[WRONG BRANCH] release: 2.80.0

desktop/src-tauri/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "opencodex-desktop"
-version = "2.79.0"
+version = "2.80.0"
 description = "OpenCodex desktop shell"
 authors = ["OpenCodex contributors"]
 license = "MIT"

desktop/src-tauri/tauri.conf.json

@@ -1,7 +1,7 @@
 {
   "$schema": "https://schema.tauri.app/config/2",
   "productName": "OpenCodex",
-  "version": "2.79.0",
+  "version": "2.80.0",
   "identifier": "com.opencodex.desktop",
   "build": {
     "frontendDist": "../ui",

devlog/_plan/261006_lane_c_provider_compat/000_roadmap.md

@@ -0,0 +1,16 @@
+# 000 — Lane C (provider compat) roadmap
+
+Lane session 01a11129 (fork of coordinator 01a11110). Worktree /Users/jun/Developer/new/700_projects/opencodex/.tmp/lanes/sweep-c-providers,
+branch codex/sweep-c-providers at origin/dev f494403fba. This unit rides the lane's own PR; the coordinator owns devlog/_plan/261006_compat_hardening_sweep.
+
+| wp | Item | Plan | Exit |
+|---|---|---|---|
+| wp1 | roadmap | this file | audited, locked |
+| wp2 | PR #6657 (luvs01, Devin) | worktree sweep-c-providers-6657 on the PR head; merge origin/dev (no force push); gpt-6.1-sol adversarial review of src/adapters/devin.ts change against tests/providers/devin-chat-wire-fixes.test.ts; fold real defects as commits on the PR branch; bun test focused file, typecheck, structure:check; push; gh pr ready; exact-head CI | READY or NEEDS_HUMAN |
+| wp3 | PR #6654 (Ollama, branch on origin) | same shape in sweep-c-providers-6654 with focused commands: bun test tests/providers/ollama/ollama-native-tool-continuations.test.ts tests/providers/ollama/ollama-native-v4.test.ts tests/adapters/abort-race.test.ts; verify initial and rebuild 413, ordinary build error 400, cancellation precedence and zero upstream sends on refusal; also confirm the 413 mapping keeps ordinary build errors at 400 and that docs-site/structure text matches code; shared files structure/transports/responses.md and docs-site reference/adapters.md are B/C shared (second lander reconciles) | READY or NEEDS_HUMAN |
+| wp4 | issue #6674 | in the lane worktree: widen qwen38LeadingSystemTemplate in src/adapters/openai-chat/messages.ts from ^(?:Qwen/)? to ^(?:[^/]+/)* so provider-namespaced served ids (openai/Qwen3.8-27B, hosted_vllm/Qwen/Qwen3.8-27B) match while Qwen3.8-27B-FP8 and api.openai.com stay excluded; add positive cases openai/Qwen3.8-27B and hosted_vllm/Qwen/Qwen3.8-27B plus namespaced negative (openai/Qwen3.8-27B-FP8) and native OpenAI control (openai/Qwen3.8-27B on api.openai.com) to tests/adapters/openai/openai-chat-qwen38-leading-system.test.ts; update the owning structure doc if it states the matcher; one PR, Closes #6674 | READY |
+| wp4 | issue #6502 | no code change: the report is 2.76.0 and Antigravity effort families were regrouped in 2.77.0 (#6501). resolveAntigravityWireModelId resolves discovery mappings first, aliases second, identity last, so the 404 alone does not tell an endpoint, mapping or account cause apart. Comment asking for a current-version repro with a redacted discovered model list and the resolved wire id; record NEEDS_HUMAN for insufficient evidence | NEEDS_HUMAN |
+
+Verification per PR: focused bun test, bun run typecheck, bun run structure:check, no full local suite (four lanes share this Mac; coverage left to hosted CI), exact-head CI with only path/event-filtered skips.
+Bounds: tools gh/git/bun; writes only in the lane write scope plus own registry/structure/docs entries. Branch synchronization (git merge origin/dev into a PR head branch, then push) is lane-owned and granted by the packet; PR integration into dev (gh pr merge) is coordinator-only and never done here.
+

docs-site/src/content/docs/guides/claude-code.md

@@ -794,6 +794,19 @@ depend on them for long unattended runs.
 
