* feat(fleet): add openclaw fleet cell supervisor for multi-tenant hosting
* test(fleet): cover cell profile, registry, containers, and service flows
* docs(fleet): document multi-tenant hosting and the fleet CLI
* fix(fleet): verify upgrade replacements and release foreign-collision reservations
* fix(fleet): gate upgrade commit on replacement health
* docs(fleet): mark fleet experimental and pin its single-host scope
* test(fleet): make incomplete-profile casts explicit for test-type lane
* fix(state): regenerate kysely schema artifacts after rebase conflict
* feat: node-hosted plugins — dynamic tools, MCP servers, and skills
Nodes become declarative plugin hosts:
- node.pluginTools.update: node hosts publish plugin-registered agent tool
descriptors; gateway materializes them as agent tools executing via
node.invoke under the node command allowlist, with tools.effective
invalidation and node online/offline removal.
- Trusted paired-node descriptors: no gateway-side plugin registration
required; gateway.nodes.pluginTools.enabled off-switch (default on);
description/count caps; deterministic node-prefixed collision names.
- Declarative node-hosted MCP: nodeHost.mcp.servers (McpServerConfig shape)
starts MCP clients on the node host, publishes tools as pluginId node-mcp,
executes via built-in mcp.tools.call.v1 with per-layer timeouts, failure
isolation, and orphan-safe shutdown. No re-pairing when servers change.
- Node-hosted skills: node.skills.update publishes ~/.openclaw/skills
content (64 skills/64KB/512KB caps both sides); gateway merges them into
the skills snapshot while connected and exec host=node is available, with
node:// locators, node-prefixed collisions, disabled command dispatch,
and gateway.nodes.skills.enabled + nodeHost.skills.enabled switches.
- Security: node-supplied pluginIds cannot satisfy pluginId-scoped tool
allowlists unless gateway-registered; reserved node-mcp id requires the
core MCP descriptor shape; protocol registry kept out of public
plugin-sdk dts.
- E2E: pond harness proves publication, MCP round-trip, skills locator, and
disconnect/reconnect for all three surfaces.
* style: format node-plugin-tools test
* fix(skills): keep status loader unfiltered when eligibility is passed
skills.status started passing eligibility for the node-skill merge, which
flipped loadWorkspaceSkillEntries into filtered mode and dropped disabled
skills from status reports (QA plugin-lifecycle-hot-reload timeout). Status
now merges node skills explicitly around an unfiltered load. Also: regen
docs_map for new node docs sections; add the intentional node-host MCP
onclose suppression to the lint-suppression allowlist.
* feat(webui): reintroduce opt-in AI purpose titles for tool calls
Restores the chat.toolTitles path removed in #103821, gated behind the new
gateway.controlUi.toolTitles opt-in (default false) so tool rendering stays
fully deterministic with no background model calls unless an operator enables
it. Disabled gateways answer { titles: {}, disabled: true } without loading
the completion runtime, and clients stop asking for the session.
When enabled, titles use canonical utility-model routing: an explicit
utilityModel (operator-chosen provider, like every utility task), else the
session provider's declared small-model default, honoring per-session model
overrides and auth profiles; utilityModel "" disables titles and malformed
refs fail closed — never the primary model. Tool inputs are redacted with the
tools-mode redactor before cache keys or prompts, caller ids are bounded and
never reach the model, and results cache in the per-agent SQLite
cache_entries so repeat views never re-bill.
Also completes two crestodian model-input mock factories that leaked into
sibling tests under shared-registry CI shards.
Fixes#103987
* fix(webui): redact tool-title inputs before truncation
* fix(anthropic): delegate adaptive CLI effort
Strip static --effort args for adaptive runs so Claude Code resolves effort from its environment, settings, and model default. Preserve configured effort for off or absent thinking and replace it only for concrete OpenClaw levels.
Fixes#103245
Co-authored-by: Dan Korotin <korotin.daniil@gmail.com>
* fix(anthropic): make effort dispatch exhaustive
---------
Co-authored-by: Dan Korotin <korotin.daniil@gmail.com>
* feat(agents): derive a provider-declared default utility model when unset
When agents.defaults.utilityModel is not set, utility tasks (titles, progress
narration) now use the primary provider's declared small model
(modelCatalog.providers.<id>.defaultUtilityModel: OpenAI -> gpt-5.6-luna,
Anthropic -> claude-haiku-4-5). Auth is inherent because derivation follows
the agent's primary provider. Setting utilityModel to an empty string
disables utility routing entirely; narration turns on automatically when a
default resolves and stays off otherwise.
