plugins/cli
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Compare commits

merge into: plugins/cli:feat/skill-behaviors-and-version-block/main
Branches Tags
plugins/cli:master plugins/cli:develop plugins/cli:chore/marketplace-registry-sync/main plugins/cli:feat/cli-hook-rewire/main plugins/cli:feat/cli-hook-rewire/quote-description plugins/cli:feat/skill-behaviors-and-version-block/main plugins/cli:feat/skill-behaviors-and-version-block/deploy-requires-warn plugins/cli:feat/doctor-cli-runtime-tests/main plugins/cli:chore/preserve-local-develop-20260831 plugins/cli:feat/deploy-recommend-update
..
pull from: plugins/cli:develop
Branches Tags
plugins/cli:master plugins/cli:develop plugins/cli:chore/marketplace-registry-sync/main plugins/cli:feat/cli-hook-rewire/main plugins/cli:feat/cli-hook-rewire/quote-description plugins/cli:feat/skill-behaviors-and-version-block/main plugins/cli:feat/skill-behaviors-and-version-block/deploy-requires-warn plugins/cli:feat/doctor-cli-runtime-tests/main plugins/cli:chore/preserve-local-develop-20260831 plugins/cli:feat/deploy-recommend-update
29 Commits
feat/skill-behaviors-and-version-block/main .. develop
Author SHA1 Message Date
admin 62ea8022e0 Merge pull request '本機複本一個 domain 一把鎖，重啟閘門不等整輪判定' (#65) from fix/serialize-shared-clone-and-always-gate into develop 
Reviewed-on: #65
 2026-09-07 04:43:42 +00:00
jiantw83andClaude Opus 5 cd832460ee fix(deploy): 本機複本一個 domain 一把鎖，重啟閘門不等整輪判定 
實測踩到兩件事，同一輪、同一個根因。

那一份本機複本一台機器只有一份，而技能規定五支 CLI 平行部署——平行是對的，
它們寫的是不同的外掛目錄。但複本不是：五支都會來 pull 同一個目錄，連只需要
讀 manifest 的那幾支也會（相依檢查從那裡讀）。git 對同一個存取庫的併發寫入
沒有保護，於是同一輪裡兩支撞在一起，一支拿不到 ORIG_HEAD.lock、一支的遠端
refs 換不上去。

後果是最難查的那一種：兩支的整輪判定都變成 fail，而外掛其實全部裝好了——
報告說失敗、實際成功，而真正的原因跟部署無關。

改成一個 domain 一把 mkdir 鎖：那是檔案系統這一層唯一原子的建立動作。等不到
就印一行 warn 改用磁碟上的內容，別人正在拉同一份，硬等下去只是排隊。上一輪
中途死掉留下的鎖用年紀判，門檻放寬到等待秒數的四倍。

複本已經在磁碟上而 pull 拉不動的那一種，也改成只印 warn、不判整輪失敗：
內容在，只是可能比遠端舊。但一定要印出來——安靜地裝一份舊內容，是這一組
工具最怕的那種失效。clone 不存在那一種照舊算失敗，磁碟上根本沒東西可裝。

第二件事更嚴重。原本的寫法是「整輪判定成功才掛重啟閘門」，於是那一輪的
fail 把閘門一起跳過了：外掛換了一半，而唯一沒有被告知要重啟的，剛好就是
正在跑那份剛被換掉的程式碼的那一支 CLI。一道只在成功時才生效的提醒，在最
需要它的那一次不會出現。改成 install 與 update 一律先掛，再判 result。

順帶補一支安全截斷：訊息截長度用的是 cut -c，那數的是位元組，多位元組字
剛好被切成兩半會留一個替代字元，而亂碼不影響結束碼、沒有人會來報。

乾跑那一路一步都不動，連鎖都不取——取鎖是建目錄，那已經是寫入。原本改完
之後乾跑會真的去 pull，這一版修回來了。

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
 2026-09-07 12:35:27 +08:00
admin 07c1c44e5a Merge pull request 'doctor 與 setup 補上路徑守則，腳本呼叫一律改成字面絕對路徑' (#63) from fix/literal-paths-for-doctor-and-setup into develop 
Reviewed-on: #63
 2026-09-03 05:35:25 +00:00
jiantw83 56dbffa25c chore(plugin): 三份 manifest 升版至 0.3.4 
兩支技能的路徑守則要靠版號才傳得到機器端。

三份 manifest 由 sync-skill-manifest.sh 同步，只動版本欄位。
 2026-09-03 13:07:07 +08:00
jiantw83 5b0eb784b4 fix(doctor,setup): 兩支技能補上路徑守則，呼叫一律字面絕對路徑 
這兩支技能的腳本呼叫原本全是裸相對路徑，整份文件沒有任何路徑守則。相對路徑會對著操作者當下的工作目錄解，而那裡從來不是外掛根目錄，所以每一個呼叫點都是模型就地猜前綴的地方。部署技能原本三處相對路徑被補成帶變數路徑，就是同一個機制。

實際掃到的處數比預估多：體檢 18 處、修復 20 處。多出來的是兩類原本沒想到的——裸檔名的 Gitea 呼叫，以及被當成參數傳的資料檔。後者同樣會解錯地方，而且解錯了不會報錯。

兩支各補路徑守則與前置步驟，解出兩條根目錄：跨外掛走 current 那一層（解到那一層就停，再往下解會落到帶版本號的快取路徑，那種路徑進不了允許清單），技能自己的腳本與範本走 CLI 載入技能時講明的外掛基底目錄。

兩支都不靠 current 底下那條 jsc-cli 連結，但理由各自不同，不是照抄部署那一支的。體檢不能靠，因為它正是機器可疑時才跑的技能——觸發時機本身就是連結可能出錯的那幾個時刻，而連結指著舊版時拿到的是舊的掃描腳本與舊的規格表，掃出來的每一列都像真的發現，不會有任何錯誤訊息。修復更不能靠，因為它寫檔，而且它要修的其中一列正是那個決定連結位置的變數：那種機器上根目錄根本解不出來，而修它要用的腳本掛在外掛基底目錄底下、不依賴那個變數，所以解不出來既不停這一輪也不擋那一項修復。何況拿舊的寫入腳本改設定檔，再用同一份舊腳本重驗，兩邊當然對得上，整輪會報成已修。

失敗分支也與部署不同。部署解不出根目錄就停手；這兩支都不停——體檢把它記成一項發現然後跑完不需要跨外掛的檢查，因為唯讀體檢中途停掉操作者什麼都拿不到；修復把它當成待修的那一項，修好再重解一次。

