omnilane 0.7.1 → 0.8.1

package/CHANGELOG.md

@@ -6,6 +6,37 @@ semantic version tags.
 
 ## [Unreleased]
 
+## [0.8.1] - 2026-07-18
+
+### Added
+
+- Claude Code plugin `SessionStart` hook (`hooks/hooks.json`): plugin installs
+  now auto-inject the routing reminder (`hooks/routing-instruction.md`) at
+  session open (`startup|resume|clear`) via `${CLAUDE_PLUGIN_ROOT}`, with no
+  edit to the user's `~/.claude/CLAUDE.md`. The `install.sh` instruction-file
+  reminder remains the path for the other CLIs.
+
+## [0.8.0] - 2026-07-18
+
+### Added
+
+- Two new dispatch vendors: `kimi` (Moonshot Kimi Code CLI, binary `kimi`,
+  override `KIMI_BIN`) and `qwen` (Alibaba Qwen Code CLI, binary `qwen`,
+  override `QWEN_BIN`). Both runners follow the uniform
+  `MODE WORKDIR MODEL EFFORT PROMPT_FILE OUTPUT_FILE` contract: advise stays
+  read-only (kimi `--plan`; qwen default approvals), work auto-approves
+  (kimi `-y`; qwen `--approval-mode yolo`), API-key env is stripped so the
+  CLIs use their own subscription logins, and empty output is a failure.
+- `coding-overflow` now falls back grok → kimi → qwen before `off`, so the
+  quota relief valve works with any one of the three vendors installed.
+- Configurator and bash completion know the new vendors; `--vendor kimi|qwen`
+  pins them directly. The `vote` panel still accepts the original four
+  vendors only.
+- Both new runners are contract-tested (argv construction, exit codes,
+  empty-output failure) against fake binaries; real-model end-to-end runs
+  are pending community feedback — failures surface loudly, never as a
+  silent rc=0.
+
 ## [0.7.1] - 2026-07-18
 
 ### Changed
@@ -208,8 +239,11 @@ semantic version tags.
 - Initial shared routing table, cross-vendor dispatcher, runners, installer,
   and baseline lint fixes.
 
-[Unreleased]: https://github.com/Seraphim0916/omnilane/compare/v0.7.1...HEAD
+[Unreleased]: https://github.com/Seraphim0916/omnilane/compare/v0.8.1...HEAD
+
+[0.8.1]: https://github.com/Seraphim0916/omnilane/compare/v0.8.0...v0.8.1
 
+[0.8.0]: https://github.com/Seraphim0916/omnilane/compare/v0.7.1...v0.8.0
 [0.7.1]: https://github.com/Seraphim0916/omnilane/compare/v0.7.0...v0.7.1
 [0.7.0]: https://github.com/Seraphim0916/omnilane/compare/v0.6.0...v0.7.0
 [0.6.0]: https://github.com/Seraphim0916/omnilane/compare/v0.5.1...v0.6.0

package/README.ja.md

@@ -20,6 +20,27 @@
 
 ---
 
+## v0.8.1 の新機能
+
+- **Claude Code プラグインがルーティングリマインダーを自動読み込み** —
+  プラグインに `SessionStart` フック(`hooks/hooks.json`)を同梱。
+  セッション開始時(`startup|resume|clear`)にルーティングリマインダーを
+  自動注入するため、`~/.claude/CLAUDE.md` を編集せずプラグインの
+  インストールだけで有効になります。他の CLI は引き続き `install.sh` の
+  指示ファイル方式です。
+
+## v0.8.0 の新機能
+
+- **2 つの新ディスパッチベンダー** — `kimi`(Moonshot Kimi Code CLI)と
+  `qwen`(Alibaba Qwen Code CLI)が統一 runner 契約で加わりました:
+  advise は読み取り専用、work は自動承認、API キー環境変数を除去して
+  CLI 自身のサブスクリプションログインを使用、空出力は明示的な失敗。
+  `--vendor kimi|qwen` で直接指名できます。
+- **coding-overflow にフォールバックチェーン** — クォータ逃し弁が
+  grok → kimi → qwen → `off` の順にフォールバックし、3 ベンダーの
+  いずれか 1 つで動作します。runner はフェイクバイナリで契約テスト済み。
+  実モデルでの報告を歓迎します。
+
 ## v0.7.1 の新機能
 