 <!-- TODO(WP5 GUI): Add the sidecar settings-screen walkthrough after the GUI controls ship. -->
 
+## Client tools on Cursor routes
+
+When Claude Code uses a Cursor-backed model, OpenCodex preserves recognized client tool names
+such as `Bash`, `Read`, `Write`, `Edit`, `Grep`, `Glob`, and `Task` in the Cursor catalog, tool
+guidance, and replayed calls. This keeps the names consistent with Claude Code's instructions.
+Denied Cursor-native execution frames direct the model to the client tools available that turn.
+
+A mixed catalog containing a bare Codex `exec_command` or `shell_command` bridge uses wire
+aliases such as `ocx_client_Read` to avoid Cursor tool collisions; calls returned to Claude Code
+still use its original tool names. Namespaced MCP tools keep their namespaced identities.
+OpenCodex includes these aliases in Cursor's transport byte budget, so a tool whose definition
+would exceed the limit is omitted for that turn.
+
 ## Reasoning effort
 
 Claude Code's `/effort` setting is preserved across the adapter:
@@ -818,6 +831,7 @@ The proxy translates every Anthropic Messages API request into the Codex Respons
 | Assistant text | `output_text` |
 | Assistant `tool_use` | `function_call` (`input` → JSON-stringified `arguments`) |
 | User `tool_result` | `function_call_output` (`is_error` → `[tool error]` prefix) |
+| `tool_reference` in a tool result | Text `Tool loaded: <tool_name>` in the paired result; preserves loaded-tool names without adding declarations or enabling translated server-side deferral |
 | `thinking` / `redacted_thinking` replay | `reasoning` items with bounded `ocxr1` envelopes for signatures and redacted payloads |
 | Function tools | `{type: "function"}` (`web_search*` → `{type: "web_search"}`) |
 | `tool_choice` | `auto`→`auto`, `none`→`none`, `any`→`required`, named function→`{type:"function",name}`, hosted WebSearch/web_search→`{type:"web_search"}` |

docs-site/src/content/docs/guides/combos.md

@@ -474,9 +474,11 @@ The decision state is deliberately bounded: up to 500 characters of the current
 240-character previous-assistant tail, a 520-character latest-tool-output tail, the tool name, and
 boolean image/tool signals may be sent to the selected decision backend. It excludes credentials,
 request headers, raw image bytes, tool arguments, encrypted reasoning, and full conversation history.
-Use a decision method only for content you are willing to send to that backend. Recognized OpenCodex machine-context
-envelopes are removed from all three text samples, but ordinary assistant and tool-output text is
-not a secret scanner and may still contain sensitive content. TypeSafe states that Jev is not
+Use a decision method only for content you are willing to send to that backend. Recognized Codex machine-context
+envelopes and Claude Code `<system-reminder>` blocks are removed from all three text samples before
+clipping, so injected reminders do not displace the actual task. The original model request is
+unchanged. This is not a secret scanner: ordinary assistant and tool-output text may still contain
+sensitive content. TypeSafe states that Jev is not
 trained on customer requests, but its terms set no fixed retention period for submitted state and
 offer zero data retention only on enterprise plans
 ([models](https://docs.typesafe.ai/models), [legal](https://docs.typesafe.ai/legal)). TypeSafe
@@ -486,6 +488,13 @@ target/effort, a coarse decision gate, latency, optional confidence/probability,
 Automated tests use mocked TypeSafe responses plus a no-key fail-open smoke; a live TypeSafe decision
 requires an operator-supplied key and is not run implicitly.
 
+When JEV's selected effort is applied to the Responses child, request and attempt logs show a
+differing effort as `high->xhigh` or `high->low`. Later effort adjustments remain in the chain,
+such as `high->xhigh->medium`. An unchanged effort or a target without effort control keeps the
+existing label; fallback attempts do not inherit the initial JEV effort. With the opt-in
+`protocols.rollout.nativeChatCombos` path, native Chat children currently do not apply JEV's
+selected effort; their labels retain the caller's effort and any native child adjustments instead.
+
 After the Combo has served requests, open **Models → Combos → jev-auto → Stats** to inspect JEV's
 picks without replacing the normal model picker or Usage page. The tab separates backend-reported decision
 tokens from tokens reported by physical model sends, and shows decision gates, fail-open picks,

docs-site/src/content/docs/guides/desktop-app.md

@@ -71,10 +71,26 @@ registration, including installations under `Program Files`. Previously enabled
 registrations are updated once on launch. Startup entries you disabled in the tray
 or Task Manager remain disabled.
 
