];
-
-        const push = (entry: ModelPickerCatalogEntry) => {
-          const provider = normalizeProviderId(entry.provider);
-          const id = String(entry.id ?? "").trim();
-          if (!provider || !id) return;
-          const key = modelKey(provider, id);
-          if (keys.has(key)) return;
-          keys.add(key);
-          out.push({ provider, id, name: entry.name });
-        };
-
-        for (const entry of allowedModelCatalog) push(entry);
-
-        for (const rawKey of Object.keys(
-          params.cfg.agents?.defaults?.models ?? {},
-        )) {
-          const resolved = resolveModelRefFromString({
-            raw: String(rawKey),
-            defaultProvider,
-            aliasIndex,
-          });
-          if (!resolved) continue;
-          push({
-            provider: resolved.ref.provider,
-            id: resolved.ref.model,
-            name: resolved.ref.model,
-          });
-        }
-        if (resolvedDefault.model) {
-          push({
-            provider: resolvedDefault.provider,
-            id: resolvedDefault.model,
-            name: resolvedDefault.model,
-          });
-        }
-        return out;
-      })();
-
-      const items = buildModelPickerItems(pickerCatalog);
-      const index = Number.parseInt(raw, 10) - 1;
-      const item = Number.isFinite(index) ? items[index] : undefined;
-      if (!item) {
     return {
-          text: `Invalid model selection "${raw}". Use /model to list.`,
+      text: `Thinking level "xhigh" is only supported for ${formatXHighModelHint()}.`,
     };
   }
-      const picked = pickProviderForModel({
+
-        item,
+  const nextThinkLevel = directives.hasThinkDirective
-        preferredProvider: params.provider,
+    ? directives.thinkLevel
-      });
+    : ((sessionEntry?.thinkingLevel as ThinkLevel | undefined) ??
-      if (!picked) {
+      currentThinkLevel);
-        return {
+  const shouldDowngradeXHigh =
-          text: `Invalid model selection "${raw}". Use /model to list.`,
+    !directives.hasThinkDirective &&
-        };
+    nextThinkLevel === "xhigh" &&
-      }
+    !supportsXHighThinking(resolvedProvider, resolvedModel);
-      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." };
-  }
 
   if (sessionEntry && sessionStore && sessionKey) {
     const prevElevatedLevel =
@@ -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
@@ -1593,7 +1629,7 @@ export function resolveDefaultModel(params: {
   aliasIndex: ModelAliasIndex;
 } {
   const agentModelOverride = params.agentId
-    ? resolveAgentConfig(params.cfg, params.agentId)?.model?.trim()
+    ? resolveAgentModelPrimary(params.cfg, params.agentId)
     : undefined;
   const cfg =
     agentModelOverride && agentModelOverride.length > 0

src/auto-reply/reply/followup-runner.ts

@@ -1,10 +1,12 @@
 import crypto from "node:crypto";
+import { resolveAgentModelFallbacksOverride } from "../../agents/agent-scope.js";
 import { lookupContextTokens } from "../../agents/context.js";
 import { DEFAULT_CONTEXT_TOKENS } from "../../agents/defaults.js";
 import { runWithModelFallback } from "../../agents/model-fallback.js";
 import { runEmbeddedPiAgent } from "../../agents/pi-embedded.js";
 import { hasNonzeroUsage } from "../../agents/usage.js";
 import {
+  resolveAgentIdFromSessionKey,
   type SessionEntry,
   updateSessionStoreEntry,
 } from "../../config/sessions.js";
@@ -136,6 +138,10 @@ export function createFollowupRunner(params: {
           cfg: queued.run.config,
           provider: queued.run.provider,
           model: queued.run.model,
+          fallbacksOverride: resolveAgentModelFallbacksOverride(
+            queued.run.config,
+            resolveAgentIdFromSessionKey(queued.run.sessionKey),
+          ),
           run: (provider, model) =>
             runEmbeddedPiAgent({
               sessionId: queued.run.sessionId,

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

@@ -1,6 +1,8 @@
 import crypto from "node:crypto";
 import {
   resolveAgentDir,
+  resolveAgentModelFallbacksOverride,
+  resolveAgentModelPrimary,
   resolveAgentWorkspaceDir,
 } from "../agents/agent-scope.js";
 import { ensureAuthProfileStore } from "../agents/auth-profiles.js";
@@ -28,8 +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";
@@ -214,17 +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}.`,
     );
   }
 
@@ -333,9 +347,28 @@ export async function agentCommand(
     await saveSessionStore(storePath, sessionStore);
   }
 
