Published to npm as @bitkyc08/opencodex@2.53.0 with dist-tag latest.
New Features
- Show the cached subset beside every token total (#4421)
- Meta Muse Code device-authorization login, with the api version header and on-demand quota (#4428)
- Classify Grok OAuth service tier per model and relay the tier echo (#4431)
- Expose standalone transcription API (#4391)
- Expose streaming dictation and owned live calls (#4392)
- Restore opt-in plaintext V2 agent messages (#4351)
- Add optional client transport hint suppression (#4355)
- Persist exact per-model capability declarations (#4374)
- Select provider approval reviewers with reversible root stamps (#4358)
- Route explicitly text-only models through existing vision handling (#4376)
- Read connected hub usage scoped to the enrolled client key (#4373)
- Register verified Z.AI GLM list prices as expected-cost overlays (#4444)
- Carry bounded executor and hub runtime foundations (#4362)
- Prefer accounts near shared quota reset (#4369)
- Establish quota-history publication identity (#4375)
- Explain Antigravity quota refresh failures (#4401)
- Persist content-free Codex WS upstream stage records (#4191) (#4427)
- Integrate opt-in workspace dashboard and admission (#4372)
- Retain raw quota observation history (#4404)
- Native-main device reauth for headless hubs (#3898) (#4433)
- Estimate observed effective quota capacity (#4408)
- Main-card Re-login with device code (#3898) (#4441)
- Prepare Codex Spark retirement and remove its quota UI (#4334)
- Request prompt caching and give the session cache an identity (#4453)
- Register Cognition/Devin list prices for both devin surfaces (#4466)
- Merge devin-cli into devin (import-first login, deprecated alias) (#4484)
- Pull an authenticated remote catalog into local Codex state (#4481)
- Give every model the reasoning ladder its catalog already names (#4490)
- Derive routed effort ladders from models.dev and replay a refused rung (#4488)
Bug Fixes
- Enforce effort ceilings without model pins (#4349)
- Refresh the catalog when paginated history refuses injection (#4411)
- Harden the imported CLI session against path, read, and cache defects (#4418)
- Read usage from ModelUsageStats and classify cloud failures (#4419)
- Map Connect trailer codes onto the status core acts on (#4423)
- Replay the assistant's reasoning instead of dropping it (#4426)
- Alias tool names over 64 chars for Meta Muse wire compatibility (#4422)
- Compensate native restore when history migrates during writes (#4380)
- Preserve unknown native reasoning while normalizing targets (#4346)
- Preserve raw content and provider summary provenance (#4345)
- Retry canonical Console upload rejection once (#4354)
- Align Spark Lite metadata with serialized body (#4359)
- Publish DeepSeek V4.1 context window (#4440)
- Bound multipart encrypted agent task recovery (#4364)
- Diagnose auxiliary listener failures and malformed edits (#4353)
- Retry clean empty forced answers once (#4356)
- Surface first overflow and bound thread remints (#4363)
- Shrink oversized inline images while retaining source content (#4366)
- Validate upstream sideband handshake before upgrade (#4367)
- Guide browser pairing with the hub and preserve error meaning (#4378)
- Separate local management catalog authority from inference (#4402)
- Apply explicit SWE-2 effort before model suffix (#4445)
- Inherit reasoning metadata for renamed key destinations (#4370)
- Retain readable totals and disclose skipped history (#4357)
- Explain and enforce automatic plan exclusions (#4361)
- Stop the header deadline from killing turns that are still generating (#4450)
- Stop a tier suffix from earning a second effort suffix (#4459)
- Stop refusal bookkeeping from failing refusal-free turns (#4468)
- Bind hub status credentials to connection snapshots (#4476)
- Distinguish model availability from auth failure (#4480)
- Fail an in-flight search at a truncated terminal (#4472)
- Bound bridge iteration buffering (#4478)
- Scope Fast-tier catalog copy to key-auth pricing (carry of #4077) (#4474)
- Normalize the Win32 extended-length prefix in history path identity (#4486)
- Scope the Windows logon trigger and stop blaming elevation for the denial (#4483)
- Stop leftover CLI profile ARN from collapsing a second account (#4482)
- Restore namespaced tool identities (#4487)
- Bind continuations to the serving API key (#4489)
- Execute code-mode view_image through unified exec (#4475)
- Let field-masked writes reach canonical OpenAI past stored overlays (#4477)
