* perf(state): cap the per-agent SQLite handle cache with LRU eviction
Multi-tenant hosts open one WAL database per agent (~3 file descriptors
each); the process-local handle cache was unbounded, so large fleets
exhausted descriptors. Cache hits now refresh LRU recency, and cache-miss
opens evict the oldest non-transactional handle before constructing the
new one, capped at 64 open handles. Eviction closes the process-local
handle only; registry rows and durable data are untouched and evicted
databases reopen transparently on next access.
* fix(state): fleet-wide introspection reads stop opening writable agent databases
Live testing an 81-agent gateway surfaced three full-fleet sweeps that
opened every agent's SQLite database writable (schema ensure + registry
write transaction per open): the 60s health snapshot's per-agent session
summaries, usage-cost cache reads behind the gateway usage endpoints, and
the zalouser doctor detector that ran from every CLI startup. Each sweep
churned the new bounded handle cache and saturated the event loop.
Session listing gains a readonly non-registering variant (SDK exposes it
as an additive readOnly flag on listSessionEntries); health and usage
reads use it, with transient-lock tolerance owned by the health caller.
The zalouser detector gates on the channel/credentials actually existing,
collects once per pass, and detection reads go readonly while migration
apply stays writable. The usage refresh queue's flat 50ms busy retry
becomes exponential backoff capped at 5s, and refreshing summaries retain
their cache timestamp so the 30s TTL gates fleet rescans.
* perf(state): make per-agent handle eviction and reopen cheap
Eviction only works if reopen is cheap. Reopens after eviction previously
repeated full first-open work; ensureOpenClawAgentSchema takes BEGIN
IMMEDIATE and the registry upsert writes the shared state DB, so reopens
blocked in synchronous busy waits (observed 14-22s event-loop stalls with
reconcile workers holding write transactions). Owner/schema validation and
registration now run once per path per process with invalidation on
quarantine and disposal; the read-only integrity guard still runs on every
physical open. Eviction closes with a PASSIVE WAL checkpoint instead of
TRUNCATE, which waits on readers; orderly dispose keeps TRUNCATE so
sidecars are flushed before unlink.
* ci: regenerate plugin-SDK API baseline, drop prod test-seam exports, mock readonly accessor
The additive readOnly flag on the SDK listSessionEntries changed the
declaration surface, so the generated API baseline is regenerated via the
sanctioned script. Knip forbids production exports consumed only by tests:
the usage-cache testApi export is removed outright and the refresh-queue
seam moves behind an env-gated global exposed by a test-support module.
health.plugins.test's full session-accessor mock gains the newly imported
readonly listing.
* feat(channels): batch 1 producers drop media placeholder bodies
Media-placeholder program batch 1: Google Chat, Zalo, LINE, and
Mattermost stop minting <media:kind> placeholder bodies. Media-only
messages carry an empty caption plus one structured fact per native
attachment (type-only when a download fails or is rejected, so payload
positions and kind signals stay aligned). The shared
formatMediaPlaceholderText SDK formatter renders text-only carriers
(Mattermost pending-room lines) from structured facts; per-channel
placeholder builders and the expected-count side channel are deleted.
* fix(mattermost): satisfy type, deadcode, and SDK manifest gates
* feat: add session discussion panel seam
* fix: keep discussion iframe cookie-capable and changelog release-owned
* test: cover cookie-capable discussion iframe sandbox
* fix: stretch discussion panel host so the embed fills the rail
* fix: keep provider failures retryable and probe discussion availability before showing the action
* fix: block same-origin discussion embeds, show action on catalog sessions, close stale panel on reconnect
* fix: scope discussion probes and panel callbacks to the issuing connection
* fix: dedupe in-flight discussion probes per session
* fix: retry superseded discussion probes and key-scope panel results
* fix: regenerate Swift protocol models and extend advertised-method expectations
* style: format chat-pane-header
* fix: regenerate Kotlin protocol models and date discussion methods in the 2026.7 train
* chore: restore release-owned changelog to main state
* chore: keep changelog untouched relative to merge-base
* feat(browser): send pages to the main session from the Chrome extension
One-click page share in the OpenClaw Chrome extension: toolbar popup with an
optional note, page/selection context menu, and Alt+Shift+S. Capture is
selection-first with a readability heuristic, X/Twitter thread extraction, and
Google Docs plain-text export via the user's session cookies. Payloads ride the
existing paired relay WebSocket as a new pageShare message; the gateway-only
page-share sink wraps page text in the external-content safety boundary, then
enqueues a main-session system event and requests an immediate heartbeat
(hooks/wake semantics). Node-hosted relays report a clear unsupported error.