+  const agentModelPrimary = resolveAgentModelPrimary(cfg, sessionAgentId);
+  const cfgForModelSelection = agentModelPrimary
+    ? {
+        ...cfg,
+        agents: {
+          ...cfg.agents,
+          defaults: {
+            ...cfg.agents?.defaults,
+            model: {
+              ...(typeof cfg.agents?.defaults?.model === "object"
+                ? cfg.agents.defaults.model
+                : undefined),
+              primary: agentModelPrimary,
+            },
+          },
+        },
+      }
+    : cfg;
+
   const { provider: defaultProvider, model: defaultModel } =
     resolveConfiguredModelRef({
-      cfg,
+      cfg: cfgForModelSelection,
       defaultProvider: DEFAULT_PROVIDER,
       defaultModel: DEFAULT_MODEL,
     });
@@ -423,6 +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,
   });
@@ -442,6 +498,10 @@ export async function agentCommand(
       cfg,
       provider,
       model,
+      fallbacksOverride: resolveAgentModelFallbacksOverride(
+        cfg,
+        sessionAgentId,
+      ),
       run: (providerOverride, modelOverride) => {
         if (isCliProvider(providerOverride, cfg)) {
           const cliSessionId = getCliSessionId(sessionEntry, providerOverride);

src/commands/agents.ts

@@ -142,7 +142,15 @@ function resolveAgentModel(cfg: ClawdbotConfig, agentId: string) {
   const entry = listAgentEntries(cfg).find(
     (agent) => normalizeAgentId(agent.id) === normalizeAgentId(agentId),
   );
-  if (entry?.model?.trim()) return entry.model.trim();
+  if (entry?.model) {
+    if (typeof entry.model === "string" && entry.model.trim()) {
+      return entry.model.trim();
+    }
+    if (typeof entry.model === "object") {
+      const primary = entry.model.primary?.trim();
+      if (primary) return primary;
+    }
+  }
   const raw = cfg.agents?.defaults?.model;
   if (typeof raw === "string") return raw;
   return raw?.primary?.trim() || undefined;

src/commands/auth-choice.ts

@@ -1,5 +1,5 @@
 import { loginOpenAICodex, type OAuthCredentials } from "@mariozechner/pi-ai";
-import { resolveAgentConfig } from "../agents/agent-scope.js";
+import { resolveAgentModelPrimary } from "../agents/agent-scope.js";
 import {
   CLAUDE_CLI_PROFILE_ID,
   CODEX_CLI_PROFILE_ID,
@@ -152,7 +152,7 @@ export async function warnIfModelConfigLooksOff(
   options?: { agentId?: string; agentDir?: string },
 ) {
   const agentModelOverride = options?.agentId
-    ? resolveAgentConfig(config, options.agentId)?.model?.trim()
+    ? resolveAgentModelPrimary(config, options.agentId)
     : undefined;
   const configWithModel =
     agentModelOverride && agentModelOverride.length > 0

src/config/schema.ts

@@ -107,6 +107,8 @@ const FIELD_LABELS: Record<string, string> = {
   "tools.audio.transcription.args": "Audio Transcription Args",
   "tools.audio.transcription.timeoutSeconds":
     "Audio Transcription Timeout (sec)",
+  "tools.profile": "Tool Profile",
+  "agents.list[].tools.profile": "Agent Tool Profile",
   "tools.exec.applyPatch.enabled": "Enable apply_patch",
   "tools.exec.applyPatch.allowModels": "apply_patch Model Allowlist",
   "gateway.controlUi.basePath": "Control UI Base Path",

src/config/types.ts

@@ -988,7 +988,11 @@ export type QueueConfig = {
   drop?: QueueDropPolicy;
 };
 
+export type ToolProfileId = "minimal" | "coding" | "messaging" | "full";
+
 export type AgentToolsConfig = {
+  /** Base tool profile applied before allow/deny lists. */
+  profile?: ToolProfileId;
   allow?: string[];
   deny?: string[];
   /** Per-agent elevated exec gate (can only further restrict global tools.elevated). */
@@ -1053,6 +1057,8 @@ export type MemorySearchConfig = {
 };
 
 export type ToolsConfig = {
+  /** Base tool profile applied before allow/deny lists. */
+  profile?: ToolProfileId;
   allow?: string[];
   deny?: string[];
   audio?: {
@@ -1121,13 +1127,22 @@ export type ToolsConfig = {
   };
 };
 
+export type AgentModelConfig =
+  | string
+  | {
+      /** Primary model (provider/model). */
+      primary?: string;
+      /** Per-agent model fallbacks (provider/model). */
+      fallbacks?: string[];
+    };
+
 export type AgentConfig = {
   id: string;
   default?: boolean;
   name?: string;
   workspace?: string;
   agentDir?: string;
-  model?: string;
+  model?: AgentModelConfig;
   memorySearch?: MemorySearchConfig;
   /** Human-like delay between block replies for this agent. */
   humanDelay?: HumanDelayConfig;
@@ -1618,7 +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