Formatting/lint/tests verified on Testbox (oxfmt, oxlint, plugins:inventory,
docs:map, import-cycles, check:test-types, 4 test files).
* fix(ai): drop the temperature parameter for models that reject it
The GPT-5.6 family 400s on temperature via the Responses API (live-verified:
gpt-5.6-luna/-terra reject it; gpt-5.5 and gpt-5.4-mini/nano accept it).
supportsOpenAITemperature gates all three OpenAI payload builders, with a
catalog compat override (compat.supportsTemperature) declared for the 5.6
family in the openai manifest. Without this, utility tasks (titles,
narration) would fail once gpt-5.6-luna becomes the derived default.
Gates on Testbox: oxfmt, oxlint, plugins:inventory, import-cycles,
check:test-types, 6 test files. Live proof on Testbox: gpt-5.6-luna and
claude-haiku-4-5 one-shot completions succeed with temperature requested.
* fix(agents): carry the primary model's auth profile onto the derived utility default
A primary like openai/gpt-5.5@work previously reached utility tasks via the
profiled fallback; the derived default shares the provider, so its trailing
auth profile carries over instead of silently switching to default
credentials (Codex review). Gates on Testbox: oxfmt, check:test-types, tests.
* docs: realign the manifest provider-fields table after adding defaultUtilityModel
Kind-aware tool rows (terminal-style commands with wrapper stripping and
display highlighting, file edits with inline numbered diffs and diffstat,
write previews, key-value args), aggregate group summaries with live run
status, and a new batched chat.toolTitles gateway RPC that titles complex
calls via the configured utilityModel or the OpenAI Luna default (gated to
OpenAI-primary agents, cached in the per-agent SQLite cache_entries).
Also fixes two transcript pairing bugs: result blocks now inherit call
id/name/details at merge time, and results pair with any open call in the
current tool run so parallel calls render as single rows.
Fixes#103554
* feat(gateway): session worktree targeting, branch listing, and group catalog
- sessions.create accepts worktreeBaseRef/worktreeName (write scope) and
execNode (admin); worktree binding persists on the session entry as
worktree { id, branch, repoRoot } and projects onto session rows with
execNode so UIs can show checkout/branch/node state
- new worktrees.branches RPC lists local/remote branches (no fetch) for
base-ref pickers; worktrees.remove now reports snapshotError
- sessions.delete reports preserved dirty checkouts as worktreePreserved
- gateway-owned session group catalog in the shared state DB with
sessions.groups.list/put/rename/delete; rename/delete update member
categories server-side without bumping recency; sessions.patch absorbs
ad-hoc categories into the catalog
- group session display names prefer the human chat title (subject or
space #channel) over stored compact tokens
Part of #103431
* fix(gateway): route group category updates through the session accessor and regen protocol models
- bulk member-category rename/clear uses applySessionEntryReplacements
instead of legacy updateSessionStore call sites (session accessor
boundary guard)
- regenerate Swift GatewayModels for the worktrees.branches schemas,
snapshotError, and new sessions.create params
Part of #103431
* fix(gateway): resolvable remote branch refs and workspace-scoped branch listing
- worktrees.branches returns remote-only branches remote-qualified
(origin/feature-a) so every advertised name works as a worktree base ref
- write-scoped worktrees.branches callers are limited to configured agent
workspaces; other host paths require operator.admin, matching the
sessions.create cwd bar
Part of #103431
* fix(gateway): guard worktree name reuse by owner and env-scope group transactions
- managedWorktrees.create rejects a caller-chosen name whose live or
restorable record belongs to a different owner, so write-scoped
sessions.create cannot bind a session into another session's or a
manual checkout
- session group catalog writes run their SQLite transaction on the same
env-scoped handle as their statements, keeping OPENCLAW_STATE_DIR
overrides atomic and away from the default state DB
Part of #103431
* feat(ui): add plugin catalog management
* feat(gateway): add plugins.uninstall and richer plugin catalog metadata
Adds a plugins.uninstall gateway method (operator.admin, control-plane write)
backed by a lock-guarded uninstallManagedPlugin that mirrors the CLI flow:
config cleanup, install-record removal, managed file deletion, and registry
refresh. Bundled plugins stay disable-only. Catalog entries now carry a
manifest-derived category and a removable flag; ClawHub search results expose
download counts and verification tiers.