Capture heuristics adapted from Nat Eliason's MIT-licensed send-to-openclaw.
Co-authored-by: Codex <codex@openai.com>
* fix(browser): keep page-controlled metadata inside the share safety boundary
Review findings: move title/URL inside wrapExternalContent (a hostile <title>
must not become trusted header text), prefer the user's selection over the
full Google Docs export, and pass the context-menu selectionText through so
iframe selections and selections cleared during relay reconnect still win.
* fix(browser): bind context-menu shares to the click-time document
Selection shares from the context menu now send the click snapshot directly
(no recapture), so navigations during relay reconnect cannot mislabel the
source and iframe selections are preserved. The Google Docs selection probe
scans all accessible frames before falling back to the full-document export.
* test(browser): expect the page-share handler in relay server args
* fix(browser): probe only the main frame for Google Docs selections
All-frame injection rejects wholesale when one frame is inaccessible and
returns child frames in nondeterministic order, so the probe now reads the
main frame only. Child-frame selections still share correctly through the
context menu's click-time selectionText; toolbar/shortcut entry sends the
full page for that case (named tradeoff in the code comment).
* fix(browser): satisfy page-share CI gates
---------
Co-authored-by: Codex <codex@openai.com>
* fix(gateway): emit config.changed from the reload committer for every accepted write
Agent config_set, CLI, and doctor writes reach the gateway through the file
watcher's reload path, which never broadcast config.changed - only direct RPC
writes did. Move the broadcast to a single onConfigCandidateCommitted hook in
the reload committer so all writers notify connected UIs, including
runtime-skipped commits (writer-intent echo suppression, reload mode off),
and delete the duplicate RPC-side broadcasts.
* feat(ui): sync chat follow-up and commentary prefs across devices
Add chatPersistCommentary and chatFollowUpMode to the ui.prefs sync surface
(schema, types, docs) with clearable null removal for the follow-up override.
Rewrite server-prefs around a descriptor table so a synced pref is one entry
instead of five hand-written code sites. Caption synced rows/sections
(chat prefs, theme, language) and refresh the stale Settings subtitle.
* refactor(ui): dedupe settings constants, drop dead composer branch, refresh copy and docs
Derive the Security tool-profile options from PROFILE_OPTIONS and General's
thinking levels from BASE_THINKING_LEVELS instead of parallel literals.
Remove the unused composer mode:"model" branch, the orphaned
.config-view-toggle CSS family, and stale quick-settings wording. Add static
settings-search entries for the Sessions and Managed Worktrees pages plus a
provenance-row test, and update docs/cli/openclaw.md to describe the current
system-agent config-write policy (denylist roots, route-backing plugin guard,
per-agent routing escalations).
* chore: refresh plugin SDK API baseline after rebase onto current main
* fix(agents): skip runtime plugin reload when a registry is already active
ensureRuntimePluginsLoaded rebuilt and activated a replacement plugin
registry on every call site invocation (agent turns, cron isolated-agent
runs, the gateway prewarm sidecar). Activating a replacement retires the
active registry and runs its plugin host cleanup, which cron.remove()s
persistent plugin-scheduled session turns before the replacement's async
re-registration catches up — silently deleting the cron row with only a
low-level 'cron service unavailable' WARN. Plugins using
api.session.workflow.scheduleSessionTurn for recurring jobs lose their
schedule on the next agent turn after boot.