兩份範本與說明文件一併改。範本是這兩支技能在同一輪一起讀的，而且指示寫入，留著裸路徑等於留一個繞過新規則的入口。範本裡另外補掉兩個路徑洞：指向範本自己的佔位符原本沒有路徑，以及一個連目錄都沒有的裸檔名。

行為契約四列跟著改，並補上可稽核的跡象：回報裡的腳本路徑全是字面絕對路徑，跨外掛的路徑中間是 current 那一層而不是帶版本號的快取路徑。
 2026-09-03 13:07:07 +08:00
admin 45187f5173 Merge pull request '部署收尾刷新 current 連結農場，讓升版後的技能路徑不再指著舊版' (#61) from feat/refresh-current-link-farm-on-deploy into develop 
Reviewed-on: #61
 2026-09-03 04:40:24 +00:00
jiantw83 47f248adec chore(plugin): 三份 manifest 升版至 0.3.3 
連結農場的刷新要靠版號才傳得到機器端。

三份 manifest 由 sync-skill-manifest.sh 同步，只動版本欄位。
 2026-09-03 12:19:59 +08:00
jiantw83 a439636c4a feat(deploy): 部署收尾刷新 current 連結農場 
$JSC_HOME/current 是一組不帶版本的符號連結，每個外掛一條，指向快取裡帶版本號的實體目錄。技能文件裡所有跨外掛的腳本呼叫都以這一層為根，因為它不帶版本號、寫得進權限允許清單。

問題是沒有任何東西會更新這些連結，只有 wire-cli.sh 會更新 jsc-hooks 那一條。其餘幾條是人手動建的，建好之後就停在當時的版本。實際後果是部署完四個 domain 之後，快取裡是新版，連結卻還指著舊版：助理巡檢照文件的字面路徑跑，跑到的是舊腳本，而其中一個舊版底下根本沒有它要呼叫的檔案。失敗無聲，只有心跳停止，沒人盯就不會有人發現。

部署改成收尾時刷新整組連結。挑部署來做，是因為它本來就知道裝了哪些 domain、裝到哪個版本，資訊最齊。

四個設計決定：

基準 CLI 取 claude、codex、copilot、kiro 之中第一支找得到的，整輪只有那一支寫連結。連結農場只有一組，不可能同時指向五個 CLI 的副本；而五支 CLI 是平行跑的，五支都寫會互相覆寫，最後指到哪一份是隨機的、出事重現不出來。antigravity 一律不當基準，它的來源是本地 clone，而那份 clone 明文允許是維護者的開發樹，把全機器路徑指到做到一半的樹正好是這次要修的那種毛病。

版本目錄取版本排序最大、且真的有 plugin.json、且本身不是符號連結的那一層。要求 plugin.json 是因為裝到一半的目錄沒有它，挑到會讓連結指向不完整的外掛而且照樣不報錯。

解除安裝的判準是「連結還在、指向卻沒了」，不是「這輪解除安裝過這個 domain」。只解除安裝其中一支 CLI 時，連結可能還指著另一支手上完好的副本，那一條必須留著。

連結建立失敗印一行繼續，不記進失敗清單。這一段跑在外掛都裝好之後，部署本身已經成功；記成失敗會連帶跳過重啟閘門，操作者拿到的是一台明明裝好卻被說成失敗的機器。缺陷仍然看得見，因為輸出多了一行。

目標存在但不是符號連結時一律不覆寫，印 skip 要人工處理。ln -sfn 對著實體目錄下手會把連結建進那個目錄裡，農場當場壞掉還不會報錯。

新增 link 行讓呼叫端讀得到每一條連結指到哪裡，狀態五選一。技能文件與行為契約跟著更新，另修正一句因這次改動而失效的敘述：原本寫 current 底下沒有 jsc-cli，刷新之後那條連結會存在，改成講清楚它仍然靠不住，因為第一次建起它的正是這一輪。
 2026-09-03 12:19:59 +08:00
admin 253443e94e Merge pull request 'deploy 的腳本呼叫全改成字面絕對路徑，開頭解一次根目錄，無人看管的輪次不再停在第一支腳本' (#59) from fix/literal-absolute-paths-for-deploy into develop 
Reviewed-on: #59
 2026-09-03 02:52:09 +00:00
jiantw83 bd23690aa5 chore(plugin): 三份 manifest 升版至 0.3.2 
三份外掛 manifest 的版本欄從 0.3.1 升到 0.3.2，內容由同步腳本產生，其餘欄位未動。

deploy 技能的路徑守則與 Step 0 已經改完，版號要跟著動，版本守衛才判定得出這台機器落後，更新輪次才會把新的技能內容拉下來。

交給 manifest 同步腳本一次寫三份，避免三份手改而版號不一致。

影響每一台裝了這組技能的機器：版本比對與部署更新都讀這個欄位。
 2026-09-03 10:33:30 +08:00
jiantw83 b41ceb00d9 fix(deploy): 腳本呼叫改成字面絕對路徑，開頭先解一次根目錄 
技能文件新增路徑守則一節，寫明這支技能的每一次腳本呼叫都要是字面絕對路徑，不留未展開的變數，也不留波浪號。

新增 Step 0：開頭跑一次 readlink -f "$JSC_HOME/current" 解出根目錄，整輪只解這一次，之後每一處都把代稱換成那個目錄。解出來的路徑要用 [ -d ] 查過存在才算數——JSC_HOME 沒設時那道指令會印出 /current、結束碼 0，非空又是絕對路徑，只看前三項會直接放行，而那個目錄並不存在。文件同時明講解到 current 這一層就停，不要再往下解一層，因為再解就落到帶版本號的快取路徑，而那種路徑放不進允許清單。

四處跨外掛呼叫改成字面路徑寫法。其中三處原本寫成相對路徑，那才是模型自己補成帶變數路徑的源頭。技能自己的四支腳本另外處理：jsc-cli 不在 current 底下，那裡只有另外三個外掛，所以改用 CLI 載入技能時講明的外掛基底目錄，原樣照抄、不跑指令。

無人看管的輪次在第一支腳本就被權限層擋下，整輪一個外掛都沒部署。2026-09-02 到 09-03 用非互動模式比照排程環境實測多種寫法：權限層拿未展開的字面字串做靜態比對，帶變數或波浪號的路徑對不上任何允許規則，只有允許清單上的完整字面指令跑得動，連 readlink、ls 這種讀取指令都要各自有規則。放寬允許清單解不掉，因為規則本身也是靜態比對，而且路徑中段的萬用字元不匹配。

deploy 是人在現場跑的技能，開頭那一次核准詢問有人可以按，所以保留自己解路徑的做法。守則另外寫下一句擋回頭路：為了看起來整齊而把字面路徑改回帶變數的寫法，等於再一次弄壞每一個無人看管的輪次。