- Disarm sideband connect watchdog on the pre-opened path (#4496)
- Accept Pi and Anthropic-shaped image parts on the chat wire (#4497)
Documentation
- Close the Devin ACP removal unit (#4417)
- Record the #3988 carry whose merge dropped the trailer (#4432)
- Plan the Devin landing, TTFB deadline and prompt-cache work (#4446)
- Close out unpriced-model overlay unit after PR #4444 (#4449)
- Record the 36-PR lane-stack merge, regression audit and UI screenshots (#4463)
- Close the Devin unit and record why AssignModel is not implemented (#4464)
- Close the chat refusal-scope unit (#4471)
- Close the model-picker ladder unit (#4494)
Chores
- Retire the ACP adapter and give the shared adapter the tool-catalog nudge (#4415)
- Revert "feat(usage): show the cached subset beside every token total" (#4424)
- Retain phase evidence for coordinator refusal timeouts (#4414)
- Anchor the empty-data-dir fallback at the host home (#4448)
- Land god-file round 1 (cleanup, service, kiro) (#4495)
- Split cleanup.ts into cohesive leaf modules (#4491)
- Split kiro.ts into cohesive leaf modules (#4493)
- Split service.ts into platform leaves (#4492)
Other Changes
- Make v1 the default sub-agent surface and ask before base or v2 (#4462)
- Promote verified 2.53.0 product tree to main (#4507)
Changelog
Full Changelog: v2.52.0...v2.53.0
- #4334 feat(codex): prepare Codex Spark retirement and remove its quota UI @lidge-jun
- #4345 fix(responses): preserve raw content and provider summary provenance @lidge-jun
- #4346 fix(combos): preserve unknown native reasoning while normalizing targets @lidge-jun
- #4349 fix(chat): enforce effort ceilings without model pins @lidge-jun
- #4351 feat: restore opt-in plaintext V2 agent messages @lidge-jun
- #4353 fix(start): diagnose auxiliary listener failures and malformed edits @lidge-jun
- #4354 fix(responses): retry canonical Console upload rejection once @lidge-jun
- #4355 feat(responses): add optional client transport hint suppression @lidge-jun
- #4356 fix(search): retry clean empty forced answers once @lidge-jun
- #4357 fix(usage): retain readable totals and disclose skipped history @lidge-jun
- #4358 feat(catalog): select provider approval reviewers with reversible root stamps @lidge-jun
- #4359 fix(responses): align Spark Lite metadata with serialized body @lidge-jun
- #4361 fix(codex): explain and enforce automatic plan exclusions @lidge-jun
- #4362 feat(remote): carry bounded executor and hub runtime foundations @lidge-jun
- #4363 fix(cursor): surface first overflow and bound thread remints @lidge-jun
- #4364 fix: bound multipart encrypted agent task recovery @lidge-jun
- #4366 fix(openai-chat): shrink oversized inline images while retaining source content @lidge-jun
- #4367 fix(live): validate upstream sideband handshake before upgrade @lidge-jun
- #4369 feat(codex): prefer accounts near shared quota reset @lidge-jun
- #4370 fix(providers): inherit reasoning metadata for renamed key destinations @lidge-jun
- #4372 feat(remote): integrate opt-in workspace dashboard and admission @lidge-jun
- #4373 feat(usage): read connected hub usage scoped to the enrolled client key @lidge-jun
- #4374 feat(providers): persist exact per-model capability declarations @lidge-jun
- #4375 feat(codex): establish quota-history publication identity @lidge-jun
- #4376 feat(providers): route explicitly text-only models through existing vision handling @lidge-jun
- #4378 fix(dashboard): guide browser pairing with the hub and preserve error meaning @lidge-jun
- #4380 fix(codex): compensate native restore when history migrates during writes @luvs01
- #4391 feat(audio): expose standalone transcription API @lidge-jun
- #4392 feat(audio): expose streaming dictation and owned live calls @lidge-jun
- #4401 feat(accounts): explain Antigravity quota refresh failures @lidge-jun
- #4402 fix(opencode): separate local management catalog authority from inference @lidge-jun
- #4404 feat(codex): retain raw quota observation history @lidge-jun
- #4408 feat(codex): estimate observed effective quota capacity @lidge-jun
- #4411 fix(codex): refresh the catalog when paginated history refuses injection @lidge-jun
- #4414 test(clients): retain phase evidence for coordinator refusal timeouts @lidge-jun
- #4415 refactor(devin): retire the ACP adapter and give the shared adapter the tool-catalog nudge @lidge-jun
- #4417 docs(devlog): close the Devin ACP removal unit @lidge-jun
- #4418 fix(devin-cli): harden the imported CLI session against path, read, and cache defects @lidge-jun