 - **ルーティング表を更新(2026-07 モデルデータ)** — hardest-coding の第一候補を
@@ -128,7 +149,7 @@ flowchart LR
 | 📚 long-context | Gemini 3.1 Pro (High) | Claude Opus 4.8 (high) | 100 万トークン級の長文統合——分析専用、agentic ループ禁止 |
 | ⚡ fast-agentic | Gemini 3.5 Flash (High) | GPT-5.6 Luna (high) | 高速なマルチステップ agentic ループ、マルチモーダル確認 |
 | 📡 live-search | Grok 4.5 | —(off) | リアルタイム X/ウェブ検索とソーシャル文脈 |
-| 🚰 coding-overflow | Grok 4.5 | —(off) | Codex クォータ逼迫時の中級コーディング逃し弁 |
+| 🚰 coding-overflow | Grok 4.5 | Kimi K3 → Qwen3 Coder Plus | Codex クォータ逼迫時の中級コーディング逃し弁 |
 | 🗳️ arbitrate | off(オプトイン) | — | 内蔵オピニオンパネル(重大な判断用)——デフォルト無効。`routing.local.yaml` で有効化;投票者×ラウンドごとに 1 コール消費 |
 
 **バックアップ**はチェーンの次の候補——第一候補のベンダー CLI が未インストールの
@@ -229,7 +250,9 @@ CLI バージョンにより異なる場合あり)へマーカー付きの可逆
 `./install.sh --uninstall` を実行します。
 
 - **Claude Code**:プラグインとしてインストール(`/route`、`/route-jobs`
-  コマンド付き)、または `skills/omnilane` を `~/.claude/skills/` へ。
+  コマンドに加え、セッション開始時にルーティングリマインダーを自動注入する
+  `SessionStart` フック付き。CLAUDE.md の編集は不要)、または
+  `skills/omnilane` を `~/.claude/skills/` へ。
 - **Codex**:`skills/omnilane` を `~/.codex/skills/` へ配置/リンク。
 - **Grok Build**:`grok plugin install <このリポジトリ> --trust`
 - **Antigravity**:`agy plugin install <このリポジトリ>`(先に

package/README.ko.md

@@ -20,6 +20,26 @@
 
 ---
 
+## v0.8.1 새 기능
+
+- **Claude Code 플러그인이 라우팅 리마인더를 자동 로드** — 플러그인에
+  `SessionStart` 훅(`hooks/hooks.json`)이 포함되어 세션 시작 시
+  (`startup|resume|clear`) 라우팅 리마인더를 자동 주입합니다.
+  `~/.claude/CLAUDE.md` 수정 없이 플러그인 설치만으로 적용됩니다.
+  다른 CLI는 기존대로 `install.sh` 지침 파일 방식을 사용합니다.
+
+## v0.8.0 새 기능
+
+- **새 디스패치 벤더 2종** — `kimi`(Moonshot Kimi Code CLI)와
+  `qwen`(Alibaba Qwen Code CLI)이 통일 runner 계약으로 합류:
+  advise 는 읽기 전용, work 는 자동 승인, API 키 환경 변수를 제거해
+  CLI 자체 구독 로그인을 사용, 빈 출력은 명시적 실패.
+  `--vendor kimi|qwen` 으로 직접 지정할 수 있습니다.
+- **coding-overflow 폴백 체인** — 쿼터 안전 밸브가 grok → kimi → qwen
+  → `off` 순으로 폴백해 세 벤더 중 하나만 설치돼 있어도 동작합니다.
+  runner 는 페이크 바이너리로 계약 테스트 완료. 실제 모델 사용 보고를
+  환영합니다.
+
 ## v0.7.1 새 기능
 
 - **라우팅 테이블 갱신(2026-07 모델 데이터)** — hardest-coding 1순위를
@@ -128,7 +148,7 @@ flowchart LR
 | 📚 long-context | Gemini 3.1 Pro (High) | Claude Opus 4.8 (high) | 100만 토큰급 장문 통합——분석 전용, agentic 루프 금지 |
 | ⚡ fast-agentic | Gemini 3.5 Flash (High) | GPT-5.6 Luna (high) | 빠른 멀티스텝 agentic 루프, 멀티모달 확인 |
 | 📡 live-search | Grok 4.5 | —(off) | 실시간 X/웹 검색과 소셜 맥락 |
-| 🚰 coding-overflow | Grok 4.5 | —(off) | Codex 쿼터 소진 시 중급 코딩 안전 밸브 |
+| 🚰 coding-overflow | Grok 4.5 | Kimi K3 → Qwen3 Coder Plus | Codex 쿼터 소진 시 중급 코딩 안전 밸브 |
 | 🗳️ arbitrate | off(옵트인) | — | 내장 의견 패널(중대한 결정용)——기본 비활성. `routing.local.yaml` 에서 활성화;투표자×라운드마다 1콜 소모 |
 
 **백업**은 체인의 다음 후보입니다——1순위 벤더 CLI 가 설치되지 않았을 때
@@ -224,8 +244,9 @@ omnilane ui stop     # 정상 중지
 설치 프로그램이 소유한 링크와 표시된 알림을 되돌리려면
 `./install.sh --uninstall` 을 실행합니다.
 