行為說明表的 deploy 四列同步改寫，觸發時機沒動。

影響排程觸發的無人看管輪次，它們現在才跑得完第一支腳本。現場執行部署的操作者只多出開頭那一次核准詢問。
 2026-09-03 10:33:30 +08:00
admin 9d0e1ea3dd Merge pull request '體檢目錄頁改成大標題加條列，doctor 與 setup 的呼叫同步換鍵' (#58) from feat/contents-list/main into develop 
Reviewed-on: #58
 2026-09-02 10:00:37 +00:00
jiantw83 4479a090a6 chore(plugin 版本): 三份 manifest 升版至 0.3.1 
What：`plugin.json`、`.claude-plugin/plugin.json`、`.codex-plugin/plugin.json` 三份 manifest 的 `version` 從 `0.3.0` 升到 `0.3.1`，三份一起改、值保持一致，其餘欄位一個字都不動。

Why：版本號是 `jsc-hooks/hooks/version-guard.sh` 判斷機器上的 plugin 落後與否的唯一依據。目錄頁的版面與鍵已經換過，版本號不動的話 version-guard 會把機器上的舊版當成最新版，`jsc-cli:doctor` 不會把它列進待修項目，那台機器就繼續留著以裸 `HASH` 當鍵的舊敘述，寫出來的目錄頁跟新範本對不上。三份分別給不同 CLI 讀，只升其中一份會讓同一支 plugin 在不同 CLI 上報出不同版本，落後判斷跟著失準。

How：只改 `version` 這一個鍵，走修訂號一階。這一輪是既有頁面型別的呈現方式調整，沒有新增技能，也沒有改動任何腳本的呼叫介面，不到次版本號的幅度。三份的值刻意保持一致，version-guard 才比得出單一結論。

Who：屬於「wiki 目錄頁改條列式呈現」這個需求的發版收尾，與同一輪的敘述與範本改動配成一套。獨立成一個 commit 的理由是型別不同：這一組是 chore 而非 feat，改動的檔案與敘述那一組完全不重疊，版號要重切或回退時也不必動到行為契約。
 2026-09-02 17:25:37 +08:00
jiantw83 d5bc40be13 feat(wiki): 體檢目錄頁改條列式版面，鍵改用體檢頁頁名 
What：`templates/check-contents.md` 從 markdown 表格改成一台執行環境一個 H2 區塊，H2 標題就是那一台的體檢頁頁名 `CHECK_{HASH}`，原本的七個欄位改成標題底下一層 `- {欄位名}：{值}` 的條列，頁上不留任何表格。`skills/doctor/SKILL.md` 與 `skills/setup/SKILL.md` 對目錄頁的呼叫從 `wiki-contents.sh upsert CHECK 4 "{HASH}"` 改成 `upsert CHECK 1 "CHECK_{HASH}"`，`references/behaviors.md` 的關鍵步驟、完成條件與可驗證跡象，以及 `README.md` 的兩段技能敘述都跟著對齊。

Why：一列七格的表格，欄位一多就要橫向捲動，讀的人得先數欄位再對照表頭才知道哪一格是什麼；一筆一個 H2 區塊、每一條自己帶欄位名，掃過去就讀得懂。版面換成區塊之後鍵也得跟著換：表格時代的鍵是第 4 欄的裸 `HASH`，區塊沒有欄位可指，唯一能當鍵的是 H2 標題。key-col 與 key 沿用舊值會讓腳本比對不中，同一台機器每體檢一次就在頁尾多附一個區塊，舊區塊從此再也更新不到，而頁面看起來完全正常，錯得無聲無息。頁名只由 `{短主機名}/{登入帳號}` 決定，`GITEA_HOST` 換掉、`JSC_WIKI_REPO_CHECK` 搬到別的存取庫、Gitea 對頁名的編碼有差都動不到它，拿它當鍵比拿任何含網址的值都穩。

How：範本頁首的寫入語意從「一列」改寫成「一個區塊」，並補上四個參數的說明。第三個參數 `<key>` 收 H2 標題文字，也就是 `CHECK_{HASH}`；第四個參數收的是區塊檔而不是列檔，內容為 `## CHECK_{HASH}` 那一行、一個空行，再照範本的欄位順序每欄一條條列，`HASH` 那一欄照樣要寫，標題是鍵不代表欄位可以省，否則下一個讀的人讀不到。第二個參數 `1` 定位成 `<key-col>`，只在頁面還留著舊表格、需要自動轉檔時才用得到，指舊表格裡持有 `[CHECK_{HASH}](網址)` 的第 1 欄，轉檔時取那一格的文字當 H2 標題，頁面已經是條列格式就忽略它。「體檢頁」那一條維持 `[{文字}]({絕對網址})` 的人用連結寫法，H2 標題本身不放連結也不放網址。兩支技能的結束碼分流一併校正：`wiki-contents.sh` 的結束碼 1 從「寫入失敗」改寫成「組不出頁面內容或寫入失敗」，頁上找不到本機那一個區塊不算這一碼，腳本會改成附加。實際的轉檔與 upsert 邏輯都在 `jsc-gitea/tools/wiki-contents.sh`，這個存取庫只改敘述與範本。

Who：屬於「wiki 目錄頁改條列式呈現」這個需求落在 jsc-cli 的部分，也就是 `CHECK_CONTENTS` 這一型目錄頁的去表格化。內容頁維持圖表優先，不在這一輪範圍。
 2026-09-02 17:25:37 +08:00
admin 8444307534 Merge pull request '收尾寫一筆 skill-end 事件，執行狀態才回報得到助理' (#57) from feat/status-report into develop 
Reviewed-on: #57
 2026-09-02 08:04:20 +00:00
jiantw83 db2a2f8206 chore(plugin 版本): 三份 manifest 升版至 0.3.0 2026-09-02 16:01:14 +08:00
jiantw83 03e59c69bd feat(狀態回報): 收尾寫一筆 skill-end 事件 
現行紀錄只記「被叫用」，沒有成敗也沒有結束碼。跑完整輪的技能與開場就
中止的技能，在紀錄裡長得一模一樣。

start 由技能用量 hook 順手發，不必改技能文件。end 只能由技能自己在收尾
步驟寫——hook 接在技能工具呼叫上，而實際工作發生在之後的模型輪次，它在
原理上看不到成敗。有 start 沒有配對的 end，就是那一輪中止了。