* feat(ui): redesign plugins page with inventory, store shelves, and cover art
Rebuilds /settings/plugins around three tabs: Installed (category-grouped
inventory with overview stats, state filters, uninstall for external plugins,
and inline MCP server management through the shared config seam), Discover
(featured/official shelves plus one-click MCP connectors and curated ClawHub
searches), and ClawHub (search with download counts and verification badges).
Every catalog entry renders bundled cover art or a deterministic gradient
monogram tile - no more empty boxes. Artwork generated with Codex CLI, shipped
as 512px WebP under ui/public/plugin-art.
* chore(ui): regenerate locale bundles for plugins manager strings
* docs: describe plugins manager tabs, uninstall, and MCP connectors
* fix(plugins): human catalog labels and un-pinned hosted fallback ids
listManagedPlugins now prefers manifest names over registry package-name
backfill, falls back to channel catalog labels and blurbs, and stops pinning
expectedPluginId when a hosted feed entry only exposes its package name
(which rejected every legitimate install of that package). Found via live
gateway testing against ClawHub.
* fix(ui): send minimal RFC 7396 merge patches for MCP server edits
config.patch merges rather than replaces, so key removal needs an explicit
null; sending the full config back made MCP server removal a no-op. Found
via live gateway testing.
* fix(ui): write explicit MCP transports for URL servers
The MCP runtime defaults URL-only servers to SSE, so streamable HTTP
endpoints saved by the add form or connector templates would fail at
connect time. Connector templates now declare their transport and the
add form infers streamable-http unless the URL follows the /sse
convention. Flagged by autoreview against the transport resolver.
* test(ui): wait for deferred plugin requests before resolving in e2e
* feat(ui): plugins detail view, action menus, and unified ClawHub search
Reworks the plugins page from PR #103176 feedback: merges the ClawHub tab into
Discover (typing searches ClawHub inline and appends a quiet From ClawHub
section, with Browse ClawHub demoted to a header text link), makes every row and
store card open a plugin detail overlay (hero art, primary enable/install
action, metadata table), and replaces enable/disable switches with a state chip
plus an overflow menu (Enable/Disable, Remove for external plugins, View
details) matching the ChatGPT-store install+menu pattern. MCP rows use the same
menu; refresh is now icon-only.
* chore(ui): regenerate locale bundles for plugins UI iteration
* feat(ui): vetted, grouped connector catalog for the plugins store
Expands Connect your world to 28 connectors organized into use-case shelves
(Work & productivity, Coding & infrastructure, Home & media, Everyday life).
Every entry passed a three-stage subagent review: official-docs verification
plus live endpoint probes for MCP servers, ClawHub result-quality and
malware/typosquat screening for curated searches, and an adversarial pass
that dynamically registered OAuth clients to prove one-click viability.
That review removed Figma (registration allowlisted, 403) and Atlassian
(OAuth issuer-mismatch bug upstream), downgraded GitHub to PAT-based setup
(no dynamic client registration upstream), fixed Linear (/sse retired) and
Home Assistant (/api/mcp, streamable HTTP) endpoints, retargeted poisoned or
dead searches (youtube, finance, hue dropped; calendar -> google calendar;
stocks replaces finance), and added Todoist, Airtable, Canva, Stripe,
Context7, DeepWiki, Hugging Face one-click MCP servers plus Jira, PDF,
transcription, Kubernetes, Reddit, maps, translation, and notes searches.
Keyless servers get a ready-to-use success message; new cover art included.
* chore(ui): regenerate locale bundles for connector groups
* fix(plugins): suppress hosted catalog rows once their package is installed
Hosted feed entries without a declared runtime id fall back to their package
name as catalog id, which never matches the installed runtime id, so the
Discover shelf kept offering an already-installed package. Installed package
names now also suppress official rows. Flagged by autoreview.
