* fix(google): keep parallel Gemini tool responses in the turn after the model
On Gemini < 3 vision models, a parallel tool-call turn whose non-last result
returns an image split function responses across user turns. The merge heuristic
only inspected contents[last], so the separate "Tool result image:" turn landed
between two parallel responses and stranded the second one in a fresh turn. The
turn right after the model then carried fewer functionResponse parts than the
model issued functionCall parts, so Gemini returned 400 INVALID_ARGUMENT. Because
the malformed turn is persisted, every later turn re-400s and the session sticks.
Replace the contents[last] heuristic with a run-scoped accumulator: all responses
for one model turn merge into the single user turn after it, and Gemini < 3 image
turns defer to the end of the tool-result run so they trail that response turn.
Covers both google.ts and google-vertex.ts, which share this convertMessages.
* fix(google): align provider transport tool result turns
---------
Co-authored-by: Vincent Koc <25068+vincentkoc@users.noreply.github.com>
Summary:
- Strip the redundant `google/` provider prefix before embedding Google Vertex model ids under `/publishers/google/models/`.
- Keep bare Vertex model ids unchanged.
- Add regression coverage for the provider-qualified Vertex path.
Verification:
- `node_modules/.bin/oxfmt --check --threads=1 extensions/google/transport-stream.ts extensions/google/transport-stream.test.ts`
- `node scripts/run-oxlint.mjs extensions/google/transport-stream.ts extensions/google/transport-stream.test.ts`
- `node scripts/run-vitest.mjs extensions/google/transport-stream.test.ts --maxWorkers=1 -t 'strips redundant google provider prefixes from Google Vertex model paths'`
- Autoreview clean
- AWS Crabbox `run_649b209478d2` focused Node 24 regression proof
- AWS Crabbox `run_e193db2707ad` remote `check:changed`
- Exact-head CI green for `23aca6f46f596e220df37d939317b433f7044ec6`
- Contributor live Google Vertex proof recorded in the PR body
* Google: preserve Gemini CLI OAuth failure context
Port the diagnostics-only fix to the bundled Google OAuth implementation so user-facing setup errors explain why automatic Gemini CLI client-config discovery failed.
Constraint: #54289 may remove or gate automatic Gemini CLI credential extraction
Rejected: Change extraction consent behavior here | security/product decision belongs in #54289
Confidence: medium
Scope-risk: narrow
Tested: pnpm test -- extensions/google/oauth.test.ts
Tested: pnpm check
Tested: pnpm format:check extensions/google/oauth.credentials.ts extensions/google/oauth.test.ts
Not-tested: full pnpm test suite
* Google: clarify Gemini bundle fallback diagnostic comment
Keep the follow-up limited to the explanatory comment so it matches the diagnostic error preservation added around bundle traversal failures.
Constraint: comment-only cleanup after diagnostics port
Confidence: high
Scope-risk: narrow
Tested: pnpm format:check extensions/google/oauth.credentials.ts
Not-tested: tests not run; comment-only change
* fix: resolve OAuth test rebase conflict
Decode Google Vertex authorized_user ADC OAuth token refresh responses from bytes so gzip-compressed token payloads still expose access_token. Adds a regression test for the compressed token response path while preserving plain JSON handling and the custom fetch seam.
Proof: OPENCLAW_VITEST_MAX_WORKERS=1 node scripts/run-vitest.mjs extensions/google/transport-stream.test.ts; pnpm exec oxfmt --check extensions/google/vertex-adc.ts extensions/google/transport-stream.test.ts; pnpm tsgo:extensions; git diff --check origin/main...HEAD; autoreview --mode branch --base origin/main. PR CI check-test-types failure was reproduced on current origin/main 607bbe4f5c and is unrelated to this two-file Google provider change.
Thanks @liaoandi for the fix and live Google Vertex ADC proof.
Fixes#88528.
Gemini web_search now accepts successful Google Search grounding responses that include candidate text and an empty `groundingMetadata` object without `groundingChunks`, returning wrapped content with `citations: []` instead of throwing `Gemini API error: malformed JSON response`.
Proof: live direct Gemini API reproduced the empty-grounding response shape; live OpenClaw provider failed before and succeeded after; `node scripts/run-vitest.mjs extensions/google/web-search-provider.test.ts`; `pnpm lint:web-search-provider-boundaries`; targeted oxfmt check; `git diff --check`; autoreview clean.
CI note: admin bypass used for unrelated failures in memory-core/device-pair/scripts, an existing core architecture cycle, and gateway-watch; PR diff touched only the two Gemini files.
Fixes#89891.