status 五選一，每支技能各自寫明什麼情況選哪一個。找不到回報腳本就安靜
跳過，回報失敗一律不改變技能自己的結論。
 2026-09-02 16:01:14 +08:00
admin 936fc5cdaa Merge pull request '連結一律寫成 [文字](絕對網址)，並在寫入前驗證連得到' (#56) from feat/link-verification/main into develop 
Reviewed-on: #56
 2026-09-02 06:46:45 +00:00
jiantw83 9f4ed977b2 chore(plugin 版本): 三份 manifest 升版至 0.2.9 2026-09-02 14:27:18 +08:00
jiantw83 c76daa61ca feat(link): 連結一律寫成 [文字](絕對網址)，寫入前先驗證連得到 
取消 [[頁名]] 與 [[顯示文字|頁名]] 兩種同 wiki 寫法，不再分「同存取庫」與
「跨存取庫」兩條規則。那種寫法只在自己那個 wiki 內解析，寫錯不報錯，畫面上
看起來像普通文字或死連結，巡不到也修不了。

連結寫進頁面前先過 jsc-gitea 的 link-check.sh，結束碼 0 才寫。驗證一律走 API，
不看網頁狀態碼：私有存取庫的網頁網址對未登入請求一律回 404，拿狀態碼判會把
好連結判成壞的。認證失敗回 7，與死連結的 1 分開，免得金鑰一過期就把還在的頁
整批判死。
 2026-09-02 14:27:18 +08:00
admin 86225f355f Merge pull request 'feat(wiki): 體檢目錄頁改走專用存取庫，並補齊設定規格表' (#54) from feat/wiki-contents-repo/main into develop 
Reviewed-on: #54
 2026-09-02 03:27:54 +00:00
jiantw83 fb80559159 feat(wiki): 體檢目錄頁改走專用存取庫，並補齊設定規格表 
What：CHECK_CONTENTS 改由 wiki-repo CONTENTS 解析並透過 wiki-contents.sh upsert
寫入，CHECK_{HASH} 仍走 wiki-repo CHECK。目錄頁新增一欄裸 HASH 當比對鍵。主機名
改由程式取短名，不再交給模型自由填。

Why：比對鍵原本是含網址的儲存格，換主機或換存取庫就比對不到，每跑一次體檢就替同一台
機器多附一列，畫面上還看不出來。主機名短名與 FQDN 不一致時，同一台機器會分裂成兩張頁，
而助理巡檢那邊是用程式取值的，兩邊對不起來。

How：設定規格表同一輪補齊三處既有缺漏——補上漏掉的 JSC_WIKI_REPO_MONITOR，體檢本來
看不到它而孤兒掃描還會誤報；刪掉指向不存在頁面的 MAINTAIN 內容頁字樣；MAINTAIN 那一列
改成不需要使用者處理，免得體檢叫人去設一支管不到任何頁的變數。整列保留，刪掉會讓孤兒
掃描開始誤報那個變數。

Who：jsc-cli
 2026-09-02 11:03:07 +08:00
admin 225e33d2c8 Merge pull request '放行 jsc-assist 的 marketplace 條目到預設分支' (#52) from chore/marketplace-assist-registry/main into develop 
Reviewed-on: #52
 2026-09-01 04:55:09 +00:00
admin 214b8a22c5 Merge pull request 'chore/marketplace-assist-registry/sync-copies' (#51) from chore/marketplace-assist-registry/sync-copies into chore/marketplace-assist-registry/main 
Reviewed-on: #51
 2026-09-01 04:52:42 +00:00
jiantw83 e87d6b36fe chore(marketplace): 把 jsc-assist 登錄進統一 marketplace 
What：
- 兩份 marketplace 檔各加一個 jsc-assist 條目，來源網址指向 assist 存放庫。

Why：
- 準則要求每個 domain 存放庫都帶同一份 marketplace 檔，任何一個存放庫都能當註冊入口。副本之間只要有一份沒跟上，稽核就會報出不一致。
- 正本少了這個條目，各 CLI 的安裝指令就找不到 jsc-assist，這個 domain 等於發佈不出去。

How：
- 條目由 meta 的 sync-marketplace.sh 產生，同時寫進正本與每個 domain 存放庫的副本，寫完逐檔比對位元組。這一支存放庫的兩份副本就是那一輪的產物。
- 條目依名稱排序，縮排與非 ASCII 描述的處理都交給同一支腳本，不手改 JSON。
- 這一批是從最新的預設分支重新產生的。前一輪的分支基底早於監控頁型別那批改動，直接合併會把那些改動回退掉，所以整批重做而不是解衝突。

Who：
助理 domain 落地的註冊步驟在這個存放庫的同步。
 2026-09-01 12:50:04 +08:00
jiantw83 36e1bebe1f Merge pull request '收攏 models 技能的 frontmatter 語法修正' (#48) from feat/cli-hook-rewire/main into develop 2026-09-01 00:58:39 +00:00
jiantw83 749b1ad6d3 Merge pull request '修正 models 技能 SKILL.md frontmatter 的 YAML 純量語法' (#47) from feat/cli-hook-rewire/quote-description into feat/cli-hook-rewire/main 2026-09-01 00:56:08 +00:00
jiantw83 daa4bcf9e1 fix(frontmatter): 修正 models 技能 SKILL.md frontmatter 的 YAML 純量語法錯誤 
What：
- 修正 skills/models/SKILL.md frontmatter 裡 description 欄位的 YAML 語法錯誤。
- 整串 description 加上單引號，內部撇號改寫成兩個單引號，內容文字一個字都沒變。
- 同步更新 plugin.json、.claude-plugin/plugin.json、.codex-plugin/plugin.json 三個 manifest 版本號，從 0.2.6 進到 0.2.7。

Why：
- description 內含「冒號加空白」，屬於未加引號的 YAML plain scalar，違反 YAML 語法規定。
- Antigravity 解析 frontmatter 時當場中斷，整支技能被靜默丟棄，沒有任何錯誤訊息；磁碟上 34 支技能，Antigravity 只認得 28 支。
- 準則要求 description 用英文撰寫，不能把「: 」改成全形冒號迴避語法問題，只能加引號修正。

How：
- 整串 description 值加上單引號，內部撇號寫成兩個單引號跳脫，其餘字元不動。
- 用 git show HEAD: 取出改前的原始值，把改後的單引號純量還原後做字串相等比對，確認逐字相同、字元數一致。
- 執行 ste100-lint.sh、check-behaviors.sh、lint-frontmatter.sh 三支檢查腳本，退出碼皆為 0；git diff --numstat 顯示只動了 frontmatter 那一行。

