* feat(anthropic): add Claude Opus 5 model support
Wire claude-opus-5 through the Claude 5 contract seams: adaptive-by-default
thinking with the full low..max effort range, default-sampling and prefill
stripping, streaming refusal contract, model-bound thinking replay, and 1M/128k
catalog metadata across anthropic, claude-cli, Vertex, Bedrock, and Mantle.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* fix(anthropic-vertex): normalize explicit Vertex Opus 5 model rows
Extend normalizeAnthropicVertexResolvedModel so user-configured Vertex Opus 5
rows regain reasoning, image input, 1M/128k limits, and the native thinking
map; update the live-model priority expectation and regenerate docs_map.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
The root package.json shipped "author": "", so npm showed no author for the
published openclaw package, and the three published workspace packages had no
author field at all. Set all four to the Foundation, matching the LICENSE
copyright holder and the openclaw.ai footer.
Also name the Foundation on the docs homepage, which never said who develops
OpenClaw even though the README and website footer now do.
Personal credits in docs/reference/credits.md, docs/start/lore.md, and the site
footer stay as authorship history; they are not org attribution.
* refactor(ai): invert plugin coupling behind the transport host port
* fix(ai): queue custom transport registrations until the host is configured
* refactor(ai): remove relocated transport sources from src/agents
* fix(ai): source core stream types from canonical packages and fix tarball fixtures
* fix(ai): invert plugin transport host wiring
* fix(ai): harden managed transport projection
* test(ai): register synchronous stream in transport mock
* fix(ai): lazily install transport runtime host
* fix(ai): preserve completion compat detection
* fix(ai): provider-billed totals desync the cost breakdown and drop cache writes
`applyProviderReportedUsageCost` overwrote only `cost.total`, leaving the
per-component estimates in place, so `input + output + cacheRead + cacheWrite`
stopped summing to the total and the replaced estimate was unrecoverable. The
repo rebuilds that sum in four other places; this helper was the only writer
breaking it. Retain the replaced estimate in a new optional `cost.estimatedTotal`
instead of inverting the override: free OpenRouter models report `cost: 0`, and
`shouldPreserveRecordedZeroCost` depends on a provider-billed zero staying
authoritative, so holding the estimate in `total` would write phantom spend.
`parseTransportChunkUsage` also hard-coded `cacheWrite: 0` and never read
`prompt_tokens_details.cache_write_tokens`, while its plugin-sdk sibling reads it
and documents the contract. Writes were dropped from `cacheWrite` and wrongly
counted inside `input`. OpenRouter routes through this lane.
Fixing that shifts tokens out of `input`, which `resolveContextInputTokens` used
as a proxy for prompt size, so context-overflow and length-stop detection are
aligned with the Anthropic lane's `input + cacheRead + cacheWrite`.
Related: #9016
* @
refactor(ai): drop the retained estimatedTotal cost surface
ClawSweeper blocked the branch three review cycles running on the same P1: the
new `Usage.cost.estimatedTotal` field crossed both the plugin SDK and the closed
worker transcript schema, making it an upgrade contract that a bug-fix PR cannot
decide on its own.
Measurement settles it: the field was written in one place and forwarded through
four transport layers, but nothing ever read it. It also did not repair the
invariant it was introduced for -- `input + output + cacheRead + cacheWrite` still
disagrees with a provider-billed `total`; the field only parked the discarded
estimate beside it.
Removing it drops eight files, retires the compatibility merge risk, and leaves
the two fixes that carry real runtime evidence: OpenRouter `cache_write_tokens`
mapping and the context-overflow accounting that depends on it.
@
* refactor(agents): tighten OpenRouter usage contract
---------
Co-authored-by: Peter Steinberger <steipete@gmail.com>
* fix(ai): honor Retry-After when retry-after-ms is unparseable
The ChatGPT Responses retry path returned as soon as `retry-after-ms` was
present, so a malformed or empty value discarded a valid `Retry-After`
header sent alongside it and the client fell back to blind exponential
backoff. The sibling parser in `provider-transport-fetch.ts` treats the
two headers as ordered preferences and falls through on a parse failure;
this aligns the Responses path with that contract.
* test(ai): deduplicate retry header coverage
---------
Co-authored-by: Peter Steinberger <steipete@gmail.com>
* fix(ai): ChatGPT Responses retries errors it classified as non-retryable
The retry loop builds its friendly error and throws it from inside the same
try block whose catch treats unknown failures as retryable network errors.
Authentication and bad-request failures were therefore resent up to
maxRetries times, adding 7s of backoff before surfacing the same message.
Tag the already-classified failure with a private error type and rethrow it
unchanged from the catch, leaving the retryable paths untouched.
* test(ai): move retry classification cases to a dedicated file
The provider test file was already near the 1000-line oxlint max-lines
budget, so the new cases pushed it over. Split them out following the
existing <module>.<topic>.test.ts convention.
* refactor(ai): separate HTTP and transport retries
Co-authored-by: Yigtwxx <yigiterdogan023@gmail.com>
* test(ai): clarify synthetic JWT fixture
Co-authored-by: Yigtwxx <yigiterdogan023@gmail.com>
---------
Co-authored-by: Peter Steinberger <steipete@gmail.com>
* fix(agents): record token usage when a Responses turn ends incomplete
The agent-side Responses processor only had a terminal branch for
response.completed. A stream that ends with response.incomplete — the
max_output_tokens and content_filter cases, and what Azure emits on early
truncation — matched no branch at all, so the event was dropped: usage was
never recorded and stopReason was never set. The turn reports zero tokens and
zero cost, which is the drift in #100954.