Route Google Vertex `eu` and `us` multi-region locations to the REP hosts used by `@google/genai`, and keep native Vertex endpoint trust exact to those two hosts.
Verification before merge:
- Live 1Password-backed GCP service-account probe: `eu-aiplatform.googleapis.com` returned Google HTML 404; `aiplatform.eu.rep.googleapis.com` reached Vertex JSON `PERMISSION_DENIED` with the same token.
- `node scripts/run-vitest.mjs src/agents/provider-attribution.test.ts extensions/google/vertex-multi-region-host.test.ts extensions/google/api.test.ts` passed.
- `git diff --check` passed.
- `autoreview --mode branch --base origin/main` clean.
- Real behavior proof passed on latest head.
- ClawSweeper re-review: ready for maintainer review, proof sufficient.
CI note: merged with maintainer approval despite red CI because the failures were unrelated to this PR and reproduced on untouched paths: `extensions/acpx/doctor-contract-api.ts`, `extensions/device-pair/notify.ts`, script lint, and existing architecture/gateway-watch checks.
Co-authored-by: alkor2000 <200923177@qq.com>
Forward configured stop sequences to Gemini generationConfig.stopSequences in the bundled Google transport, matching the shared Google provider behavior and the @google/genai request contract.\n\nThanks @coder999999999.
* fix(providers): use native reasoning mode for direct Gemini API, keep CLI tagged
Gemini 2.5+ delivers reasoning via native thinkingParts (thinkingConfig.
includeThoughts). Having tagged mode active at the same time injects a
<think>…</think>/<final>…</final> directive into the system prompt; the
model opens a <think> block before a tool call, never closes it, and
returns an empty post-tool turn (content:[], payloads=0 error, #69220).
Fix: override resolveReasoningOutputMode in buildGoogleProvider() only —
not in the shared GOOGLE_GEMINI_PROVIDER_HOOKS. The Gemini CLI backend
(google-gemini-cli) runs gemini --output-format json and parses a text
response field, not native thought parts; it must stay on tagged mode.
A regression test confirms google-gemini-cli remains "tagged".
Also remove the dead BUILTIN_REASONING_OUTPUT_MODES entry keyed on
"google-generative-ai" from provider-utils.ts — that string is only
ever the transport model.api value, never the provider id passed to
resolveReasoningOutputMode, so the map was unreachable.
Fixes#69220
* docs: clarify Gemini reasoning output modes
* fix(google): keep Antigravity reasoning tagged
* fix(google): default direct reasoning checks to native
* fix(google): import reasoning context from plugin entry
---------
Co-authored-by: Peter Steinberger <steipete@gmail.com>
* fix(google): add gemini-3.1-flash-lite to provider catalog
Adds the missing gemini-3.1-flash-lite model definition to the
GOOGLE_GEMINI_TEXT_MODELS array. This resolves the ProviderFailoverError
when configuring google-vertex/gemini-3.1-flash-lite.
Fixes#89390
* test(google): cover Gemini flash lite catalog row
---------
Co-authored-by: Peter Steinberger <steipete@gmail.com>
Fix live model inference edge cases across provider streaming, model switching, outbound delivery, and gateway tool resolution.
Includes live/provider issue fixes and leaves #89100 explicitly partial for the remaining FM-2 group routing case.
When a configured Google provider/model row had no explicit
but had a baseUrl set, the fallback defaulted to openai-completions,
causing Gemini requests to route through the OpenAI Responses
transport instead of the native @google/genai transport.
Made resolveConfiguredProviderDefaultApi provider-aware: for the
google provider, the default API is now google-generative-ai.
Root cause: the generic fallback assumed any provider with a baseUrl
should use openai-completions, which is incorrect for Google's native
Gemini API.
Co-authored-by: xin <1052326311+xin@users.noreply.github.com>
Bumps OpenClaw release metadata to 2026.5.31 across package manifests, app version files, plugin metadata, changelog headings, and generated shrinkwraps.
Verification:
- pnpm plugins:sync:check
- pnpm ios:version:check
- pnpm deps:shrinkwrap:check
- git diff --check
- stale 2026.5.30/build-code scan across changed files
- autoreview clean: no accepted/actionable findings
- PR CI green for real gates: Checks, security scans, dependency guard, app lanes, real behavior proof
Known non-code workflow issue:
- label workflow failed because this PR hits GitHub's 100-label issue cap before the size-label step.
* feat: add Claude Opus 4.8 support
* fix: omit Vertex Opus sampling overrides
* fix: preserve Opus adaptive thinking levels
* fix: clamp Anthropic max effort support
* fix: use sha256 for QA mock call ids
* fix: type Anthropic transport test model metadata
* test: update PDF model default for Opus 4.8