Who：
- 本次修到 cli 技能組的 models 技能，屬盤點各 CLI 可用模型與能力標籤的功能。
 2026-08-31 19:02:43 +08:00
jiantw83 7d58afa2ee Merge pull request '收攏技能行為清單與相依版本阻擋改動' (#45) from feat/skill-behaviors-and-version-block/main into develop 2026-08-31 08:10:33 +00:00
17 changed files with 688 additions and 100 deletions
Expand all files Collapse all files
.agents/plugins/marketplace.json
+8
View File
@@ -13,6 +13,14 @@
},
"description": "決策樹問詢與問詢紀錄（QUESTION_* wiki 頁）"
},
{
"name": "jsc-assist",
"source": {
"source": "url",
"url": "https://gitea.jsc.idv.tw/plugins/assist.git"
},
"description": "助理：事件收攏、健康巡檢與待辦簿（MONITOR_* wiki 頁）"
},
{
"name": "jsc-cli",
"source": {
.claude-plugin/marketplace.json
+8
View File
@@ -13,6 +13,14 @@
},
"description": "決策樹問詢與問詢紀錄（QUESTION_* wiki 頁）"
},
{
"name": "jsc-assist",
"source": {
"source": "url",
"url": "https://gitea.jsc.idv.tw/plugins/assist.git"
},
"description": "助理：事件收攏、健康巡檢與待辦簿（MONITOR_* wiki 頁）"
},
{
"name": "jsc-cli",
"source": {
.claude-plugin/plugin.json
+1 -1
View File
@@ -1,6 +1,6 @@
{
"name": "jsc-cli",
"version": "0.2.6",
"version": "0.3.5",
"description": "CLI 偵測、模型能力標籤、子代理派工與技能庫批次部署",
"skills": "./skills",
"author": {
.codex-plugin/plugin.json
+1 -1
View File
@@ -1,6 +1,6 @@
{
"name": "jsc-cli",
"version": "0.2.6",
"version": "0.3.5",
"description": "CLI 偵測、模型能力標籤、子代理派工與技能庫批次部署",
"skills": "./skills",
"jsc": {
README.md
+6 -5
View File
@@ -23,7 +23,7 @@ Marketplace 統一為 `jsc`（https://gitea.jsc.idv.tw/plugins/meta.git），安
| 工具 | 用途 |
| --- | --- |
| `tools/detect-clis.sh` | 列出已安裝的 AI CLI 與執行檔路徑（TSV：name / path / version；antigravity 的執行檔為 `agy`、kiro 為 `kiro-cli`） |
| `tools/deploy.sh` | 對單一 CLI 執行安裝、更新或解除安裝（`deploy.sh [-n] {mode} {cli} {domain}...`，mode 為 install / update / uninstall）；印出每個指令與其結束碼，最後一行 `result` 標 ok 或 fail。`-n` 只印指令不執行。上表五個 CLI 的指令差異全部收在這支腳本裡。Codex 更新 `jsc-cli` 與 `jsc-hooks` 後會把舊版快取路徑補成指向新版的相容連結，避免正在跑的部署流程找不到 helper 腳本，也避免尚未重啟的工作階段在 Stop hook 階段找不到舊路徑。install 或 update 全數成功時，收尾轉呼叫 `jsc-hooks` 的 `restart-gate.sh require` 掛上重啟閘門，並印一行 `restart` 標出狀態檔位置；uninstall 不寫。尋找 `restart-gate.sh` 時優先用 `$JSC_HOME/current/jsc-hooks`、本地 clone 與 Kiro skills，最後才掃各 CLI 快取，避免部署收尾綁死單一 CLI 的版號路徑。狀態檔的路徑、格式與判讀全在 `restart-gate.sh`，這支腳本不自己拼——格式只留一個真實來源。站台取自 `GITEA_HOST`，本地 clone 目錄取自 `JSC_LOCAL_PLUGINS`，兩者的預設值見下表 |
| `tools/deploy.sh` | 對單一 CLI 執行安裝、更新或解除安裝（`deploy.sh [-n] {mode} {cli} {domain}...`，mode 為 install / update / uninstall）；印出每個指令與其結束碼，最後一行 `result` 標 ok 或 fail。`-n` 只印指令不執行。上表五個 CLI 的指令差異全部收在這支腳本裡。Codex 更新 `jsc-cli` 與 `jsc-hooks` 後會把舊版快取路徑補成指向新版的相容連結，避免正在跑的部署流程找不到 helper 腳本，也避免尚未重啟的工作階段在 Stop hook 階段找不到舊路徑。收尾還會刷新 `$JSC_HOME/current/` 那一組不帶版本號的符號連結（技能文件的跨外掛路徑都以那一層當根）：install 與 update 把每一條指到這次裝的版本目錄，uninstall 清掉指向已消失的那幾條，一條印一行 `link`，第五欄就是指向。一台機器只有一組農場，所以只有基準 CLI 那一輪會動它，基準取 claude、codex、copilot、kiro 之中第一支裝得到的；連結刷新失敗只記一行、不讓部署變成失敗。install 或 update 全數成功時，收尾轉呼叫 `jsc-hooks` 的 `restart-gate.sh require` 掛上重啟閘門，並印一行 `restart` 標出狀態檔位置；uninstall 不寫。尋找 `restart-gate.sh` 時優先用 `$JSC_HOME/current/jsc-hooks`、本地 clone 與 Kiro skills，最後才掃各 CLI 快取，避免部署收尾綁死單一 CLI 的版號路徑。狀態檔的路徑、格式與判讀全在 `restart-gate.sh`，這支腳本不自己拼——格式只留一個真實來源。站台取自 `GITEA_HOST`，本地 clone 目錄取自 `JSC_LOCAL_PLUGINS`，兩者的預設值見下表 |