-- **Claude Code**: 플러그인으로 설치(`/route`, `/route-jobs` 명령 포함),
-  또는 `skills/omnilane` 을 `~/.claude/skills/` 에 배치.
+- **Claude Code**: 플러그인으로 설치(`/route`, `/route-jobs` 명령과 함께
+  세션 시작 시 라우팅 리마인더를 자동 주입하는 `SessionStart` 훅 포함,
+  CLAUDE.md 수정 불필요), 또는 `skills/omnilane` 을 `~/.claude/skills/` 에 배치.
 - **Codex**: `skills/omnilane` 을 `~/.codex/skills/` 에 배치/링크.
 - **Grok Build**: `grok plugin install <이 저장소> --trust`
 - **Antigravity**: `agy plugin install <이 저장소>`(먼저

package/README.md

@@ -20,6 +20,26 @@ Every subtask goes to the model that is actually best at it — across<br/>
 
 ---
 
+## What's new in v0.8.1
+
+- **Claude Code plugin auto-loads the routing reminder** — the plugin now
+  ships a `SessionStart` hook (`hooks/hooks.json`) that injects the routing
+  reminder at session open (`startup|resume|clear`), so plugin installs get
+  the persistent reminder with no edit to `~/.claude/CLAUDE.md`. The
+  `install.sh` instruction-file reminder still covers the other CLIs.
+
+## What's new in v0.8.0
+
+- **Two new dispatch vendors** — `kimi` (Moonshot Kimi Code CLI) and `qwen`
+  (Alibaba Qwen Code CLI) join the vendor set with the uniform runner
+  contract: advise stays read-only, work auto-approves, API-key env is
+  stripped so the CLIs use their own subscription logins, and empty output
+  is a loud failure. Pin them with `--vendor kimi|qwen`.
+- **coding-overflow grows a chain** — the quota relief valve now falls back
+  grok → kimi → qwen before `off`, so it works with any one of the three
+  vendors installed. Runners are contract-tested against fake binaries;
+  real-model reports welcome.
+
 ## What's new in v0.7.1
 
 - **Routing refresh (2026-07 model data)** — hardest-coding now dispatches
@@ -130,7 +150,7 @@ flowchart LR
 | 📚 long-context | Gemini 3.1 Pro (High) | Claude Opus 4.8 (high) | 1M-token synthesis — analysis only, never agentic loops |
 | ⚡ fast-agentic | Gemini 3.5 Flash (High) | GPT-5.6 Luna (high) | Fast multi-step agentic loops, multimodal checks |
 | 📡 live-search | Grok 4.5 | — (off) | Realtime X/web search and social context |
-| 🚰 coding-overflow | Grok 4.5 | — (off) | Codex-quota relief valve for mid-tier coding |
+| 🚰 coding-overflow | Grok 4.5 | Kimi K3 → Qwen3 Coder Plus | Codex-quota relief valve for mid-tier coding |
 | 🗳️ arbitrate | off (opt-in vote panel) | — | Built-in opinion panel for big calls — disabled by default; enable it in `routing.local.yaml`, one call per voter per round |
 
 The **backup** is the next candidate in the lane's `routing.yaml` chain — what
@@ -231,7 +251,9 @@ Rollback the installer-owned links and marked reminders with
 `./install.sh --uninstall`.
 
 - **Claude Code**: install as a plugin (ships the skill + `/route`,
-  `/route-jobs` commands), or drop `skills/omnilane` into `~/.claude/skills/`.
+  `/route-jobs` commands, and a `SessionStart` hook that auto-injects the
+  routing reminder at session open — no CLAUDE.md edit needed), or drop
+  `skills/omnilane` into `~/.claude/skills/`.
 - **Codex**: drop/symlink `skills/omnilane` into `~/.codex/skills/`.
 - **Grok Build**: `grok plugin install <this repo> --trust`
 - **Antigravity**: `agy plugin install <this repo>` (check first with