-## Startup safety on macOS
+## Startup safety on macOS and Linux
 
 Startup safety reports **Desktop app** protection when OpenCodex's recorded ownership,
 **Start at Login** registration, and live supervision of its bundled proxy all match.
+On Linux, the pinned autostart backend writes the login entry to
+`~/.config/autostart/OpenCodex.desktop`, even when `$XDG_CONFIG_HOME` is set. Startup
+safety reads that entry and the desktop install-id under `~/.config`. When
+`$XDG_CONFIG_HOME` points elsewhere, the login session searches a different autostart
+directory, so startup safety stays **At risk** instead of crediting the entry.
+The entry counts only while it is not marked `Hidden=true` or
+`X-GNOME-Autostart-enabled=false`, and has no `OnlyShowIn`, `NotShowIn`, or `TryExec`
+condition. Its `Exec` must be an unquoted absolute path to `opencodex-desktop`, without
+spaces, escapes, or field codes, followed by exactly `--autostart`. The resolved
+executable must still be named `opencodex-desktop`, with its bundled `ocx` beside it.
+Startup safety reads the full evidence chain twice and grants protection only when
+both reads agree.
+
+AppImage installations remain **At risk**: the autostart backend registers the outer
+AppImage path, while the live desktop process runs inside its mount. Startup safety
+cannot verify that relationship and does not credit AppImage protection.
 A missing or stale check remains **At risk**. If the desktop app owns the proxy but
 protection cannot be verified, reopen OpenCodex and check **Start at Login**. Service
 and launcher installation or repair stays disabled while that ownership remains;

docs-site/src/content/docs/ko/guides/claude-code.md

@@ -524,6 +524,7 @@ Claude Code의 `/effort` 설정은 어댑터에서도 유지돼요.
 | Assistant 텍스트 | `output_text` |
 | Assistant `tool_use` | `function_call`(`input` → JSON 문자열로 변환한 `arguments`) |
 | 사용자 `tool_result` | `function_call_output`(`is_error` → `[tool error]` 접두사) |
+| 도구 결과의 `tool_reference` | 짝지어진 결과에 `Tool loaded: <tool_name>` 텍스트로 보존합니다. 도구 선언을 추가하거나 번역 경로의 서버 측 지연 로딩을 활성화하지는 않습니다. |
 | `thinking` / `redacted_thinking` 재생 | 서명과 비공개 페이로드를 제한된 `ocxr1` 봉투에 담은 `reasoning` 항목 |
 | Function 도구 | `{type: "function"}`(`web_search*` → `{type: "web_search"}`) |
 | `tool_choice` | `auto`→`auto`, `none`→`none`, `any`→`required`, 이름 지정 함수→`{type:"function",name}`, 호스팅 WebSearch/web_search→`{type:"web_search"}` |

docs-site/src/content/docs/reference/adapters.md

@@ -55,7 +55,8 @@ transport; it does not infer subscription attribution from the inbound protocol.
   tool message as the anchor.
 - **Rewrites Codex's GPT-5 identity prompt** to a model-agnostic intro so routed models don't claim to
   be OpenAI.
-- For translated `Qwen3.8-27B` requests, a text-only developer reminder after the leading system
+- For translated `Qwen3.8-27B` requests (including gateway-namespaced ids such as
+  `openai/Qwen3.8-27B`), a text-only developer reminder after the leading system
   message stays in its conversation slot but is sent as `user`. The model's
   [chat template](https://huggingface.co/Qwen/Qwen3.8-27B/blob/1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0/chat_template.jinja)
   rejects later `system` messages and does not accept `developer`, while later `user` messages
@@ -543,7 +544,13 @@ compatibility pair: `agent.v1.AgentService/RunSSE` for server output and
   Foreground `shellArgs` and `shellStreamArgs` are an exception: both are rejected before spawn
   on every platform until kernel-backed descendant ownership is available. Use client shell tools;
   background-shell execution and other native operations retain their existing policy.
-- The denial reply is a silent redirect whose wording follows the request catalog.
+- The denial reply follows the request catalog. Explicit Claude-family Cursor models receive
+  factual wording that names the available tools without instructions to hide or avoid discussing
+  the redirect. Factual routing commentary can precede the next tool call without ending the turn.
+  When no client or configured MCP tools are available, Claude models are asked to answer without
+  tools or report the limitation instead of calling nonexistent shell/edit tools.
+  Other models, including Auto/default, retain the existing silent-redirect wording and commentary
+  guard. Tool availability, execution policy, and approval/sandbox behavior are unchanged.
   In code mode — a freeform unified `exec` and no bare shell bridge — the redirect points inside `exec`, where
   shell, file, search, and fetch are nested `tools.<name>(...)` helpers of the JavaScript cell,
   and never recommends the top-level shell bridge code mode does not expose. A flat catalog that

docs-site/src/content/docs/reference/architecture.md

@@ -47,8 +47,9 @@ full owner inventory for the Responses surface.
 `server/index/serve-options.ts` owns the HTTP boundary and delegates the Responses data plane to
 the `server/responses.ts` facade and its `server/responses/*.ts` modules:
 