* fix(plugins): pin declared runtime ids and surface connector errors in place
The runtime-id pin now keys off explicitly declared catalog ids (plugin,
channel, or provider) instead of string-comparing against the package name,
so declared ids that equal their package name stay enforced while entry-id
fallbacks stay unpinned. Connector add failures on Discover now render on the
triggering card instead of the Installed tab's MCP section. Both flagged by
autoreview; regression tests included.
* feat(ui): full inventory artwork, pulse header, and two-column plugin list
Every bundled plugin now ships distinctive cover art (113 new Codex CLI
illustrations; 172 total, ~2.1MB WebP), so inventory rows and detail views
never fall back to monogram tiles. The four stat cards give way to a compact
inventory pulse: a segmented enabled/disabled/issues meter whose legend and
counts live inside the filter chips. Inventory, MCP, and search rows flow
into two columns when the panel is wide enough.
* chore(ui): regenerate locale bundles for pulse header
* fix(ui): omit stdio args from the MCP server row target
Stdio MCP args routinely carry tokens, and the inventory is visible to
read-only operators; mirror the config page and show only the command.
Flagged by autoreview; regression test included.
* fix(merge): point crestodian setup at relocated plugin commit/refresh modules
* fix(merge): add bootstrapToken to plugins page test gateway harness
* fix(plugins): name catalog install-action branches so Swift emits the union
* fix(ui): satisfy strict lint on plugins page form parsing and mocks
* chore(build): regen docs map, raise plugin-sdk declaration budget for new protocol surface
* fix(ui): type the plugins page patch mock with its real call signature
Paired devices can now carry a durable operator-assigned label: device.pair.rename { deviceId, label } (schema-bounded, admin/ownership-gated) stores an operatorLabel that persists in the shared SQLite state DB, survives device repair and re-approval, and takes display precedence over the client-reported name in CLI devices list and the Control UI inventory (operatorLabel, then displayName, then clientId, then deviceId). The label was previously dropped on write because the pairing store had no column for it. CLI: openclaw devices rename --device <id> --name <label>. Docs cover the command and precedence.
Fixes#13870
Thanks to @bladin for the contribution.
Co-authored-by: heichl_xydigit <1740879+bladin@users.noreply.github.com>
Co-authored-by: Ayaan Zaidi <hi@obviy.us>
In messages.visibleReplies "message_tool" sessions, a successful agent turn that produced a substantive private final without calling message(action=send) previously left the user with silence and only an operator log. The gateway now enqueues one protected front-of-queue retry prompting the model to deliver the reply, and falls back to a sanitized visible diagnostic when the retry cannot be enqueued or also strands. Queue overflow protection is unified with the in-flight-aware drop policy (skip in-flight or protected items, reject when nothing is droppable), rejected overflow no longer refreshes the drain debounce, heartbeat turns are excluded from recovery, and recovery retries no longer share the client turn's queued-turn lifecycle.
Fixes#85714
Thanks to Eva (@100yenadmin) for the contribution.
* feat(channels): narrated progress drafts + activity receipt on the final answer
Progress mode replaces raw tool lines with short utility-model narration of
what the agent is doing (streaming.progress.narration, default on, requires
an explicit utilityModel). On Discord the final answer now carries the
-# activity receipt and the working draft is deleted once the answer lands,
so busy channels keep no orphaned tool log above the reply.
Formatting verified remotely via Testbox oxfmt --check (hook bypassed:
no node_modules in this worktree).
* fix(channels): keep narration toggle independent of channel-default stream mode
Discord resolves its own progress default, so the resolver must not re-derive
mode with the generic partial fallback (narration was off for unset config).
* fix(auto-reply): honor status-only command text in narration model input
streaming.progress.commandText: "status" hides raw exec/bash text from the
channel draft; narration input now mirrors that policy so the utility model
never receives more command detail than the draft shows (Codex review P2).
Formatting verified remotely via Testbox oxfmt --check.
* fix(auto-reply): share the draft's command-tool set for narration and regen channel metadata
Reuse isCommandToolName (exec|shell|bash) so narration's commandText policy
matches the draft formatter exactly, and regenerate bundled channel config
metadata for the new streaming.progress.narration key (Codex review round 2).
Gates verified on Testbox: config:channels:check, oxfmt --check, 4 test shards.