package/README.zh-CN.md

@@ -20,6 +20,23 @@
 
 ---
 
+## v0.8.1 新功能
+
+- **Claude Code 插件开场自动载入路由提醒** — 插件新增 `SessionStart`
+  hook(`hooks/hooks.json`),在会话开始(`startup|resume|clear`)时自动
+  注入路由提醒,安装插件即生效,无需修改 `~/.claude/CLAUDE.md`。其他
+  CLI 仍使用 `install.sh` 的指令文件提醒。
+
+## v0.8.0 新功能
+
+- **两个新派工 vendor** — `kimi`(Moonshot Kimi Code CLI)与 `qwen`
+  (Alibaba Qwen Code CLI)加入,沿用统一 runner 契约:advise 只读、
+  work 自动批准、剥除 API key 环境变量改用 CLI 自身订阅登录、空输出
+  视为失败。可用 `--vendor kimi|qwen` 直接指定。
+- **coding-overflow 新增备选链** — 额度溢流道改为 grok → kimi → qwen
+  再到 `off`,三家装任一家即可用。runner 以假可执行文件完成契约测试;
+  欢迎反馈真实模型实测结果。
+
 ## v0.7.1 新功能
 
 - **路由表更新(2026-07 模型数据)** — hardest-coding 首选改为 GPT-5.6 Sol
@@ -117,7 +134,7 @@ flowchart LR
 | 📚 long-context | Gemini 3.1 Pro (High) | Claude Opus 4.8 (high) | 百万 token 长文整合——仅限分析,不派 agentic 长链 |
 | ⚡ fast-agentic | Gemini 3.5 Flash (High) | GPT-5.6 Luna (high) | 快速多步骤 agentic 循环、多模态检查 |
 | 📡 live-search | Grok 4.5 | —(off) | 实时 X/网络搜索与社群脉络 |
-| 🚰 coding-overflow | Grok 4.5 | —(off) | Codex 额度吃紧时的中量级编码溢流道;事实性声明须另行查证 |
+| 🚰 coding-overflow | Grok 4.5 | Kimi K3 → Qwen3 Coder Plus | Codex 额度吃紧时的中量级编码溢流道;事实性声明须另行查证 |
 | 🗳️ arbitrate | off(可选评审团) | — | 内置意见评审团,重大决定用——默认关闭,要用在 `routing.local.yaml` 打开;每评审每轮烧一次额度 |
 
 **备选模型**是候选链的下一位——首选那家的厂商 CLI 没装时,派发就降到它。
@@ -202,7 +219,8 @@ Esc。服务器发送事件(SSE)会实时更新，又不会重建当前聚焦的
 `--dry-run`，可先预览每个由这份 checkout 拥有的文件动作。
 要回滚安装器拥有的链接与标记提示，执行 `./install.sh --uninstall`。
 
-- **Claude Code**:以插件安装(附 `/route`、`/route-jobs` 命令),
+- **Claude Code**:以插件安装(附 `/route`、`/route-jobs` 命令,并内置
+  `SessionStart` hook,会话开始时自动注入路由提醒,无需修改 CLAUDE.md),
   或把 `skills/omnilane` 放进 `~/.claude/skills/`。
 - **Codex**:把 `skills/omnilane` 放进或链接到 `~/.codex/skills/`。
 - **Grok Build**:`grok plugin install <本仓库路径> --trust`

package/README.zh-TW.md

@@ -20,6 +20,23 @@
 
 ---
 
+## v0.8.1 新功能
+
+- **Claude Code 外掛開場自動載入路由提醒** — 外掛新增 `SessionStart`
+  hook(`hooks/hooks.json`),於開場(`startup|resume|clear`)自動注入
+  路由提醒,裝外掛即生效,不必修改 `~/.claude/CLAUDE.md`。其他 CLI
+  仍走 `install.sh` 的指令檔提醒。
+
+## v0.8.0 新功能
+
+- **兩個新派工 vendor** — `kimi`(Moonshot Kimi Code CLI)與 `qwen`
+  (Alibaba Qwen Code CLI)加入,沿用統一 runner 契約:advise 唯讀、
+  work 自動核准、剝除 API key 環境變數改用 CLI 自身訂閱登入、空輸出
+  視為失敗。可用 `--vendor kimi|qwen` 直接點名。
+- **coding-overflow 長出備援鏈** — 額度溢流道改為 grok → kimi → qwen
+  再到 `off`,三家裝任一家即可用。runner 以假執行檔完成契約測試;
+  歡迎回報真實模型實測結果。
+
 ## v0.7.1 新功能
 
