* fix(google): declare bundled Gemini models in the plugin manifest model catalog
* fix(google): scope manifest catalog mirror to canonical provider and declare shared-model tiers
Address ClawSweeper review on #139243:
- drop google-gemini-cli / google-vertex manifest mirrors so legacy CLI refs
keep their Doctor migration hint and CLI/Vertex rows stay out of runtime
catalog planning (neither declares runtime discovery)
- declare explicit capable code-mode tiers on the five sibling rows newly
matched by the shared-upstream-model contract (opencode, github-copilot, gmi)
- strengthen the drift guard to full-row equality against the runtime static
catalog and pin the CLI/Vertex no-mirror boundary
- add regression test keeping the google-gemini-cli migration hint alive
* fix(google): preserve native Gemini transport in the manifest catalog mirror
* fix(google): derive catalog from manifest
* fix(google): tighten manifest catalog validation
---------
Co-authored-by: Vincent Koc <vincentkoc@ieee.org>
Related: #136257. Builds on merged #139357.
Make all bundled shared-helper catalog callers report actual acquisition failures and authoritative empty results. Preserve shipped external advisory defaults through the same transport/cache implementation.
The existing publication owner retains compatible learned inventory or degraded provider-owned starters. Static preparation consumes registered hooks with their authoritative plugin identity, without activating unknown runtimes or widening successful empty membership. No operator option, persistence schema, new cache, or auth re-observation is introduced.
Proof: retained built baseline failure, packaged CLI/Gateway fault/recovery/empty checks, public SDK/provider API compatibility, fresh/upgrade starter checks, and the decisive seven-clause downstream composite on exact shared head 5e0ecef411cc7bcabbcab2672227910a05209ffd. Composite 296bfa19ec7663044ce50b8ebb3652555900d531 adds only the unchanged separate DeepInfra/NVIDIA producer patch; that patch is not part of this PR. Controlled clock advancement and fixture-rejected background inference are disclosed. No real vendor inference or UI claim.
Production net +199; tests/support +313; docs +34; ratchet metadata -1; generated/lockfile delta zero. Growth preserves shipped API and shared failure/empty/ownership contracts rather than adding parallel provider policy.
Co-authored-by: Ayaan Zaidi <hi@obviy.us>
* chore(release): carry shipped 2026.9.2 state onto main
Align main and official plugin versions with the published stable release and
preserve the exact tagged changelog. All ten canonical runtime/qualification
fix commits are already on main; retain its newer SQLite worker and Watch UI.
Refresh generated UI and native locale artifacts through their canonical
owners after provider failures left main's current inventories out of sync.
The release tag and published package bytes remain unchanged.
* chore(release): isolate generated locales from version closeout
Keep release version metadata and the exact published changelog in this PR.
Generated UI locales are covered by #139382; native locales use #139364.
Set main to the shipped stable version 2026.9.1 (root, apps, plugins, and
version-owned generated metadata via `pnpm release:prepare`), make the
`## 2026.9.1` changelog section identical to the tagged release branch, and
carry the native locale refresh that the release preflight requires.
Release: https://github.com/openclaw/openclaw/releases/tag/v2026.9.1
npm: openclaw@2026.9.1 (latest, beta); 89 @openclaw/* plugins at 2026.9.1.
* fix(opencode-go): send OpenClaw User-Agent on native OpenCode Go routes
OpenCode Go's documented client-identification contract requires coding
agents to identify themselves with a specific User-Agent. opencode-go
requests previously carried only the generic OpenAI SDK fallback.
Add a vendor-documented attribution policy for opencode-go that sets
User-Agent: openclaw/<version> only on the verified native OpenCode Go
endpoint; Zen and custom-proxy routes are unchanged. Route the Anthropic
transport through the same central policy via the plugin stream wrapper
so both Go transports identify consistently, with caller headers winning.