- #4419 fix(devin): read usage from ModelUsageStats and classify cloud failures @lidge-jun
- #4421 feat(usage): show the cached subset beside every token total @lidge-jun
- #4422 fix(responses): alias tool names over 64 chars for Meta Muse wire compatibility @lidge-jun
- #4423 fix(devin): map Connect trailer codes onto the status core acts on @lidge-jun
- #4424 Revert "feat(usage): show the cached subset beside every token total" @lidge-jun
- #4426 fix(devin): replay the assistant's reasoning instead of dropping it @lidge-jun
- #4427 feat(responses): persist content-free Codex WS upstream stage records (#4191) @lidge-jun
- #4428 feat(oauth): Meta Muse Code device-authorization login, with the api version header and on-demand quota @lidge-jun
- #4431 feat(xai): classify Grok OAuth service tier per model and relay the tier echo @lidge-jun
- #4432 docs(credits): record the #3988 carry whose merge dropped the trailer @lidge-jun
- #4433 feat(codex): native-main device reauth for headless hubs (#3898) @lidge-jun
- #4440 fix(opencode-go): publish DeepSeek V4.1 context window @martinbcg
- #4441 feat(gui): main-card Re-login with device code (#3898) @lidge-jun
- #4444 feat(usage): register verified Z.AI GLM list prices as expected-cost overlays @lidge-jun
- #4445 fix(devin): apply explicit SWE-2 effort before model suffix @lidge-jun
- #4446 docs(devlog): plan the Devin landing, TTFB deadline and prompt-cache work @lidge-jun
- #4448 test(devin-cli): anchor the empty-data-dir fallback at the host home @lidge-jun
- #4449 docs(devlog): close out unpriced-model overlay unit after PR #4444 @lidge-jun
- #4450 fix(devin): stop the header deadline from killing turns that are still generating @lidge-jun
- #4453 feat(devin): request prompt caching and give the session cache an identity @lidge-jun
- #4459 fix(devin): stop a tier suffix from earning a second effort suffix @lidge-jun
- #4462 Make v1 the default sub-agent surface and ask before base or v2 @lidge-jun
- #4463 docs(devlog): record the 36-PR lane-stack merge, regression audit and UI screenshots @lidge-jun
- #4464 docs(devlog): close the Devin unit and record why AssignModel is not implemented @lidge-jun
- #4466 feat(usage): register Cognition/Devin list prices for both devin surfaces @lidge-jun
- #4468 fix(chat): stop refusal bookkeeping from failing refusal-free turns @lidge-jun
- #4471 docs(devlog): close the chat refusal-scope unit @lidge-jun
- #4472 fix(bridge): fail an in-flight search at a truncated terminal @lidge-jun
- #4474 fix(xai): scope Fast-tier catalog copy to key-auth pricing (carry of #4077) @lidge-jun
- #4475 fix(responses): execute code-mode view_image through unified exec @lidge-jun
- #4476 fix(cli): bind hub status credentials to connection snapshots @lidge-jun
- #4477 fix(providers): let field-masked writes reach canonical OpenAI past stored overlays @lidge-jun
- #4478 fix(images): bound bridge iteration buffering @lidge-jun
- #4480 fix(codex): distinguish model availability from auth failure @lidge-jun
- #4481 feat(codex): pull an authenticated remote catalog into local Codex state @lidge-jun
- #4482 fix(kiro): stop leftover CLI profile ARN from collapsing a second account @lidge-jun
- #4483 fix(service): scope the Windows logon trigger and stop blaming elevation for the denial @lidge-jun
- #4484 feat(providers): merge devin-cli into devin (import-first login, deprecated alias) @lidge-jun
- #4486 fix(codex): normalize the Win32 extended-length prefix in history path identity @lidge-jun
- #4487 fix(devin): restore namespaced tool identities @lidge-jun
- #4488 feat(reasoning): derive routed effort ladders from models.dev and replay a refused rung @lidge-jun
- #4489 fix(web-search): bind continuations to the serving API key @lidge-jun
- #4490 feat(devin): give every model the reasoning ladder its catalog already names @lidge-jun
- #4491 refactor(storage): split cleanup.ts into cohesive leaf modules @lidge-jun
- #4492 refactor(service): split service.ts into platform leaves @lidge-jun
- #4493 refactor(adapters): split kiro.ts into cohesive leaf modules @lidge-jun
- #4494 docs(devlog): close the model-picker ladder unit @lidge-jun
- #4495 refactor: land god-file round 1 (cleanup, service, kiro) @lidge-jun
- #4496 fix(live): disarm sideband connect watchdog on the pre-opened path @lidge-jun
- #4497 fix(chat): accept Pi and Anthropic-shaped image parts on the chat wire @lidge-jun
- #4507 release: promote verified 2.53.0 product tree to main @lidge-jun