 - **路由表更新(2026-07 模型數據)** — hardest-coding 首選改為 GPT-5.6 Sol
@@ -117,7 +134,7 @@ flowchart LR
 | 📚 long-context | Gemini 3.1 Pro (High) | Claude Opus 4.8 (high) | 百萬 token 長文整合——僅限分析,不派 agentic 長鏈 |
 | ⚡ fast-agentic | Gemini 3.5 Flash (High) | GPT-5.6 Luna (high) | 快速多步驟 agentic 迴圈、多模態檢查 |
 | 📡 live-search | Grok 4.5 | —(off) | 即時 X/網路搜尋與社群脈絡 |
-| 🚰 coding-overflow | Grok 4.5 | —(off) | Codex 額度吃緊時的中量級編碼溢流道;事實性宣稱須另行查證 |
+| 🚰 coding-overflow | Grok 4.5 | Kimi K3 → Qwen3 Coder Plus | Codex 額度吃緊時的中量級編碼溢流道;事實性宣稱須另行查證 |
 | 🗳️ arbitrate | off(選配評審團) | — | 內建意見評審團,重大決定用——預設關閉,要用在 `routing.local.yaml` 開;每評審每輪燒一次額度 |
 
 **備選模型**是候選鏈的下一位——首選那家的廠商 CLI 沒裝時,派工就降到它。
@@ -203,7 +220,8 @@ omnilane ui stop     # 正常停止
 
 要回滾安裝器擁有的連結與標記提示，執行 `./install.sh --uninstall`。
 
-- **Claude Code**:以外掛安裝(附 `/route`、`/route-jobs` 指令),
+- **Claude Code**:以外掛安裝(附 `/route`、`/route-jobs` 指令,並內建
+  `SessionStart` hook,開場自動注入路由提醒,不必修改 CLAUDE.md),
   或把 `skills/omnilane` 放進 `~/.claude/skills/`。
 - **Codex**:把 `skills/omnilane` 放進或連結到 `~/.codex/skills/`。
 - **Grok Build**:`grok plugin install <本 repo 路徑> --trust`

package/VERSION

@@ -1 +1 @@
-0.7.1
+0.8.1

package/completions/omnilane.bash

@@ -64,7 +64,7 @@ _omnilane() {
       route|dispatch)
         case "$prev" in
           --mode) words="advise work" ;;
-          --vendor) words="codex claude grok gemini exec vote" ;;
+          --vendor) words="codex claude grok gemini kimi qwen exec vote" ;;
           --effort) words="low medium high xhigh max" ;;
           --workdir)
             COMPREPLY=()

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "omnilane",
-  "version": "0.7.1",
+  "version": "0.8.1",
   "description": "One routing table, every harness — classify subtasks into lanes and dispatch each lane to the best vendor's agentic CLI (Codex, Claude, Gemini, Grok) using your existing subscription logins.",
   "bin": {
     "omnilane": "bin/omnilane"

package/routing.yaml

@@ -22,7 +22,7 @@ ui-draft:        codex gpt-5.6-sol xhigh | claude claude-opus-4-8 high  # only w
 long-context:    gemini "Gemini 3.1 Pro (High)" - | claude claude-opus-4-8 high | codex gpt-5.6-sol high  # 1M-token synthesis ONLY — never agentic loops (AA agentic 21.4)
 fast-agentic:    gemini "Gemini 3.5 Flash (High)" - | codex gpt-5.6-luna high  # speed + agentic tool loops
 live-search:     grok grok-4.5 - | off          # native X/web search lane; no real substitute
-coding-overflow: grok grok-4.5 - | off          # codex-quota relief valve: mid-tier coding (AA 72.45); verify factual claims (hallucination 0.54)
+coding-overflow: grok grok-4.5 - | kimi kimi-k3 - | qwen qwen3-coder-plus - | off  # codex-quota relief valve: mid-tier coding; verify factual claims (grok hallucination-prone). kimi/qwen model fields are CLI aliases — adjust to your login.
 arbitrate:       off - -                        # opinion panel is OPT-IN: it costs one call per voter per round.
                                                 # Enable: `arbitrate: vote codex,claude,grok -` (any 1-4 of codex/claude/grok/gemini)
                                                 # Debate round (each voter rebuts the others): set the effort field to 2.