Guard with getActiveRuntimePluginRegistry(): when a registry is already
active every caller's request is satisfied, so return before building a
replacement.
* fix(agents): make active-registry reuse compatibility-aware, not presence-only
The original guard (1de432a9) returned early whenever ANY registry was
active, without checking whether it was actually compatible with this
call's workspace, plugin scope, config, or runtime mode. A later call
requesting a different workspace could silently keep an unrelated
workspace's registry active instead of loading its own — a correctness
and workspace-isolation issue (caught in upstream review, PR #107752).
Fix: compute workspaceDir/startupPluginIds/loadOptions first, then check
getLoadedRuntimePluginRegistry({loadOptions, workspaceDir, requiredPluginIds})
before deciding to skip — reusing the existing loader cache-key
compatibility rules (workspace, config, runtime subagent mode, required
plugin IDs) that already govern normal cache-hit reuse elsewhere. Only
skips the reload when an active registry is genuinely compatible; still
loads a fresh one otherwise.
Tests: replaced the presence-only regression test with three cases —
compatible-registry reuse (skip), incompatible/different-workspace (still
loads), and startup-scoped plugin IDs passed through to the compatibility
check. All 5 acceptance criteria from the review verified locally:
runtime-plugins.test.ts, active-runtime-registry.test.ts,
standalone-runtime-registry-loader.test.ts, and the full cron lane
(33 files / 460 tests, including run.runtime-plugins.test.ts) all green;
pnpm build completes clean (5m14s, no errors).
* fix(plugins): preserve live scheduled turns
Scope dynamic scheduler cleanup to the retiring registry owner and keep runtime workflow APIs callable while their registry remains pinned and live.\n\nCo-authored-by: Dan Scarafoni <dan@poweredbyzon.com>
* test(plugins): prove recurring cron ownership
* chore(plugin-sdk): refresh api baseline
---------
Co-authored-by: Peter Steinberger <steipete@gmail.com>
* improve(diagnostics): surface run startup phases as session-correlated events
The embedded runner reports ordered startup milestones (workspace, auth,
context_engine, ..., model_call_started) through the private
onExecutionPhase callback, so they reach typing indicators and cron
watchdogs but never the diagnostic bus. External status surfaces that
already consume run.attempt / run.progress / model.call.* cannot see
where a turn is during startup — exactly the window that matters when a
run hangs before its first model call.
Wrap params.onExecutionPhase once at the runner entry so every phase
transition also emits a session-correlated run.execution_phase event
(runId, sessionId, sessionKey, closed EmbeddedAgentExecutionPhase +
provider/model/tool context). The event rides the async diagnostic lane
like the other high-frequency lifecycle events, so ordering against
model.call.* is preserved and no subscriber work runs on the hot path.
The wrapper tracks session rotation via onSessionIdChanged so
post-compaction phases carry the current session id. Stability records
project the milestone into their dedicated phase/provider/model fields.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* fix(diagnostics): preserve trusted phase observability
* chore(plugin-sdk): refresh diagnostic API baseline
* fix(diagnostics): publish execution phases to plugins
* refactor(logging): keep public diagnostic recorder path
* fix(plugin-sdk): keep deprecated budget exact
* fix(plugin-sdk): refresh diagnostic surface budget
* fix(diagnostics): preserve execution phase coverage
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Co-authored-by: Peter Steinberger <steipete@gmail.com>
* refactor(ui): restructure Settings IA around user-facing pages
Settings sidebar shrinks to settings-only destinations grouped by attention:
General/Appearance/Notifications up top, Connections, Agents & Tools,
Privacy & Security, System. Sessions, Worktrees, Activity, and Memory
Import leave the takeover: Sessions/Activity return to the workspace
sidebar, Worktrees becomes a Sessions hub tab, Memory Import is reached
from the Agents memory panel. Model Setup becomes a Models subpage.
