* feat(cli): openclaw resume attaches the TUI to a recent session
Implements docs/plan/runners.md milestone 2.
* fix(cli): resume covers global sessions and preflights TTY
* fix(cli): require a terminal before resume
* fix(cli): adapt resume gateway disconnect state
* fix(cli): preserve global resume ownership
* test(cli): prove qualified global resume crosses the gateway boundary
Assign queue-owned opaque IDs to modern system-event snapshots so copied stale snapshots cannot acknowledge a later identical event. Keep structural matching only for shipped legacy ID-less consumers.
* fix(macos): let AI setup proceed when the login keychain is unavailable
A missing/locked login keychain made GatewayActivationBindingKeyStore
unable to mint the activation binding key, and onboarding refused every
candidate and manual-key activation with 'Secure storage is unavailable'.
The fingerprint only protects the crash/relaunch resume receipt, so
degrade instead of refusing: write an ownerless pending record (still
matched exactly, never as a wildcard), keep the full activation-ambiguity
window, and skip only restart reconciliation, which needs fingerprint
proof by design. A relaunch then repeats activation rather than trusting
the receipt.
Reported by Peter Steinberger.
* fix(macos): render proper provider brand icons in onboarding AI setup
Nearly every provider icon URL the gateway sends is a simpleicons.org
SVG, which AsyncImage cannot decode, so the Connect your AI page fell
back to generic symbols for every tool and provider; the two GitHub
avatar PNGs that did render clashed as full-color rasters.
Bundle nine monochrome template marks (Simple Icons CC0; xAI from
LobeHub icons, MIT) and resolve them local-first by brand id/kind,
including composed choice ids like xai-oauth. Remote icons now load via
NSImage(data:), which decodes SVG, with vector payloads tinted as
templates so plugin-supplied icons blend too. Every glyph sits in a
uniform rounded well; install cards show host-only links, hover states,
and a softer translucent surface shared by all setup sections instead
of controlBackgroundColor.
Requested by Peter Steinberger.
* fix(macos): use failable UTF-8 decode in icon vector sniffing (swiftlint)
* fix(macos): address ClawSweeper review of onboarding icon/keychain changes
- Refuse relaunch handoff from ownerless (keychain-unavailable) completed
receipts at the reconciliation trust boundary; such receipts can belong
to replaced credentials, so setup repeats a fresh activation instead.
- Decode and pass the canonical brandId the gateway sends for candidates,
auth options, and manual providers so bundled marks resolve for opaque
choice ids.
- Scan the bounded XML prolog (comments, declarations, doctype) when
sniffing remote SVG payloads; comment-prefixed vectors now tint as
templates.
* refactor(macos): move setup error enum to support file (swiftlint file length)
* chore(i18n): refresh native source inventory
* fix(macos): give keychain-unavailable activations attempt-specific unbound leases
A nil owner made concurrent unbound attempts indistinguishable: a stale
attempt's delayed response could complete or clear a newer attempt's
record. Unbound attempts now mint a random per-attempt lease id with a
sentinel fingerprint — live matching stays attempt-exact, restart
reconciliation's fingerprint guard rejects them, and relaunch
verification refuses unbound receipts before any handoff.
* chore(macos): ship third-party artwork notices with provider icons
Simple Icons (CC0) attribution and the LobeHub Icons MIT license text
for the xAI mark now travel inside the bundled ProviderIcons directory.
* fix(ci): refresh plugin-sdk API manifest for private-type hashing
#120975 changed the baseline generator to hash private types but landed
without regenerating the manifest; the check-plugin-sdk-api-baseline
lane was gated off on that PR and first failed here. Regenerated via
pnpm plugin-sdk:api:gen with no SDK surface change in this branch.
Propagates the canonical dispatcher step from openclaw/clawsweeper#1083
byte-identically: the acknowledgement step waits 15s after a clean
marker check and rechecks immediately before posting, so near-simultaneous
opened and ready_for_review runs (draft PR marked ready within seconds,
as on #120974) post exactly one receipt ack.
* perf(plugins): close the last kysely closure chains and guard reachability transitively
Follow-up to #120698/#120811/#120882: the closure guard's enumerated barrel
bans cannot catch new heavy edges, and two closures still statically reached
kysely on main.
- guard: add a transitive kysely-reachability test that walks static value
imports from every doctor-contract and legacy-setup closure through plugin,
plugin-sdk, and relative core graphs, failing with the full import chain;
type-only and lazy dynamic imports stay allowed
- llm-task/model refs: manifest-model-id-normalization reads snapshots
through a registration-slot runtime bridge (snapshot modules register at
eval; require fallback covers cold processes) and
current-plugin-metadata-state moves its process-scoped facts onto a
globalThis singleton so dual module instances share published state
- telegram: split thread-bindings-store.ts (pure record shapes + legacy-file
readers) out of the acp-runtime-heavy manager, delete the consumer-less
testing export, move the pure bot-user-id token parse to
token-fingerprint.ts, and lazy-import token.js in the async update-offset
detector
llm-task enumeration drops to ~0.8s/157 modules cold; every closure is now
statically kysely-free and stays that way by construction.