package/scripts/configure.sh

@@ -16,6 +16,8 @@ CLAUDE_MODELS=("claude-opus-4-8" "claude-fable-5" "claude-sonnet-5" "claude-haik
 CLAUDE_EFFORTS=("max" "xhigh" "high" "medium" "low" "-")
 GEMINI_MODELS=("Gemini 3.1 Pro (High)" "Gemini 3.1 Pro (Low)" "Gemini 3.5 Flash (High)" "Gemini 3.5 Flash (Medium)" "Gemini 3.5 Flash (Low)")
 GROK_MODELS=("grok-4.5" "grok-4.3")
+KIMI_MODELS=("kimi-k3" "kimi-k2.7-code")
+QWEN_MODELS=("qwen3-coder-plus" "qwen3-coder-flash")
 
 custom_value_is_safe() {
   case "$1" in
@@ -74,7 +76,7 @@ while true; do
     echo "$(msgf cfg_pick_range "${#LANES[@]}")"; continue
   fi
   lane="${LANES[$((n - 1))]}"
-  vendor="$(pick "$(msgf cfg_vendor_for "$lane")" codex claude grok gemini "vote (multi-model panel)" "exec (your own script/gate)" off)"
+  vendor="$(pick "$(msgf cfg_vendor_for "$lane")" codex claude grok gemini kimi qwen "vote (multi-model panel)" "exec (your own script/gate)" off)"
   [[ "$vendor" == exec* ]] && vendor="exec"
   [[ "$vendor" == vote* ]] && vendor="vote"
   if [[ "$vendor" == "off" ]]; then
@@ -86,6 +88,8 @@ while true; do
     claude) model="$(pick "$(msg cfg_model)" "${CLAUDE_MODELS[@]}")"; effort="$(pick "$(msg cfg_effort)" "${CLAUDE_EFFORTS[@]}")" ;;
     gemini) model="$(pick "$(msg cfg_model)" "${GEMINI_MODELS[@]}")"; effort="-" ;;
     grok)   model="$(pick "$(msg cfg_model)" "${GROK_MODELS[@]}")";   effort="-" ;;
+    kimi)   model="$(pick "$(msg cfg_model)" "${KIMI_MODELS[@]}")";   effort="-" ;;
+    qwen)   model="$(pick "$(msg cfg_model)" "${QWEN_MODELS[@]}")";   effort="-" ;;
     vote)   while true; do
               count="$(pick "$(msg cfg_voters_count)" "1" "2" "3" "4")"
               [[ "$count" =~ ^[1-4]$ ]] && break

package/scripts/dispatch.sh

@@ -52,7 +52,7 @@ flags:
                          before any provider call or job state
   --mode advise|work     advise (read-only, default) or work (may edit files)
   --workdir DIR          working directory handed to the vendor CLI
-  --vendor V             pin one configured vendor (codex|claude|grok|gemini)
+  --vendor V             pin one configured vendor (codex|claude|grok|gemini|kimi|qwen)
   --model M              override the routed model
   --effort E             override the routed effort
   --timeout SECONDS      cap each CLI call (default 600)
@@ -363,7 +363,7 @@ validate_routing() {
           lane_invalid=1
           break
         fi
-        if ! [[ "$vendor" =~ ^(codex|claude|grok|gemini|exec|off)$ ]]; then
+        if ! [[ "$vendor" =~ ^(codex|claude|grok|gemini|kimi|qwen|exec|off)$ ]]; then
           printf 'FAIL %s unknown-vendor=%s\n' "$lane" "$vendor"
           lane_invalid=1
           break
@@ -489,8 +489,8 @@ TASK="$2"
 # A typo like --mode advice must not fall through to the write-enabled branch.
 [[ "$MODE" == "advise" || "$MODE" == "work" ]] || { echo "omnilane: invalid --mode (advise|work)" >&2; exit 2; }
 if [[ -n "$OVERRIDE_VENDOR" ]] &&
-   ! [[ "$OVERRIDE_VENDOR" =~ ^(codex|claude|grok|gemini)$ ]]; then
-  echo "omnilane: invalid vendor (codex|claude|grok|gemini)" >&2
+   ! [[ "$OVERRIDE_VENDOR" =~ ^(codex|claude|grok|gemini|kimi|qwen)$ ]]; then
+  echo "omnilane: invalid vendor (codex|claude|grok|gemini|kimi|qwen)" >&2
   exit 2
 fi
 