Co-authored-by: VACInc <3279061+VACInc@users.noreply.github.com>
* fix(opencode-go): send OpenClaw User-Agent on native OpenCode Go routes
Worked on by:
- @VACInc
Co-authored-by: VACInc <3279061+VACInc@users.noreply.github.com>
OpenClaw-Publication: 2182e49f-af02-4250-864c-f7777a89eb83
* test(opencode-go): prove User-Agent precedence
Cover configured User-Agent replacement on native OpenCode Go requests across OpenAI-compatible, Anthropic stream, and simple-completion paths. Custom proxy routes retain the configured value.
Co-authored-by: VACInc <3279061+VACInc@users.noreply.github.com>
* fix(opencode-go): send OpenClaw User-Agent on native OpenCode Go routes
Worked on by:
- @VACInc
Co-authored-by: VACInc <3279061+VACInc@users.noreply.github.com>
OpenClaw-Publication: 41cb2f26-9a12-4b40-b903-2d5b2fc3ccbd
* fix(opencode-go): scope attribution to Go routes
Worked on by:
- @VACInc
Co-authored-by: VACInc <3279061+VACInc@users.noreply.github.com>
* fix(ai): merge transport headers case-insensitively
Worked on by:
- @VACInc
Co-authored-by: VACInc <3279061+VACInc@users.noreply.github.com>
* fix(opencode-go): send OpenClaw User-Agent on native OpenCode Go routes
Worked on by:
- @VACInc
Co-authored-by: VACInc <3279061+VACInc@users.noreply.github.com>
OpenClaw-Publication: e353cea0-b205-4dde-bded-b9bf49b93cec
* test(opencode): keep Zen live replay on free model
Worked on by:
- @VACInc
Co-authored-by: VACInc <3279061+VACInc@users.noreply.github.com>
* Publish openclaw/openclaw-openclaw-vacinc-issue-135538
Worked on by:
- @VACInc
Co-authored-by: VACInc <3279061+VACInc@users.noreply.github.com>
OpenClaw-Publication: 2d3d8e21-99a1-4cd1-a90e-9ee00e647982
* Publish openclaw/openclaw-openclaw-vacinc-issue-135538
Worked on by:
- @VACInc
Co-authored-by: VACInc <3279061+VACInc@users.noreply.github.com>
OpenClaw-Publication: 5bffe897-9d10-4b1b-b7de-550c3b400168
* test(opencode): use replay-capable free Zen model
Worked on by:
- @VACInc
Co-authored-by: VACInc <3279061+VACInc@users.noreply.github.com>
* test(opencode): use interleaved free Zen model
Worked on by:
- @VACInc
Co-authored-by: VACInc <3279061+VACInc@users.noreply.github.com>
* test(opencode): stabilize free Zen live smoke
Worked on by:
- @VACInc
Co-authored-by: VACInc <3279061+VACInc@users.noreply.github.com>
---------
Co-authored-by: roboclaw-bot <309084314+roboclaw-bot@users.noreply.github.com>
Co-authored-by: VACInc <3279061+VACInc@users.noreply.github.com>
* fix(models): refresh native pricing without pinning defaults
Use complete native OpenCode and Venice pricing schedules while preserving
authored zero, partial, flat and tiered costs across preparation and lookup.
Avoid generated price pins during fresh merge-mode Venice onboarding.
Follow-up to #133349 (cached long-context accounting) and #129694
(Venice runtime pricing); related to #113549 (broader OpenCode catalog drift).
* fix(pricing): preserve native schedules through resolution
Merge prepared costs beneath authored display prices, retain validated native
free schedules under exact owner policy, and compile OpenRouter overrides
with strict thresholds and source-order inheritance for each price key.
Preserve zero, partial, flat and tiered user pricing without new settings
or storage. Follow-up repairs for #133695.
* perf(plugins): keep provider policy artifacts on leaf module graphs
Provider policy artifacts (provider-policy-api.js) load eagerly whenever a
provider is resolved, but five of them imported the provider-model-shared
barrel at runtime, dragging the transports/compat/state graph into every
policy load. In contexts without a native TS require hook (Vitest workers,
non-tsx source runs) jiti compiled that whole graph: ~65s of event-loop
starvation on the first embedded run, which is what pushed
run.session-permissions.test.ts past its 120s timeout before #129582.