| `tools/check-requires.sh` | `check-requires.sh {cli} {manifest}` 檢查 manifest 的 `jsc.requires` 最低版本。沒有宣告就通過；版本不符或缺相依 plugin 就回 `status=blocked` 與結束碼 1。`deploy.sh update` 在每個 domain 更新前呼叫它一次：結束碼 1 只印一行 `warn`，那個 domain 照樣更新——跳過會讓落後的 domain 永遠等不到相依版本，也就永遠更新不到，真正的阻擋由 `jsc-hooks` 的 `version-guard.sh` 在技能被叫用時執行；結束碼 2 以上是檢查腳本自己出錯，讀不到結論就不當成通過，印 `skip` 並跳過該 domain |
| `tools/write-guides.sh` | 產生這台機器專屬的更新指引 `$JSC_HOME/update-guide.md` 與移除指引 `$JSC_HOME/remove-guide.md`（`write-guides.sh [-n] {install\|update} {domain}...`），一輪部署跑一次。CLI 清單取自 `detect-clis.sh`，每支 CLI 的指令字面直接取自 `deploy.sh -n` 的輸出，所以指引寫的就是實際會跑的指令；kiro 走不走本地複製退路也依實際偵測結果標注 |
| `tools/list-models.sh` | 讀各 CLI 設定檔列出模型（TSV：cli / model / in-use）；設定檔缺失就不輸出該 CLI 的列，一律 exit 0。設定檔位置只寫在這支腳本裡 |
@@ -32,7 +32,7 @@ Marketplace 統一為 `jsc`（https://gitea.jsc.idv.tw/plugins/meta.git），安
| `tools/scan-config.sh` | 依規格表盤點設定現況（`scan {global\|project\|all}` 印 TSV 與 summary、`spec` 印規格表、`orphans` 找出漏登錄的變數）；`-o` 為離線模式，需要連 Gitea 的檢查一律標 skipped。唯讀，不寫任何設定；帶 TOKEN 的項目只印 set 或 unset |
| `tools/apply-config.sh` | 把設定寫進 shell rc 檔（`set {KEY} {VALUE}`、`unset {KEY}`）或建立目錄（`mkdir {PATH}`）；`show` 印出目前設定，`rcfiles` 印出會寫入的檔案。內容一律收在 `# jsc-config` 標記段落之間，整段重寫不疊加，段落外不動。動檔案前先備份到 `$JSC_HOME/backup/config/{yyyyMMdd_HHmmss}/`，備份失敗就不寫；寫完重讀驗證。fish 自動改用 `set -gx` 語法 |
| `tools/model-tags.sh` | 解析 `references/model-tags.md` 的能力標籤與 SDLC 階段必要標籤（`dump`、`sync`、`stage {階段}`、`model {模型 id}`、`gate {階段} {模型 id}`）；`sync` 寫出 `$JSC_HOME/model-tags.tsv` 供 `jsc-hooks` 的 sdlc-gate 讀取。`gate` 的結束碼 0 為 PASS、1 為缺標籤、2 為模型或階段不在表上，`/jsc-cli:delegate` 依這三碼決定收下或退回 |
| `tools/build-todo.sh` | 把三支檢查腳本的輸出合併成一張「待修項目」表（`--config` 收 `scan-config.sh scan`、`--wiring {cli}=` 收 `wire-cli.sh status`、`--version` 收 `version-guard.sh report`）；類別固定排成 missing、invalid、unwired、落後，一項都沒有時仍印一列「無」。`/jsc-cli:doctor` 與 `/jsc-cli:setup` 共用這一份合併規則，兩邊各寫一次就會各自漂移 |
| `tools/build-todo.sh` | 把三支檢查腳本的輸出合併成一張「待修項目」表（`--config` 收 `scan-config.sh scan`、`--wiring {cli}=` 收 `wire-cli.sh status`、`--version` 收 `version-guard.sh report`）；類別固定排成 missing、invalid、unwired、落後，一項都沒有時仍印一列「無」。`/jsc-cli:doctor` 與 `/jsc-cli:setup` 共用這一份合併規則，兩邊各寫一次就會各自漂移。兩支都以 `{CLI_ROOT}/tools/build-todo.sh` 這種字面絕對路徑呼叫它，`{CLI_ROOT}` 是 CLI 載入技能時講明的 jsc-cli 外掛基底目錄；不留 `$JSC_HOME`、波浪號或裸的相對路徑，帶變數的路徑進不了允許清單，相對路徑則會對著操作者當下的工作目錄解 |
## Skills 目錄
@@ -54,11 +54,11 @@ Marketplace 統一為 `jsc`（https://gitea.jsc.idv.tw/plugins/meta.git），安
### `doctor`
一次體檢執行環境，只讀不改。四項檢查**同時啟動，各一個 sub agent**：技能版本（`jsc-hooks/hooks/version-guard.sh report`）、Hook 接線（`jsc-hooks/tools/wire-cli.sh status`，唯讀子命令）、設定現況（`tools/scan-config.sh scan all` 比對 `tools/config-spec.tsv`，一次掃完全域與專案兩個範圍）、漏登錄變數（`scan-config.sh orphans`）。唯讀契約下放程式層：呼叫 `wire-cli.sh` 一律帶 `JSC_READONLY=1`，打錯子命令也不會改到機器。每項各出一張表，專案那張一定寫出掃的是哪個目錄；待修項目由 `tools/build-todo.sh` 合併三份輸出並排序。整份結果寫進 wiki `CHECK_{HASH}`，`HASH` 取 `{主機名}/{登入帳號}`，只保留最新一次。修復交給 `/jsc-cli:setup`，體檢本身不動任何設定。
