* feat(sessions): stamp agent identity on spawned sessions and return spawn receipts
Agent-spawned sessions recorded the requesting session key as createdActor.id,
so the Control UI creator chip rendered an opaque key. Spawn producers now stamp
the canonical requester agent id; parent-authority validation moves to a new
trusted requesterSessionKey field. projectSessionActor enriches agent actors
with configured identity name/avatar at read time, and visible sessions_spawn
returns a sessionUrl + owner receipt with URL-first acknowledgement guidance.
* feat(sessions): assignable session ownership with owner facet and menus
GitHub-assignee-style ownership: sessions get a mutable owner (defaulting to
the immutable createdActor) stored in additive bare-nullable SQLite columns
with first-use lazy ensure. New operator.write sessions.assignOwner validates
targets, requires an identified caller, authorizes by session visibility, and
records assignedBy/assignedAt inside the write transaction. The sessions agent
tool gains assign_owner; the Control UI adds Assign-to-me/Assign-to menus in
sidebar rows and chat headers, renders the effective owner chip, and the
creator facet/filter now keys on effective owner. Sharing authority stays
anchored on createdActor.
* feat(sessions): record session participants and stack them in the owner chip
Records every distinct external prompter (human profile/channel sender, or a
requesting agent) per session in an additive session_participants table at the
turn-admission boundary — best-effort, deferred, never blocking the turn; the
session's own agent and viewers are never recorded, capped at 32 per session.
The session row projects a bounded participants list (owner excluded) plus a
total count with the same actor enrichment as owner/createdActor. The sidebar
chip becomes a pair-stack when others have prompted (owner front, one peeking
participant or +N behind), the chat header shows the full facepile, and an
authenticated involvingMe list filter adds an Involving-me sidebar predicate.
Participant projection is excluded from logical-session CAS equality so display
history never invalidates session writes.
* fix(sessions): identify built-in agent tool callers for owner assignment
The sessions tool's assign_owner dispatched through the in-process synthetic
client, which carries neither a signed agent-runtime identity nor a human
profile, so agent-initiated reassignment always failed with FORBIDDEN. The
tool now captures its trusted requester agent identity and carries it across
in-process dispatch as internal client state (never wire params); the handler
derives assignedBy as signed runtime identity, then trusted agent-tool caller,
then authenticated human. Live-verified end-to-end on a dev gateway.
* fix(ci): split oversized session modules and refresh prompt snapshots
Split the max-lines offenders at concept boundaries for session equality, tool overrides, and protocol owner schemas. Remove the redundant Number conversion from the node:sqlite participant count. Refresh prompt snapshots after drift from the sessions and sessions_spawn tool description updates.
* fix(ci): restore solo-mode chip suppression and conform new method descriptors
Solo-mode root cause: owner-assignment submenu options reused the permanent owner-chip custom element, so hidden menu avatars were counted as attribution chrome. Menus now use viewer avatars while gateway-gated owner chips remain exclusive to collaborative sessions.
Conform sessions.assignOwner to the 2026.8 descriptor and append-only advertised-method inventories, and regenerate the Swift and Kotlin protocol surfaces.
Keep historical v15/v14 fixtures frozen by stripping the new owner columns; the existing range already excludes the participant table. Replace the new raw SQLite schema probes with synchronous Kysely queries.
Clear max-lines by splitting the organizer host contract, pure agent-navigation projections, and ownership/filtering sidebar cases at their concept boundaries.
* fix(ci): integrate ownership series with latest main surfaces
Wire the sessions-page assign-owner action, merge capability imports, narrow the navigation export scope, and apply sessions-create formatting.
The owner-presence regression came from hidden assign-owner menu avatars emitting data-viewer-id, so owner and menu chrome now opt out of presence markers while real facepiles retain them.
* fix(sessions): scope the involving-me filter to profile-backed participants
Session participant history mixed channel-native sender ids with authenticated Gateway profile ids, so involving-me missed real sessions and could accept numeric collisions.