#109615 fixed the same split on the package-side processor by finalizing
completed and incomplete through one finalizeResponse. This does the same for
the agent path, which #109615 did not touch: both terminal events now record
usage, cost and service-tier pricing through one helper.
The helper moves to its own module rather than growing openai-responses-transport.ts,
which is a legacy file the max-lines ratchet will not let grow; extracting the
block drops it from 2502 to 2444 lines.
Content-filtered turns are mapped to a provider error instead of a plain length
stop, matching what the package side already does, so the two terminal surfaces
do not disagree.
* fix(agents): keep Responses output backfill on completed turns
Terminal handling now covers response.incomplete, which also routed those
events into backfillCompletedResponseOutput. That reconstruction exists to
recover a final answer when item events never arrived; an incomplete turn has
no final answer, so replaying its partial output persisted truncated text the
streaming path never emitted. Usage and stop-reason recording still apply to
both terminal events.
* refactor(ai): share one Responses terminal usage mapper
The agent transport mapped terminal usage buckets, cost, and stop reasons in
parallel with the package-side processor, so the two could drift on token
buckets, service-tier pricing, or future terminal-event semantics.
Both now call one canonical mapper. openai-responses-shared.ts cannot be
imported across the package boundary, so the mapper lives in its own module
re-exported through the existing internal/openai entry point; no new subpath
is introduced. The agent module keeps only the reasoning-token accounting the
package path does not track.
Merging the two revealed a real disagreement on totalTokens: the package took
the reported total, the agent summed the split buckets. The canonical rule is
now max(bucket sum, reported total), which keeps the reported value while
covering both payloads that omit total_tokens and payloads whose cached_tokens
exceed input_tokens, where clamping leaves the reported total short.
* fix(agents): preserve partial incomplete Responses output
Co-authored-by: Yigtwxx <yigiterdogan023@gmail.com>
* test(agents): cover incomplete terminal output backfill boundaries
The head's partial-output preservation is only safe while it stays a recovery
path. Pin both sides of that guard: text that already streamed must not be
replayed from the terminal payload, and a non-length incomplete stop must not
surface partial text as an answer.
* fix(agents): avoid shadowing Responses options
Co-authored-by: Yigtwxx <yigiterdogan023@gmail.com>
* style(agents): apply oxfmt to the incomplete backfill test
* test(agents): register incomplete Responses suite
Co-authored-by: Yiğit ERDOĞAN <yigiterdogan023@gmail.com>
---------
Co-authored-by: Peter Steinberger <steipete@gmail.com>
* fix(ai): harden OpenAI-compatible provider compat and error surfacing
* test(ai): add live completions compat coverage
* test(infra): mirror packages live glob in live-config test
* fix(ai): track Responses output items per index and require terminal stream events
* test(ai): add live Responses stream coverage
* test(infra): mirror packages live glob in live-config test
* fix(ai): reject noncanonical array indices in local schema refs
* test(ai): harden local schema ref indices
---------
Co-authored-by: Peter Steinberger <steipete@gmail.com>
* fix(ai/gemini): coerce numeric/boolean enum values to strings
Gemini Function Calling only accepts string enum values (see
https://ai.google.dev/gemini-api/docs/function-calling). Tool schemas
containing enum: [1,2,3] or enum: [true,false] were passed through
unchanged, causing Gemini to reject the entire request with 400
TYPE_STRING errors at 'function_declarations[X].parameters...enum[N]'.
Extend cleanSchemaForGeminiWithDefs to coerce enum and const values
via a shared helper (number/boolean/bigint -> string, drop null and
non-coercible entries, de-duplicate), and retype the containing schema
as 'string' since Gemini's validator only accepts enum on string-typed
schemas.
Adds seven vitest cases covering numeric/boolean enum, const, nested
schemas, and dedup/empty-after-coercion edge cases.
Fixes#104566
Co-Authored-By: Claude <noreply@anthropic.com>
* fix(ai/gemini): make enum retyping property-order independent
Address @clawsweeper review on #104567:
The initial patch set cleaned.type = 'string' inline while visiting the
'enum' or 'const' key. For schemas where 'type' appears AFTER 'enum' /
'const' in the input object — a valid ordering since JSON Schema property
order is not semantic — a later 'type' pass would overwrite the coercion
via the tail 'cleaned[key] = value' branch, and Gemini would still receive
an invalid declaration.
Move the type override out of the loop: track whether any enum/const was
coerced via a boolean flag, and force cleaned.type = 'string' after the
loop finishes. This makes the fix independent of input property order.
Also add curly braces around single-line ifs to satisfy check-lint
(eslint(curly)) which failed on the previous commit.
New reversed-order regression tests added.
Co-Authored-By: Claude <noreply@anthropic.com>
* fix(ai): preserve Gemini enum schema types
---------
Co-authored-by: leonme <leonme@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Peter Steinberger <steipete@gmail.com>
* feat(tooling): extend strict-ratchet lane to all remaining leaf packages
* fix(packages): burn down indexed-access debt in ai and agent-core
All 196 noUncheckedIndexedAccess errors fixed behavior-identically
(iteration, .at()/slices, DataView byte math, guarded queue peeks) plus
cleanup finds: a sparse-tool diagnostic bug, deduplicated Responses
call/item ID parsing with an honest string|undefined splitter, and a
clearer local failure for redacted-thinking blocks missing signatures.
Removes all 17 remaining non-null assertions across ai, gateway-client,
and llm-core so the assertion ban aligns 1:1 with the ratchet lane.
speech-core joins memory-host-sdk as structurally deferred (imports
src/** via plugin-sdk path mappings).
Jason (Json)