v2.51.0

Published to npm as @bitkyc08/opencodex@2.51.0 with dist-tag latest.

New Features

• Let an operator keep a downgraded account out of pool rotation (#4211) (#4238)
• One-port hub — same-port loopback companion listener, honest hub-gate messages (#4250)
• Provision the data-plane token, add ocx hub invite, and one hub block in ocx status (#4252)
• A connected client reports the hub's state, not its own (#4255)

Bug Fixes

• Give sessionless requests an isolated session lane (#4226)
• Refuse a write that would take a config file from its owner (#4227)
• Warm up new ChatGPT Free accounts instead of blaming the credential (#4126) (#4230)
• Admit the Responses Coding Plan preset to the quota reader (#4231)
• Name the account that left the pool instead of only dropping it (#4212) (#4233)
• Support native pnpm self-updates (#4235)
• Classify Codex WS failures instead of restating them (#4232)
• Name the real refusal cause instead of asserting ownership (#4237)
• Refuse vendor CLI scaffolding in routed output (#4234)
• Refuse a hub catalog the local Codex CLI cannot parse (#4240)
• MacOS repair no longer evicts a healthy launchd job, and verifies the one it loads (#4249)
• Route the hub's own local clients through their two destinations (management ingress + loopback inference) (#4251)
• Emit max/ultra efforts unconditionally and expire stale clamp diagnostics (#4257)
• Keep the retryable main-refresh refusal an overload, not a bad key (#4269)

Documentation

• Record the 2.50.0 regression audit and release train (#4196)
• Plan the 260911 lane dispatch round (#4217)
• Publish the 260911 lane branches and refresh the round ledger (#4220)
• Mechanize the lane tracking refresh and fix the wp4 merge gate (#4221)
• Fold the seven-lane feasibility audit into lane ownership and scope (#4223)
• State which account each provider login spends (#4239)
• Make the setup runnable on a fresh config and close the macOS data plane (#4241)
• The one-port hub recipe, invite flow, and launchd semantics (en+ko, skill) (#4254)
• Provider: seed GLM-5.3-Flash on the BigModel Responses preset (#4244)
• Make the remote hub guide runnable in every locale (#4247)

Other Changes

• Open dev at 2.51.0 before releasing 2.50.0 (#4194)
• Pool: name the account when a refresh fails or its models vanish (#4248)
• Client: report local Codex readiness instead of bare connected state (#4246)
• Promote verified 2.51.0 product tree to main (#4271)

Changelog

Full Changelog: v2.50.0...v2.51.0

• #4194 chore(release): open dev at 2.51.0 before releasing 2.50.0 @github-actions[bot]
• #4196 docs(devlog): record the 2.50.0 regression audit and release train @lidge-jun
• #4217 docs(devlog): plan the 260911 lane dispatch round @lidge-jun
• #4220 docs(devlog): publish the 260911 lane branches and refresh the round ledger @lidge-jun
• #4221 docs(devlog): mechanize the lane tracking refresh and fix the wp4 merge gate @lidge-jun
• #4223 docs(devlog): fold the seven-lane feasibility audit into lane ownership and scope @lidge-jun
• #4226 fix(opencode-go): give sessionless requests an isolated session lane @lidge-jun
• #4227 fix(integrations): refuse a write that would take a config file from its owner @lidge-jun
• #4230 fix(codex): warm up new ChatGPT Free accounts instead of blaming the credential (#4126) @lidge-jun
• #4231 fix(bigmodel): admit the Responses Coding Plan preset to the quota reader @lidge-jun
• #4232 fix(responses): classify Codex WS failures instead of restating them @lidge-jun
• #4233 fix(codex): name the account that left the pool instead of only dropping it (#4212) @lidge-jun
• #4234 fix(qoder): refuse vendor CLI scaffolding in routed output @lidge-jun
• #4235 fix(update): support native pnpm self-updates @lidge-jun
• #4237 fix(stop): name the real refusal cause instead of asserting ownership @lidge-jun
• #4238 feat(codex): let an operator keep a downgraded account out of pool rotation (#4211) @lidge-jun
• #4239 docs(providers): state which account each provider login spends @lidge-jun
• #4240 fix(client): refuse a hub catalog the local Codex CLI cannot parse @lidge-jun
• #4241 docs(remote-hub): make the setup runnable on a fresh config and close the macOS data plane @lidge-jun
• #4244 provider: seed GLM-5.3-Flash on the BigModel Responses preset @lidge-jun
• #4246 client: report local Codex readiness instead of bare connected state @lidge-jun
• #4247 docs(i18n): make the remote hub guide runnable in every locale @lidge-jun
• #4248 pool: name the account when a refresh fails or its models vanish @lidge-jun
• #4249 fix(service): macOS repair no longer evicts a healthy launchd job, and verifies the one it loads @lidge-jun
• #4250 feat(server): one-port hub — same-port loopback companion listener, honest hub-gate messages @lidge-jun
• #4251 fix(claude): route the hub's own local clients through their two destinations (management ingress + loopback inference) @lidge-jun
• #4252 feat(hub): provision the data-plane token, add ocx hub invite, and one hub block in ocx status @lidge-jun
• #4254 docs(remote-hub): the one-port hub recipe, invite flow, and launchd semantics (en+ko, skill) @lidge-jun
• #4255 feat(client): a connected client reports the hub's state, not its own @lidge-jun
• #4257 fix(codex): emit max/ultra efforts unconditionally and expire stale clamp diagnostics @lidge-jun
• #4269 fix(responses): keep the retryable main-refresh refusal an overload, not a bad key @lidge-jun
• #4271 release: promote verified 2.51.0 product tree to main @lidge-jun