Record the actor_source namespace at each producer, carry it through the internal SQLite projection, and match authenticated viewers only against profile-backed human participants. Legacy NULL sources fail closed for filtering, while channel ids remain available for display.
* build(ui): raise startup budget baseline for session ownership surfaces
Ownership chips, assignment menus, and the participant stack add ~0.7 KiB
gzip to the startup path; CI compression landed just over the previous
baseline+tolerance. Hard cap (350 KiB) unchanged.
* refactor(sessions): drop raw NULL projection for the lazy actor_source column
The Kysely guardrail rejects typed raw sql snippets outside allowlisted
boundaries; select the lazily-ensured column only when present and let the
row projection treat its absence as unknown/legacy.
* build(ui): refresh combined startup baseline
`agentic-agents-core-models` carried the 36s scaled estimate while measuring
56.3s (n=6, p90 58.6s) across 260 compact jobs on 2026-08-16. Packed beside
`agentic-agents-core-runtime-hosted-1` (60.4s) it built the only bin running
>=1.25x its prediction: compact-large-19 ran 122s of work priced at 88s, which
made it the most frequent slowest job in the workflow.
Add the measured value to the hybrid hint map, which already exists for exactly
these Blacksmith-specific observations. The packer then separates the two, and
the tallest actual compact bin drops 122s -> 109s with no bin above 1.25x and
no change to row counts (48 push / 56 pull-request), so this costs no extra
runner registrations.
Two things deliberately left alone. The `agentic-gateway-core-3` 140s pin looks
like a 2x over-prediction against its 68.5s median, but run 31938297538 really
did take 138.0s (109.03s + 28.98s across its two configs) on a run whose fleet
slowdown factor was 0.98 -- a genuine tail on a healthy run, so unpinning it
would rebuild a >200s bin whenever the tail lands. Every other bin already sits
within 1.24x of its prediction.
The replaced guard pinned one bin arrangement (`runtime-hosted-1` not sharing a
job with `agents-core-tools`) with no stated failure mode, so any honest refit
broke it. It now asserts the property that comment was reaching for -- both
weight sources survive rebalancing under the body ceiling -- plus the specific
regression this fixes. Mutation-checked: dropping the hint fails the new guard.
Measurement note for the next refit: sum a shard's per-config Duration lines
before taking a median. Pooling them reads as a large over-prediction that is
not there.
* refactor(ai): give transport streams an honest writer type
* test(ai): use canonical transport stream fixtures
* fix(ai): preserve partial-less stream deltas
* feat(workers): run device sessions from Gateway bundles
Install the current Gateway bundle before a device environment becomes ready, verify it at attach and tunnel boundaries, launch only from the immutable namespaced bundle directory, and retire stale environments for idempotent reprovisioning. Remove the local execution mode and preserve the node-local build claim only as temporary inventory metadata for the final projection/cleanup slice.
* docs(runners): record Gateway bundle cutover
* test(ci): repair runner validation fixtures
# Conflicts:
# src/scripts/test-projects.test.ts
* fix(workers): surface outdated node recovery
Keep legacy runner inventory diagnostic-only while exposing the update-and-reconnect action through node, environment, provider, placement, and Control UI surfaces.
* fix(workers): reject legacy inventory with recovery
* fix(workers): bundle worker deploy closure
* test(workers): close bundle cutover gates
* fix(workers): compose browser runtime at build
* fix(workers): satisfy bundle cutover gates
* fix(workers): route temp runtime through infra
* docs(workers): align bundle host guidance
* fix(ui): fence outdated session destinations
Run 31856622489's compact-small-14 owned a 335s wall because the
storage-state hosted split packed state-migrations (27s), sqlite-snapshot
(24s), and session-cost-usage (10s) into one stripe against uniform default
weights. Anchor those files with CI checkmark walls from run 31814517685 and
lift the group hint to its observed 235s so the halves pack realistically.