Add openclaw/plugin-sdk/claude-model-runtime, a narrow family-level and
local-only subpath re-exporting the Claude identity/thinking helpers from
their leaf owners (@openclaw/llm-core, plugins/provider-claude-thinking).
Switch anthropic, anthropic-vertex, and opencode policy artifacts to it, and
amazon-bedrock plus ollama to the already-plugin-visible
@openclaw/model-catalog-core leaves. The barrel keeps re-exporting the same
symbols, so no existing consumer changes.
Measured on the embedded-runner host route (first run, Vitest worker):
65540ms -> 6627ms; jiti self-time 23.4s -> 1.3s, statSync 18.6s -> 0.7s.
run.shared-integration.test.ts drops from 167s to 65s as a side effect.
Also pin run.inherited-auth-owner.test.ts to the mocked plugin-harness route
(its assertions are provider-agnostic; 37.6s -> sub-second test time) and
document the no-provider default-route trap on overflowBaseRunParams.
Follow-up to #129582.
* chore(plugins): register claude-model-runtime boundary aliases
The extension package boundary contract requires every local-only plugin-sdk
entrypoint to carry a d.ts path alias in the shared boundary map and xai's
derived override set; CI's contracts-plugin lane caught the missing entries.
* chore(release): exclude claude-model-runtime declarations from the pack
Local-only plugin-sdk entrypoints ship runtime .js only; the release check
derives the required pack exclusion from the local-only registry and CI's
core-tooling lane caught the missing package.json files entry.
* test(agents): assert the mocked harness route in auth-owner proof
ClawSweeper P2: without the agentHarnessId assertion a silent fall-back to
the built-in host harness would still pass the auth-owner assertions while
proving the wrong route; fail loudly like run.session-permissions.test.ts.
* fix(gateway): tools.invoke must carry the caller's host-minted role authority
The connect handshake resolves each connection's authority once and stores it
server-side (shared-secret operator owners mint system authority there).
tools.invoke discarded that fact and re-derived ownership from scopes, so a
shared-secret caller with no durable profile resolved to the deny-by-default
role and was refused dispatch on its own agents — while the same connection
could still mutate sessions directly.
Carry client.internal.operatorRoleActor into the synthetic dispatch client and
keep the scope-derived fallback for callers that have no connection actor
(HTTP). Regression test fails pre-fix with the FORBIDDEN agent-allowlist error.
* test(opencode): close the fake CLI before exec to stop ETXTBSY flakes
The catalog suite wrote the fake opencode executable and spawned it
immediately. Under parallel CI shards the write handle could still be open
at exec time, so the launch failed with ETXTBSY and failed the shard.
Write through an explicit file handle with an fsync before close so the
binary is fully durable before the first spawn.
* fix(ci): repair red main type and lint gates
Two gates were failing on main independently of this branch:
- extensions/qa-lab cleanup tests still built OpenClawCrablineChannelDriverSelection
with the retired smokeArtifactPath and a stale capabilityMatrixPath, so
check:test-types failed after the readiness-artifact change (#124189).
Align both fixtures with the current type and its pinned constants.
- scripts/github/release-validation-campaign.d.mts declared the Actions Octokit
client as any (#129726), tripping no-explicit-any. Declare the structural
subset the publisher actually calls instead of suppressing the rule.
Verified failing on clean origin/main before the fix.
* refactor(plugin-sdk): dedupe session-catalog cursor paging into session-catalog-runtime
The acpx Pi and opencode session catalogs carried byte-identical
boundedLimit/encodeCursor/decodeCursor/optionalRawCursor/transcriptPage
scaffolding (~114 duplicated lines each). Move one canonical copy into the
private-local session-catalog-runtime SDK subpath both plugins already may
import, and keep the plugin-named cursor guards as direct aliases.