package/scripts/jobs.sh

@@ -132,7 +132,7 @@ read_job_pid() {
 
 parse_stats_metadata() {
   local value="$1"
-  local metadata_re='^\{"lane":"([a-z][a-z0-9-]*)","vendor":"(codex|claude|grok|gemini|exec)"(,|})'
+  local metadata_re='^\{"lane":"([a-z][a-z0-9-]*)","vendor":"(codex|claude|grok|gemini|kimi|qwen|exec)"(,|})'
   STATS_LANE=""
   STATS_VENDOR=""
   [[ "$value" =~ $metadata_re ]] || return 1
@@ -143,7 +143,7 @@ parse_stats_metadata() {
 parse_audit_metadata() {
   local value="$1" metadata_re
   local json_string='([^"\\]|\\["\\/bfnrt]|\\u[0-9a-fA-F]{4})*'
-  metadata_re="^\\{\"lane\":\"[a-z][a-z0-9-]*\",\"vendor\":\"(codex|claude|grok|gemini|exec)\",\"model\":\"${json_string}\",\"effort\":\"${json_string}\",\"timeout\":(0|[1-9][0-9]*),\"job_timeout\":(null|0|[1-9][0-9]*),\"mode\":\"(advise|work)\",\"workdir\":\"${json_string}\",\"candidate\":\"[1-9][0-9]*/[1-9][0-9]*\",\"started\":\"[0-9]{4}-[0-9]{2}-[0-9]{2}T[0-9]{2}:[0-9]{2}:[0-9]{2}Z\"\\}$"
+  metadata_re="^\\{\"lane\":\"[a-z][a-z0-9-]*\",\"vendor\":\"(codex|claude|grok|gemini|kimi|qwen|exec)\",\"model\":\"${json_string}\",\"effort\":\"${json_string}\",\"timeout\":(0|[1-9][0-9]*),\"job_timeout\":(null|0|[1-9][0-9]*),\"mode\":\"(advise|work)\",\"workdir\":\"${json_string}\",\"candidate\":\"[1-9][0-9]*/[1-9][0-9]*\",\"started\":\"[0-9]{4}-[0-9]{2}-[0-9]{2}T[0-9]{2}:[0-9]{2}:[0-9]{2}Z\"\\}$"
   [[ "$value" =~ $metadata_re ]]
 }
 
@@ -252,7 +252,7 @@ case "${1:-}" in
     read_public_metadata "$JOB_DIR/meta.json" || {
       die 1 "cannot retry: unreadable job metadata"
     }
-    retry_re='^\{"lane":"([a-z][a-z0-9-]*)","vendor":"(codex|claude|grok|gemini|exec)","model":"([^"\\]*)","effort":"([^"\\]*)","timeout":([1-9][0-9]*),"job_timeout":([1-9][0-9]*|null),"mode":"(advise|work)","workdir":"([^"\\]*)",'
+    retry_re='^\{"lane":"([a-z][a-z0-9-]*)","vendor":"(codex|claude|grok|gemini|kimi|qwen|exec)","model":"([^"\\]*)","effort":"([^"\\]*)","timeout":([1-9][0-9]*),"job_timeout":([1-9][0-9]*|null),"mode":"(advise|work)","workdir":"([^"\\]*)",'
     [[ "$PUBLIC_METADATA" =~ $retry_re ]] || {
       die 1 "cannot retry: job metadata is not safely parseable"
     }

package/scripts/lib/common.sh

@@ -70,6 +70,8 @@ vendor_bin() { # vendor -> binary (honors local.sh overrides)
     claude) echo "${CLAUDE_BIN:-claude}" ;;
     grok)   echo "${GROK_BIN:-grok}" ;;
     gemini) echo "${AGY_BIN:-agy}" ;;
+    kimi)   echo "${KIMI_BIN:-kimi}" ;;
+    qwen)   echo "${QWEN_BIN:-qwen}" ;;
     *)      echo "" ;;
   esac
 }