New config pages: Notifications (web push), Privacy & Security (curated
policy rows + security/approvals schema sections), Advanced (uncurated
schema sections + raw editor, replacing the General quick/advanced
toggle). General slims to model defaults, language, and gateway host;
identity moves to Profile, lobster appearance rows to Appearance.
* feat(agents): widen system-agent config writes to operator parity
Every config surface the Control UI can edit is now agent-writable behind
the exact-operation human approval gate, except a small documented
denylist ($include, auth, env, models, secrets). tools.* and
plugins.entries.* writes are allowed; plugin_uninstall works for plugins
that provably do not back the active default inference route (fail closed
otherwise); set_default_model takes an optional agentId and live-tests a
per-agent model with the same rigor as the default route. Per-agent
routing fields on non-default agents are direct-writable; the default
agent's route keeps the verified set_default_model/onboard paths. A new
config-write-parity contract test forces explicit classification of every
top-level config key.
* feat(ui): sync operator display prefs through config ui.prefs
Theme, theme mode, text scale, locale, and chat display prefs get a
canonical server-side home in config ui.prefs. The Control UI mirrors
them into localStorage for instant boot, writes local changes through to
the gateway (best-effort; viewer-scope or offline clients stay
device-local), and applies server-side deltas on connect and on every
config snapshot reload — so an agent changing the theme through the
approval gate takes effect in the UI, and prefs follow the operator
across devices.
* chore: prune quick.ts from max-lines baseline after slim-down
* style: oxfmt changed files
* chore: regenerate plugin-sdk api baseline for ui.prefs config surface
* fix: dedupe runPluginUninstall dep type and narrow locale extraction
* docs: update Control UI settings structure and ui.prefs sync
* fix: address autoreview findings on parity guards and prefs sync
- plugins.entries.* config writes get the same active-route ownership
check as plugin_uninstall (fail closed when unprovable)
- system-agent prompt now describes the guarded write policy instead of
the old blanket refusals
- per-agent set_default_model logs the agent-scoped label
- legacy /config?section=... deep links redirect to the Advanced page
- prefs sync: pushes coalesce and drain serially (no CAS races dropping
updates), replaced config hashes mark pre-patch snapshots as stale, and
the last-seen server value persists per gateway scope so reloads cannot
revert unpushable viewer-scope edits
- lint: theme membership checks use Sets
* fix: harden uninstall route guard and scope prefs queue per gateway
- plugin uninstall ownership now covers execution owners (runtime policy,
embedded harness override) in addition to the provider, and re-verifies
inside the commit boundary so a concurrent route switch cannot slip a
destructive uninstall past the pre-approval check
- prefs push queue binds to one gateway client; switching gateways drops
the old queue (device-local fallback) instead of cross-writing prefs
* style: top-level type import in config-page test
* fix: close review races in config-write guards and prefs reconciliation
- config_set/config_set_ref re-verify the inference-route guard inside
the commit boundary so conditional verdicts (per-agent routing, plugin
entries) cannot be flipped by a concurrent edit after approval
- server prefs apply per changed field, so a server change to one pref
no longer reverts unpushable local edits of the others
- replaced-hash staleness retires once post-patch state is observed, so
a genuine restore of the pre-patch config becomes authoritative again
- moved-section deep links (communications notifications, automation
approvals) redirect to their new pages
* style: satisfy no-useless-assignment and loop-condition lints in prefs sync
* fix(ci): resolve unhandled profile-test rejections and knip unused exports
The profile page's new identity section reads context.config; the test
context mock now provides it, eliminating the unhandled rejections that
failed checks-ui and the compact-large shard. Exports flagged by the
dependency gate (extract/patch helpers, section-key and prop types) go
module-private; extraction coverage moves behind applyServerUiPrefs.
* refactor(agents): move config-write policy to its own module
The static path classifier and documented denylist move to
config-write-policy.ts, shared by the execution guard and the parity
contract test — no test-only exports left for the dependency gate to
flag. Blocked-root refusals now cite the denylist's documented
escalation for that root.
Marcus Castro