Net -87 production LOC; no behavior change (error strings, cursor canonical
form, and byte budgets are unchanged).
* refactor(core): replace private sleep() clones with canonical sleep helpers
managed-linux readiness polling now uses @openclaw/retry sleepWithAbort with
ref:false (preserving the clone's timer.unref behavior), and the embedded
agent runner's async-task wait uses src/utils/sleep.ts. Both clones matched
canonical semantics on real inputs (positive integer poll intervals).
The update-managed-service-handoff copy stays: it lives inside a serialized
standalone handoff script (String.raw template) that cannot import repo
modules.
* fix(sessions): isolated gateways no longer inherit HOME external session catalogs
A gateway on isolated state (custom OPENCLAW_STATE_DIR/CONFIG_PATH/OPENCLAW_HOME,
relocated home, or any named profile) listed, read, continued, archived, and
reopened the operator's real Claude Code/Codex/OpenCode/Pi sessions from the
process HOME. External catalogs now require the default install identity for
process-HOME scans: every catalog verb receives the isolation policy and rejects
HOME-fallback local targets, unknown providers fail closed unless they declare
supportsProcessHomeIsolation, and one structured warning records the skip.
Paired-node hosts and explicitly rooted stores (CLAUDE_CONFIG_DIR, CODEX_HOME,
OPENCODE_DB, Pi session dirs) keep working; default-identity gateways are
unchanged.
* fix(sessions): inject catalog HOME-isolation fact at registry construction
* chore(sdk): regenerate plugin API baselines after rebase
* chore(sdk): regenerate plugin API baselines after rebase
* fix(opencode): migrate retired free model refs
Repair the shipped beta upgrade path through the provider-owned Doctor contract. Activate the migration from the canonical configured-model selector inventory so every supported selector is covered consistently.
Co-authored-by: samson1357924 <samson1357924@gmail.com>
* fix(opencode): migrate media model preferences
---------
Co-authored-by: samson1357924 <samson1357924@gmail.com>
* feat(plugins): catch code-mode drift between catalogs sharing one model
Adds a contract test that groups bundled catalog rows by shared upstream
model and requires every row in a group to declare compat.codeMode once any
sibling does. Rows sharing a model id group automatically; rows under
different ids opt in with the new manifest-only `upstreamModel` marker.
Moves the kimi catalog into its manifest so the scan can see it, and records
the tier reseller catalogs were silently missing as explicit "capable".
* docs: regenerate docs map for the shared-model code-mode section
* fix(plugins): import adopted catalog history into the OpenClaw transcript
Import native Pi and OpenCode history when a catalog session is adopted so the OpenClaw transcript reflects the full conversation immediately.
Mark imported user rows with mirrorOrigin because they are transcript mirrors, not new external turns. This preserves watcher deduplication and prevents adopted history from being reported as fresh upstream input.
* refactor(plugins): split catalog history importer
Keep the session catalog contract module leaf-like by moving transcript mutation into a dedicated runtime module. This breaks the registry-to-transcript import cycle while preserving the Plugin SDK entrypoint and importer behavior.
* feat(plugins): support additional upstream session kinds
* feat(acpx): monitor upstream Pi sessions
* feat(opencode): monitor upstream OpenCode sessions
* fix(opencode): ignore hidden user text activity
* refactor(plugins): share upstream echo filtering
* fix(opencode): preserve marker tuple narrowing
* refactor(opencode): detect upstream turns via event_sequence cursor
Replace timestamp/grace-window completeness heuristics with OpenCode's own
per-session event_sequence.seq cursor, which advances inside the same
transaction as the message/part projections. Detects change instead of
inferring completion, so staged part writes can no longer drop a turn.
Handles cursor regression from OpenCode migrations that clear event_sequence,
suppresses compaction replay and summary re-publish, and restricts human-turn
classification to visible non-synthetic text.
* docs(plugins): record why the adoption coordinator requires complete
* fix(opencode): align cursor proof with strict types
* fix(acpx): keep Pi JSON parser private