一次體檢執行環境，只讀不改。開跑先解出兩條根目錄，整輪的腳本呼叫一律填成字面絕對路徑：`{JSC_ROOT}` 取 `readlink -f "$JSC_HOME/current"`（解到那一層就停，不再往下解成帶版本號的快取路徑），`{CLI_ROOT}` 取 CLI 載入技能時講明的外掛基底目錄。這一支解不出 `{JSC_ROOT}` 不停手，直接把 `JSC_HOME` 記成一項發現：唯讀體檢中途停掉，操作者什麼都拿不到。技能自己的 `tools/` 與 `templates/` 一律走 `{CLI_ROOT}`，不走 `current` 底下那條 `jsc-cli` 連結——體檢正是機器可疑時才跑的，連結指著舊版時拿到的是舊的 scan-config.sh 與舊的 config-spec.tsv，掃出來的每一列都像真的發現，而且不會有任何錯誤訊息。四項檢查**同時啟動，各一個 sub agent**：技能版本（`{JSC_ROOT}/jsc-hooks/hooks/version-guard.sh report`）、Hook 接線（`{JSC_ROOT}/jsc-hooks/tools/wire-cli.sh status`，唯讀子命令）、設定現況（`{CLI_ROOT}/tools/scan-config.sh scan all` 比對 `{CLI_ROOT}/tools/config-spec.tsv`，一次掃完全域與專案兩個範圍）、漏登錄變數（`{CLI_ROOT}/tools/scan-config.sh orphans`）。唯讀契約下放程式層：呼叫 `wire-cli.sh` 一律帶 `JSC_READONLY=1`，打錯子命令也不會改到機器。每項各出一張表，專案那張一定寫出掃的是哪個目錄；待修項目由 `{CLI_ROOT}/tools/build-todo.sh` 合併三份輸出並排序。整份結果寫進 wiki `CHECK_{HASH}`，`HASH` 取 `{短主機名}/{登入帳號}`，兩個值都由程式取，主機名一律切掉網域，只保留最新一次。目錄頁 `CHECK_CONTENTS` 住另一個庫（`JSC_WIKI_REPO_CONTENTS`），版面是 H1 加 `>` 引言，再一台機器一個 H2 區塊，頁上沒有 markdown 表格；改由 `{JSC_ROOT}/jsc-gitea/tools/wiki-contents.sh upsert CHECK 1 CHECK_{HASH}` 只寫本機那一個區塊，欄位是 `- {欄位名}：{值}` 的條列。鍵是 H2 標題，也就是體檢頁頁名 `CHECK_{HASH}`，不是任何含網址的值：網址會隨站台、存取庫與頁名編碼改變，頁名只由主機加帳號決定，拿網址當鍵就比不中，同一台機器每體檢一次就多附一個區塊。第二個參數 `1` 是 `<key-col>`，只在頁面還是舊表格、需要自動轉檔時用得到，指舊表格裡持有 `[CHECK_{HASH}](網址)` 的第 1 欄。「體檢頁」那一條的連結給人點，填絕對網址。修復交給 `/jsc-cli:setup`，體檢本身不動任何設定。
### `setup`
修復 `/jsc-cli:doctor` 找出的問題，一次一項，逐項確認才動手。待修清單優先讀 wiki `CHECK_{HASH}`，沒有頁面就當場重掃：**三支檢查腳本併行跑**，再用 `tools/build-todo.sh` 合併成同一張表。依修法分流：`auto` 用 `tools/apply-config.sh` 直接寫、`ask` 先用決策樹問到值再寫、`manual` 印出步驟交給操作者。複合修復交回原主：版本落後找 `/jsc-cli:deploy`（連同已確認的模式與版本報告一起傳過去，不讓它重問重查）、hook 未接線找 `/jsc-hooks:hooks-install`、缺 `model-tags.tsv` 找 `/jsc-cli:models`。逐項確認維持循序，**寫完的重驗併行**；環境變數類只驗「rc 段落裡確實有那一行」，環境層面交給下一次 `/jsc-cli:doctor`。最後覆寫 CHECK 頁。
修復 `/jsc-cli:doctor` 找出的問題，一次一項，逐項確認才動手。路徑寫法與 doctor 相同：`{JSC_ROOT}` 與 `{CLI_ROOT}` 各解一次，之後每一支腳本都用字面絕對路徑呼叫。這一支寫檔，所以更不能走 `current` 底下那條 `jsc-cli` 連結——`JSC_HOME` 本身就是它要修的一列，那種機器上 `{JSC_ROOT}` 根本解不出來，而修它要用的 apply-config.sh 掛在 `{CLI_ROOT}` 底下，不靠 `JSC_HOME`，所以解不出來既不停這一輪也不擋那一項修復；何況拿舊的 apply-config.sh 寫 rc 檔，再用同一份舊的 `show` 重驗，兩邊當然對得上，整輪會報成已修。待修清單優先讀 wiki `CHECK_{HASH}`，沒有頁面就當場重掃：**三支檢查腳本併行跑**，再用 `{CLI_ROOT}/tools/build-todo.sh` 合併成同一張表。依修法分流：`auto` 用 `{CLI_ROOT}/tools/apply-config.sh` 直接寫、`ask` 先用決策樹問到值再寫、`manual` 印出步驟交給操作者。複合修復交回原主：版本落後找 `/jsc-cli:deploy`（連同已確認的模式與版本報告一起傳過去，不讓它重問重查）、hook 未接線找 `/jsc-hooks:hooks-install`、缺 `model-tags.tsv` 找 `/jsc-cli:models`。逐項確認維持循序，**寫完的重驗併行**；環境變數類只驗「rc 段落裡確實有那一行」，環境層面交給下一次 `/jsc-cli:doctor`。最後覆寫體檢頁 `CHECK_{HASH}`，並用 `{JSC_ROOT}/jsc-gitea/tools/wiki-contents.sh upsert CHECK 1 CHECK_{HASH}` 以 H2 標題（也就是體檢頁頁名）當鍵，更新目錄頁 `CHECK_CONTENTS` 的本機那一個區塊，兩頁分屬不同存取庫。
<!-- JSC-SKILLS:END -->
@@ -71,7 +71,8 @@ Marketplace 統一為 `jsc`（https://gitea.jsc.idv.tw/plugins/meta.git），安
| `JSC_LOCAL_PLUGINS` | antigravity 與 kiro 退路用的本地 clone 目錄 | 用 `$JSC_HOME/plugins`（即 `~/.jsc/plugins`） |
| `JSC_KIRO_SKILLS` | kiro 退路複製 skills 的目標目錄 | 用 `~/.kiro/skills` |
| `JSC_DEPLOY_DRYRUN` | 設為 `1` 等同 `deploy.sh -n`，只印指令不執行 | 照常執行 |
| `JSC_WIKI_REPO_CHECK` | 體檢頁 `CHECK_CONTENTS`、`CHECK_{HASH}` 所在的 `{owner}/{repo}` | 退回 `JSC_WIKI_REPO`；兩個都沒有就略過寫入，並把這一項列進待修 |
| `JSC_WIKI_REPO_CHECK` | 體檢內容頁 `CHECK_{HASH}` 所在的 `{owner}/{repo}`；只管內容頁，管不到目錄頁 | 退回 `JSC_WIKI_REPO`；兩個都沒有就略過寫入，並把這一項列進待修 |
| `JSC_WIKI_REPO_CONTENTS` | 目錄頁 `CHECK_CONTENTS` 所在的 `{owner}/{repo}`。目錄頁與內容頁分屬不同存取庫：所有 `{TYPE}_CONTENTS` 一律住這一個庫，內容頁才看各自的型別變數 | 退回 `JSC_WIKI_REPO`；兩個都沒有就略過目錄頁寫入，並把這一項列進待修。不退回 `JSC_WIKI_REPO_CHECK` |
| `JSC_CONFIG_SPEC` | 改讀別份設定規格表（測試 `scan-config.sh` 時用） | 用 `tools/config-spec.tsv` |