-1. `server/index/serve-options.ts` applies CORS and API authentication, rejects new work while draining, and
-   records request lifecycle metadata. It serves `GET /v1/models`, `POST /v1/responses`,
+1. `server/index/serve-options.ts` applies CORS and API authentication, rejects new work while draining
+   with HTTP 503, JSON `error.code: "server_restarting"`, and `Retry-After: 5`, and
+   records request lifecycle metadata. Drain responses use the receiving listener's CORS policy. It serves `GET /v1/models`, `POST /v1/responses`,
    `POST /v1/responses/compact`, `POST /v1/images/generations` / `POST /v1/images/edits`
    (relayed to an OpenAI-family upstream by `server/images.ts` for codex's built-in `image_gen`
    tool), `POST /v1/live` / `POST /v1/realtime/calls` (ChatGPT / Codex App voice and OpenAI

docs-site/src/content/docs/reference/cli/lifecycle.md

@@ -879,6 +879,11 @@ Migration restores the recorded native launcher without replacing a newer launch
 then installs the private wrapper. If private installation fails after native restoration, the native
 launcher stays restored and the operation can be retried. Missing or unusable native launchers require
 package-manager repair; OpenCodex does not guess another installation or rewrap the manager's path.
+Migration also refuses when the recorded `codex-shim.json` is not a regular file owned by you or is
+group- or world-writable. Older releases wrote that file with the process umask, so a umask of `002`
+left it at mode `0664`. The refusal names the file and the `chmod 600` that clears only the
+group- or world-writable refusal. Symlink and foreign-owner refusals remain; OpenCodex does not
+change the permissions itself.
 
 Launcher installation alone does not prove that Codex requests will use OpenCodex. After a runnable
 install, the command checks the current Codex routing and reports a warning instead of a green result

docs-site/src/content/docs/reference/management-api.md

@@ -662,6 +662,12 @@ whether to star the repository.
 
 ### System lifecycle
 
+During a restart drain, new data-plane requests receive HTTP 503 with JSON
+`error.type: "server_error"`, `error.code: "server_restarting"`, and the message
+"OpenCodex is restarting; retry this request." Responses retain `Retry-After: 5`
+and the receiving listener's CORS policy. This code lets every Codex version retry
+the 503 without reporting model capacity; provider overload errors retain their separate mapping.
+
 | Method and path | Purpose | Notable errors |
 | --- | --- | --- |
 | `GET /api/system/memory` | Return scalar process, heap, stream, response-state, watchdog, and active-turn metrics. Response-state diagnostics include spill-write status, consecutive failures, fixed privacy-safe failure class, and last failure/success timestamps. `spillLastWriteFailureOrigin` is `retry_returned_timeout`, `timeout_memo_refusal`, or null; cumulative `spillAclRetryReturnedTimeouts` and `spillAclTimeoutMemoRefusals` count terminal failed publications. See [Windows spill diagnostics](/troubleshooting/windows-memory/) for process-local semantics. Raw errors and paths are never returned. | — |

docs-site/src/content/docs/troubleshooting/windows-memory.md

@@ -78,7 +78,8 @@ runtime the leak itself remains an upstream problem:
   continuation-size fields from this endpoint and offers a confirm-gated
   **Drain & restart** action: it shows
   the current active-turn count, waits up to 60s for active turns (reusing
-  the existing 503 + `Retry-After` drain), then aborts any remaining turns.
+  the existing HTTP 503 JSON `server_restarting` + `Retry-After: 5` drain),
+  then aborts any remaining turns.
   The running proxy owns restart authorization and drain coordination, then
   exits; an installed service manager launches the replacement when applicable.
   The action reports success only after a different, identity-verified process

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@bitkyc08/opencodex",
-  "version": "2.79.0",
+  "version": "2.80.0",
   "description": "Universal provider proxy for OpenAI Codex & Claude Code — use any LLM with Codex CLI/App/SDK and Claude Code",
   "type": "module",
   "main": "./bin/package-main.mjs",