* fix(telegram): repoint the native-command menu state at the token-fingerprint parser
Propagate the receipt-acknowledgment steps from the canonical
ClawSweeper dispatch template (openclaw/clawsweeper#1080): mint a
minimal issues:write App token and post an idempotent
clawsweeper-pr-ack marker comment for non-draft opened and
ready_for_review pull requests, before review dispatch.
Use channels.telegram.groups as the shared default when an account omits groups. Explicit account maps, including {}, remain full replacements. This keeps chat admission and sender restrictions on one policy path and fixes silent multi-account authorization failures.
Co-authored-by: Ayaan Zaidi <hi@obviy.us>
Abandon unexpected acceptance failures instead of leaking accepting slots. Reject non-Git or blank suggestions before side effects, and protect unseen pending suggestions ahead of accepted replay state.
* fix(doctor): migrate legacy agent databases discovered on disk and bound registry writes to the active state dir
* fix(doctor): discover configured agent databases without registry rows
* fix(doctor): preserve filesystem and configured database identity
* fix(doctor): prioritize configured agent database identity
* fix(doctor): prefer recorded agent database ownership
* fix(gateway): rebuild chat metadata when auth-profile snapshots change
chat.metadata cached a prepared generation built before the runtime
auth-profile store snapshot was published (empty-store fallback) and
generationFactsMatch never compared auth state, so the Control UI showed
"No models available" after a gateway restart until an unrelated config
edit. Capture per-agent auth snapshot revisions in the generation facts,
subscribe the metadata lifecycle to auth-store mutations, and run one
awaited revision-aware catch-up refresh after listener registration so
publications that precede registration are still observed.
* fix(gateway): bound cloud worker tunnel startup and surface dispatch failure detail
Live stress-testing the Control UI cloud flow found sessions.dispatch
hanging unbounded (observed 12+ min) when the worker SSH tunnel could
not connect: the runner's exited promise settled only on "close" (a
spawn error never settled it), the reconnect loop swallowed every
failure with no logging, the poisoned ready promise was re-handed to
every later dispatch, and dispatch errors dropped the actionable reason
recorded in worker_environments.last_error.
- settle exited on the real "exit" event, wire the owner abort signal
into spawn, bound stop()'s post-SIGKILL wait, and fail stop honestly
when termination is unconfirmed instead of fabricating an exit
- add a 60s per-attempt readiness deadline and log each failed connect
attempt (bounded, redacted); keep an unconfirmed child tracked and
wait for its real exit before retrying
- add a 3-minute epoch-fenced startTunnel deadline with a typed,
actionable error; detach its cleanup so the deadline holds
- append the bounded recorded reason to the five dispatch-visible
worker environment error messages (docs already promise these)
- remote socket setup: drop "--" from chmod (BSD/macOS chmod treats it
as a filename), which blocked every tunnel to a macOS worker host
- workspace quiescence: tolerate EPERM without crashing the protocol
while keeping unsignalable freeze targets counted as live so
quiescence fails closed
* test(gateway): type worker child kill mock
Replies, previews, and media in channel Direct Messages topics now remain in their originating topic. Bot-private and forum topic routing remains unchanged.
* perf(agents): keep turn-path model catalog reads off the full live build
First agent turns (embedded and cron) resolved thinking capability through
loadPreparedModelCatalogSnapshot without readOnly, which materialized the
full live model-runtime catalog: ambient synthetic-auth discovery fanned out
to every registered provider and loaded plugin discovery modules through
jiti source transform (3,172 TS modules, 36s event-loop block, +600MB heap,
58.7s model-selection on a cold gateway).
- add loadProviderScopedThinkingCatalog: manifest metadata first, then a
provider-scoped read-only static catalog, then scoped live discovery only
for runtime-discovery providers (preserves #116584 Ollama semantics)
- route scopedLiveProviderDiscovery through the scoped read-only loader
- scope live-mode ambient synthetic-auth refs to the requested providers
- bound the last-resort synthetic-auth sweep to discovery entry modules
- memoize per-turn plugin skill dir resolution/republish (single-slot,
lifecycle-cleared; was a full walk + symlink republish every turn)
Cold first turn 72.7s -> ~22s wall (remaining cost is provider prefill of
the ~19.5k-token default prompt); model-selection 58,726ms -> 124ms.
* test(agents): align model-catalog.runtime mocks with scoped thinking catalog seam
Explicit vi.mock factories must export every binding prod touches; the new
loadProviderScopedThinkingCatalog export is now mocked everywhere the module
is stubbed, and the live-model-switch Ollama hydration test asserts the new
provider-scoped seam instead of the retired unscoped snapshot call shape.
* test(agents): export scoped thinking catalog from every prepared-catalog mock; split synthetic-auth helpers
- add loadProviderScopedThinkingCatalog to all explicit prepared-model-catalog
and model-catalog.runtime mock factories (vi.mock factories must export every
binding prod touches)
- move synthetic-auth ref scoping/resolution into
prepared-model-runtime.synthetic-auth.ts; keeps facts under the max-lines cap
* test(agents): prove scoped thinking hydration for runtime-only models
Boundary proof for the ClawSweeper review gap: the three-tier helper stops at
manifest or scoped-static when they resolve, and runs provider-scoped live
discovery (no broad fanout) only for runtime-only models; cron selection
hydrates through the same scoped helper and skips it entirely for thinking=off.
* test(agents): accept rest args in scoped thinking catalog mocks
Peter Steinberger