* perf(ci): cut hosted CI critical path toward five-minute walls
Stripe the serial core test-type graphs across two hosted jobs and drop the
duplicated tsgo:test:root pass; gate the six-part QA Smoke matrix off pull
requests unless a QA-owned surface changed; split the fat multi-config Node
shards (cli/cli-process, unit-fast isolated/fake-timers, infra
logging-process/runtime-config) and lower the hosted split ceiling to 150
predicted seconds so no compact lane owns a ~280s wall; expand tooling to
seven stripes.
* perf(ci): widen hosted test-type striping to three jobs
Run 31825922122 measured ~40s per core test-type graph on loaded hosted
runners (282s worst stripe body of the two-way split); three stripes keep
each lane near 150s body under load.
Hosted CI runners restored the boundary-artifact cache and rebuilt it anyway: fresh checkouts re-stamp every input mtime, so mtime freshness never passed. Stamp files now record the input content digest and byte-identical inputs skip the rebuild (~60s saved per hosted lint/boundary job, 0.17s verify). Telegram CI shards pack ten files per job instead of five now that per-file import cost is back to seconds (#123607), halving the ~42-job fanout.
Five-file Telegram jobs finished the first file, then isolate re-imported the next graph in silence until the 300s watchdog killed the worker. Recycle the Vitest process after each file and keep five files per CI job.
Bound Telegram extension tests to five files per Vitest process across explicit config, directory, and full-suite routes while preserving serial isolated execution.
Co-authored-by: Ayaan Zaidi <hi@obviy.us>
* feat(docker): schedule image refreshes
* docs(docker): explain weekly image refreshes
* test(scripts): gate workflow-step execution on bash 4 mapfile support
Stock macOS bash 3.2 lacks mapfile; CI truth is Linux bash 5.
* test: cover docker-release suffix threading and sanctioned second caller
* test(codex): wire run-attempt-state into the attempt-extra project
#123345 added the file without a project owner; the full-suite coverage
guard fails for any PR that runs it.
* fix(ci): run build-artifacts PR validation on hosted runners
ci-build-artifacts-testbox.yml pinned PR runs to blacksmith-16vcpu and
ran Testbox lifecycle steps unconditionally, so the prepare-run landing
gate starved for every PR during a Blacksmith outage even with
OPENCLAW_CI_RUNNER_BACKEND=github. PR events now build on ubuntu-24.04
with dispatch-only Testbox steps, mirroring ci-check-testbox.yml.
* test(ci): align build-artifacts dispatch guard
Move llama.cpp chat and local embeddings onto a verified externally managed llama-server runtime. Remove the in-process native runtime, forked embedding workers, and node-llama-cpp dependency while preserving guided setup, local GGUF models, tool-capable agent runs, diagnostics, and operator docs.
* ci: run push compacts without tooling lanes and stripe giants three-way
* fix(ci): keep CompactNodeTestPlanMode internal
knip --production flags the exported type as unused; no external module
imports it.
* ci: shard hosted lint and node tests
Split core oxlint across hosted runners and use backend-specific compact Node admission caps while preserving Blacksmith behavior.
* test(ci): guard plugin prerelease planner isolation
The PR-changed test planner fails safe to the compact full-suite plan for any diff touching packages/**, but that compact plan excludes all extension test configs, so mixed package+extension PRs landed with zero extension test execution (escapes: PR #120534 breaking extensions/codex run-attempt.native-hook-relay.test.ts, PRs #122163/#121522 and cd7b7f639d breaking media-understanding-provider.test.ts and thread-lifecycle.test.ts on main full runs). The preflight now appends whole-config shards for the diff's touched extensions whenever the precise plan fails safe; whole configs (not precise targets) because the fail-safe cause leaves the non-extension diff's extension impact unbounded.
Peter Steinberger