* fix(channels): clear stale narration when the narrator stops mid-turn
An empty narration update now falls the draft back to raw tool lines, and the
narrator emits that clear when it disables after consecutive failures or the
per-turn cap, so drafts never pin stale status text (Codex review round 3).
Gates verified on Testbox: oxfmt --check + 4 test shards.
* chore(config): regen bundled channel metadata after rebase onto main
* chore: CI fixups — lint nits, test harness types, docs map, SDK surface budget
Two deliberate public SDK additions (resolveChannelStreamingProgressNarration,
isCommandToolName via the streaming wildcard re-export) bump the pinned
public-surface budgets to current counts (exports 10488, callable 5235).
Gates verified on Testbox: oxfmt, targeted oxlint, docs:map:check,
config:channels:check, check:test-types, 5 test shards.
* refactor(pairing): move device pairing store to shared SQLite state DB
Device pairing, pending requests, and bootstrap tokens now live in the
device_pairing_* / device_bootstrap_tokens tables of state/openclaw.sqlite
instead of devices/{paired,pending,bootstrap}.json. Gateways import legacy
paired records once at startup (before the node-surface fold) and archive
the JSON files with a .migrated suffix; transient pending/bootstrap rows
are dropped. The unshipped node_pairing_* tables are removed from the
schema and dropped from existing DBs. Doctor now flags un-imported legacy
store files instead of corrupt-JSON reads.
* refactor(pairing): drop stale awaits now that store persistence is synchronous
* refactor(pairing): extract leaf record types to break store/domain module cycle
Adds client-capability-gated tool availability: gateway clients declare
capabilities at connect (new inline-widgets cap), chat.send stamps them into
the run context, and every tool assembly path (embedded runner, queued
followups, Codex app-server harness, plugin-only construction plans) drops
tools whose requiredClientCaps the originating client did not declare. The
Canvas plugin ships the first such tool, show_widget: agents pass SVG or an
HTML fragment plus a title; the plugin hosts it as a bounded, retention-scoped
Canvas document and returns the existing canvas preview handle, which web chat
renders as a sandboxed iframe fitted to the widget's reported content height.
Widget frames never get allow-same-origin (per-preview sandbox ceiling,
including the sidebar path) and the Canvas host serves widget documents with a
CSP sandbox header so direct navigation runs in an opaque origin. Verified
live end-to-end on a Testbox with gpt-5.5 (screenshots on the PR). CLI-backed
model backends do not carry client caps yet and stay fail-closed (#102577).
Closes#101790
Summary:
- audit Doctor lint default selection after the full lint-family backfill
- make legacy state, skills readiness, session transcripts, and session snapshots explicit-only for default lint
- document default lint vs --all/--only and keep plugin/SDK public contracts unchanged
Validation:
- Galin review: no blocking findings
- maintainer accepted the default-lint compatibility tradeoff in PR comment
- exact-head hosted gates passed for d5d88a0db1: CI#28976811444 and Workflow Sanity#28976811343
- local broad pnpm check was blocked by a shrinkwrap guard failure that reproduces on origin/main and is unrelated to this five-file Doctor diff
Co-authored-by: Gio Della-Libera <235387111+giodl73-repo@users.noreply.github.com>
* feat(secrets): resolve SecretRef model credentials at egress via process-local sentinels
SecretRef-managed model-provider credentials now travel as opaque
oc-sent-v1 sentinels through auth storage, stream options, and SDK
config; the guarded model fetch injects real values into headers and
URLs immediately before the SSRF-guarded send and fails closed on
unknown sentinels. packages/ai adapters converge on the host guarded
fetch where the SDK supports custom fetch and unwrap at construction
where it does not. Resolved values (and their percent-encoded forms)
register for exact-value log redaction. Kill switch:
OPENCLAW_SECRET_SENTINELS=off. Also fixes a pre-existing unhandled
rejection race in capNonOkResponseBodyLazily (pipeThrough writer leak).
* test(plugin-sdk): update public surface budget
* fix: show exported tool results in trace viewers
* fix(diagnostics-otel): emit semconv response key and execute_tool identity so trace viewers show tool results
---------
Co-authored-by: Alex Knight <15041791+amknight@users.noreply.github.com>