package/scripts/runners/run-kimi.sh

@@ -0,0 +1,53 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+# omnilane runner: Moonshot Kimi Code CLI
+#
+# Usage: run-kimi.sh MODE WORKDIR MODEL EFFORT PROMPT_FILE OUTPUT_FILE
+#   MODE = advise (read-only, plan mode) | work (may edit files in WORKDIR)
+#
+# MODEL is a Kimi Code model alias from the user's config.toml (e.g. kimi-k3);
+# an unknown alias fails at the CLI, not here. Kimi has no reasoning-effort
+# knob, so EFFORT is accepted for interface parity only.
+
+source "$(dirname "${BASH_SOURCE[0]}")/../lib/common.sh"
+
+MODE="$1"; WORKDIR="$2"; MODEL="$3"; EFFORT="$4"; PROMPT_FILE="$5"; OUTPUT_FILE="$6"
+: "$EFFORT" # parity with the uniform runner interface
+
+KIMI_BIN="${KIMI_BIN:-kimi}"
+RUN_TIMEOUT="${OMNILANE_TIMEOUT:-600}"
+
+truncate_payload "$PROMPT_FILE" 140000
+
+# --plan starts plan mode (read-only); -y auto-approves actions for work mode.
+ARGS=(-p "$(cat "$PROMPT_FILE")" --output-format text)
+[[ -n "$MODEL" && "$MODEL" != "-" ]] && ARGS+=(-m "$MODEL")
+if [[ "$MODE" == "advise" ]]; then
+  ARGS+=(--plan)
+else
+  ARGS+=(-y)
+fi
+
+set +e
+(
+  cd "$WORKDIR" || exit 127
+  # Subscription-login path: strip API-key env so the CLI uses its own auth.
+  run_with_timeout "$RUN_TIMEOUT" env \
+    -u MOONSHOT_API_KEY -u KIMI_API_KEY \
+    OMNILANE_DEPTH=1 \
+    "$KIMI_BIN" "${ARGS[@]}" \
+    > "${OUTPUT_FILE}.tmp" 2> "${OUTPUT_FILE}.stderr.log"
+)
+RC=$?
+set -e
+
+# Empty output is a failure, not a silent rc=0 success.
+if ! grep -q '[^[:space:]]' "${OUTPUT_FILE}.tmp" 2>/dev/null; then
+  echo "omnilane: kimi produced no output" >> "${OUTPUT_FILE}.stderr.log"
+  [[ "$RC" -eq 0 ]] && RC=1
+fi
+
+[[ -f "${OUTPUT_FILE}.tmp" ]] && mv "${OUTPUT_FILE}.tmp" "$OUTPUT_FILE"
+[[ -s "${OUTPUT_FILE}.stderr.log" ]] || rm "${OUTPUT_FILE}.stderr.log" 2>/dev/null || true
+exit "$RC"

package/scripts/runners/run-qwen.sh

@@ -0,0 +1,48 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+# omnilane runner: Alibaba Qwen Code CLI
+#
+# Usage: run-qwen.sh MODE WORKDIR MODEL EFFORT PROMPT_FILE OUTPUT_FILE
+#   MODE = advise (default approvals: write tools blocked non-interactively)
+#        | work  (--approval-mode yolo: may edit files in WORKDIR)
+#
+# Qwen Code has no reasoning-effort knob, so EFFORT is parity-only.
+
+source "$(dirname "${BASH_SOURCE[0]}")/../lib/common.sh"
+
+MODE="$1"; WORKDIR="$2"; MODEL="$3"; EFFORT="$4"; PROMPT_FILE="$5"; OUTPUT_FILE="$6"
+: "$EFFORT" # parity with the uniform runner interface
+
+QWEN_BIN="${QWEN_BIN:-qwen}"
+RUN_TIMEOUT="${OMNILANE_TIMEOUT:-600}"
+
+truncate_payload "$PROMPT_FILE" 140000
+
+ARGS=(-p "$(cat "$PROMPT_FILE")" --output-format text)
+[[ -n "$MODEL" && "$MODEL" != "-" ]] && ARGS+=(-m "$MODEL")
+[[ "$MODE" == "work" ]] && ARGS+=(--approval-mode yolo)
+
+set +e
+(
+  cd "$WORKDIR" || exit 127
+  # Subscription-login path: strip OpenAI-compatible env overrides so the CLI
+  # uses its own Qwen OAuth instead of a hijacked endpoint.
+  run_with_timeout "$RUN_TIMEOUT" env \
+    -u OPENAI_API_KEY -u OPENAI_BASE_URL -u DASHSCOPE_API_KEY \
+    OMNILANE_DEPTH=1 \
+    "$QWEN_BIN" "${ARGS[@]}" \
+    > "${OUTPUT_FILE}.tmp" 2> "${OUTPUT_FILE}.stderr.log"
+)
+RC=$?
+set -e
+
+# Empty output is a failure, not a silent rc=0 success.
+if ! grep -q '[^[:space:]]' "${OUTPUT_FILE}.tmp" 2>/dev/null; then
+  echo "omnilane: qwen produced no output" >> "${OUTPUT_FILE}.stderr.log"
+  [[ "$RC" -eq 0 ]] && RC=1
+fi
+
+[[ -f "${OUTPUT_FILE}.tmp" ]] && mv "${OUTPUT_FILE}.tmp" "$OUTPUT_FILE"
+[[ -s "${OUTPUT_FILE}.stderr.log" ]] || rm "${OUTPUT_FILE}.stderr.log" 2>/dev/null || true
+exit "$RC"