@@ -839,6 +839,15 @@ const ToolPolicySchema = z
   })
   .optional();
 
+const ToolProfileSchema = z
+  .union([
+    z.literal("minimal"),
+    z.literal("coding"),
+    z.literal("messaging"),
+    z.literal("full"),
+  ])
+  .optional();
+
 // Provider docking: allowlists keyed by provider id (no schema updates when adding providers).
 const ElevatedAllowFromSchema = z
   .record(z.string(), z.array(z.union([z.string(), z.number()])))
@@ -868,6 +877,7 @@ const AgentSandboxSchema = z
 
 const AgentToolsSchema = z
   .object({
+    profile: ToolProfileSchema,
     allow: z.array(z.string()).optional(),
     deny: z.array(z.string()).optional(),
     elevated: z
@@ -932,14 +942,20 @@ const MemorySearchSchema = z
       .optional(),
   })
   .optional();
-
+const AgentModelSchema = z.union([
+  z.string(),
+  z.object({
+    primary: z.string().optional(),
+    fallbacks: z.array(z.string()).optional(),
+  }),
+]);
 const AgentEntrySchema = z.object({
   id: z.string(),
   default: z.boolean().optional(),
   name: z.string().optional(),
   workspace: z.string().optional(),
   agentDir: z.string().optional(),
-  model: z.string().optional(),
+  model: AgentModelSchema.optional(),
   memorySearch: MemorySearchSchema,
   humanDelay: HumanDelaySchema.optional(),
   identity: IdentitySchema,
@@ -964,6 +980,7 @@ const AgentEntrySchema = z.object({
 
 const ToolsSchema = z
   .object({
+    profile: ToolProfileSchema,
     allow: z.array(z.string()).optional(),
     deny: z.array(z.string()).optional(),
     audio: z
@@ -1233,6 +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

@@ -1,6 +1,7 @@
 import crypto from "node:crypto";
 import {
   resolveAgentConfig,
+  resolveAgentModelFallbacksOverride,
   resolveAgentWorkspaceDir,
   resolveDefaultAgentId,
 } from "../agents/agent-scope.js";
@@ -31,7 +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 {
@@ -366,6 +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,
@@ -449,6 +459,10 @@ export async function runCronIsolatedAgentTurn(params: {
       cfg: cfgWithAgentDefaults,
       provider,
       model,
+      fallbacksOverride: resolveAgentModelFallbacksOverride(
+        params.cfg,
+        agentId,
+      ),
       run: (providerOverride, modelOverride) => {
         if (isCliProvider(providerOverride, cfgWithAgentDefaults)) {
           const cliSessionId = getCliSessionId(

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/providers/plugins/actions/discord.ts

@@ -13,6 +13,14 @@ import type {
 
 const providerId = "discord";
 
+function readParentIdParam(
+  params: Record<string, unknown>,
+): string | null | undefined {
+  if (params.clearParent === true) return null;
+  if (params.parentId === null) return null;
+  return readStringParam(params, "parentId");
+}
+
 export const discordMessageActions: ProviderMessageActionAdapter = {
   listActions: ({ cfg }) => {
     const accounts = listEnabledDiscordAccounts(cfg).filter(
@@ -462,8 +470,7 @@ export const discordMessageActions: ProviderMessageActionAdapter = {
       const guildId = readStringParam(params, "guildId", { required: true });
       const name = readStringParam(params, "name", { required: true });
       const type = readNumberParam(params, "type", { integer: true });
-      const parentId =
+      const parentId = readParentIdParam(params);
-        params.parentId === null ? null : readStringParam(params, "parentId");
       const topic = readStringParam(params, "topic");
       const position = readNumberParam(params, "position", { integer: true });
       const nsfw = typeof params.nsfw === "boolean" ? params.nsfw : undefined;
@@ -489,8 +496,7 @@ export const discordMessageActions: ProviderMessageActionAdapter = {
       const name = readStringParam(params, "name");
       const topic = readStringParam(params, "topic");
       const position = readNumberParam(params, "position", { integer: true });
-      const parentId =
+      const parentId = readParentIdParam(params);
-        params.parentId === null ? null : readStringParam(params, "parentId");
       const nsfw = typeof params.nsfw === "boolean" ? params.nsfw : undefined;
       const rateLimitPerUser = readNumberParam(params, "rateLimitPerUser", {
         integer: true,
@@ -525,8 +531,7 @@ export const discordMessageActions: ProviderMessageActionAdapter = {
       const channelId = readStringParam(params, "channelId", {
         required: true,
       });
-      const parentId =
+      const parentId = readParentIdParam(params);
-        params.parentId === null ? null : readStringParam(params, "parentId");
       const position = readNumberParam(params, "position", { integer: true });
       return await handleDiscordAction(
         {

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