| `JSC_HOME` | hook 資料目錄，兩份指引與重啟狀態檔都寫在這裡 | 用 `~/.jsc` |
| `JSC_RESTART_GATE` | 設成 `off` 可略過部署後的重啟提示閘門（判讀在 `jsc-hooks`，`jsc-cli` 只負責寫狀態檔） | 照常提示重啟 |
plugin.json
+1 -1
View File
@@ -1,6 +1,6 @@
{
"name": "jsc-cli",
"version": "0.2.6",
"version": "0.3.5",
"description": "CLI 偵測、模型能力標籤、子代理派工與技能庫批次部署",
"skills": "./skills/",
"jsc": {
references/behaviors.md
+20 -20
View File
@@ -7,47 +7,47 @@
| 項目 | 內容 |
| --- | --- |
| 觸發時機 | 一件邊界清楚的工作要交給另一支已安裝的 AI agent CLI 執行時用。盤點模型（models）、部署技能組（deploy）、必須留在目前 agent 手上的工作都不用；給不出通過或失敗判準的目標也不用 |
| 關鍵步驟 | 寫下一句目標與可判定通過或失敗的驗收條件、同時起跑 detect-clis.sh 與 list-models.sh 取得 CLI 清單與模型清單、先選定目標 CLI 再用 model-tags.sh 的 gate 或 model 驗證模型能力標籤、組出帶目標、驗收條件、目標 CLI、模型代號、最小脈絡、寫入範圍與 TSV 輸出合約的提示、開一個 sub agent 執行、查驗回傳的結束碼與 result、summary、criterion、wrote 各行、把成功、失敗、需要使用者補充分三段回報 |
| 外部呼叫 | jsc-cli/tools/detect-clis.sh、jsc-cli/tools/list-models.sh、jsc-cli/tools/model-tags.sh（gate 與 model 兩個子命令）、一個代跑工作的 sub agent |
| 完成條件 | 目標 CLI 與模型各只有一個，且模型通過能力要求；sub agent 回傳的每一條驗收條件都有判定；每個 wrote 路徑都落在寫入範圍內；報告分三段列出結果。中途停下時，停下的原因要寫在報告裡 |
| 可驗證跡象 | 技能自己不寫檔。可檢查的是 sub agent 依寫入範圍實際改動的檔案，逐條列在回傳的 wrote 行上，照那些路徑去看即可比對；寫入範圍為空的唯讀委派沒有寫入跡象，只有回報內容 |
| 關鍵步驟 | 寫下一句目標與可判定通過或失敗的驗收條件、同時起跑 detect-clis.sh 與 list-models.sh 取得 CLI 清單與模型清單、先選定目標 CLI 再用 model-tags.sh 的 gate 或 model 驗證模型能力標籤、組出帶目標、驗收條件、目標 CLI、模型代號、最小脈絡、寫入範圍與 TSV 輸出合約的提示、開一個 sub agent 執行、查驗回傳的結束碼與 result、summary、criterion、wrote 各行、把成功、失敗、需要使用者補充分三段回報，最後呼叫 jsc-hooks/tools/report-status.sh skill-end jsc-cli:delegate 寫下這一輪的結果。收尾那一筆走每一條出口，連停在能力閘門那一條也要寫；status 五選一，sub agent 回 0 且每條驗收條件都 pass、wrote 也都在範圍內是 ok，能力鎖擋下所有候選模型、或一支 CLI 都沒偵測到、整輪沒開 sub agent 是 blocked，sub agent 非零退出、回傳格式不符、有驗收條件 fail、或 wrote 越界是 failed，sub agent 回 needs-input、工作只做一半等使用者補資料是 degraded，目標給不出通過或失敗判準、或請求裡包了兩個以上獨立目標而主動停手是 aborted。detail 只放目標 CLI 與模型代號，不放 sub agent 的輸出。腳本不在這台機器就安靜跳過，回報失敗不得改變這支技能的結果 |
| 外部呼叫 | jsc-cli/tools/detect-clis.sh、jsc-cli/tools/list-models.sh、jsc-cli/tools/model-tags.sh（gate 與 model 兩個子命令）、jsc-hooks/tools/report-status.sh skill-end、一個代跑工作的 sub agent |
| 完成條件 | 目標 CLI 與模型各只有一個，且模型通過能力要求；sub agent 回傳的每一條驗收條件都有判定；每個 wrote 路徑都落在寫入範圍內；報告分三段列出結果。中途停下時，停下的原因要寫在報告裡。這一輪還要留下一筆 skill-end 事件，或是腳本不在而略過，兩者都算收好；略過不影響這支技能的結束碼 |
| 可驗證跡象 | 技能自己只寫這一筆事件。$JSC_HOME/usage/events.jsonl 會多一筆 {kind:skill,phase:end} 事件，name 是 jsc-cli:delegate，status 與 exit 就是這一輪的結果。另一項可檢查的是 sub agent 依寫入範圍實際改動的檔案，逐條列在回傳的 wrote 行上，照那些路徑去看即可比對；寫入範圍為空的唯讀委派沒有其他寫入跡象，只有回報內容與那一筆事件 |
## deploy
| 項目 | 內容 |
| --- | --- |
| 觸發時機 | 整組 jsc 技能要在這台機器的每一支已安裝 CLI 上安裝、更新或解除安裝時用。只處理單一技能不用；只想知道版本落後與否，看 doctor 就夠 |
| 關鍵步驟 | 同時取得三項事實（detect-clis.sh 的 CLI 清單、version-guard.sh 的 report 版本表與 recommend 結論、marketplace.json 的 domain 清單）、把版本表原樣秀出並定出建議、依 jsc-ask 決策樹問出模式（呼叫端已帶模式就沿用並標明來源）、每支 CLI 各開一個 sub agent 同時跑 tools/deploy.sh {mode} {cli} {domain}...、安裝或更新後把 CLI 清單交給 jsc-hooks:hooks-install、整台機器跑一次 tools/write-guides.sh、彙整每支 CLI 的結果並要求重新啟動工作階段 |
| 外部呼叫 | jsc-cli/tools/detect-clis.sh、jsc-cli/tools/deploy.sh、jsc-cli/tools/write-guides.sh、jsc-cli/tools/check-requires.sh（由 deploy.sh 在每個 domain 更新前轉呼叫）、jsc-hooks/hooks/version-guard.sh 的 report 與 recommend、jsc-hooks/hooks/restart-gate.sh require（由 deploy.sh 收尾轉呼叫）、jsc-gitea/tools/gitea.sh 讀 plugins/meta 的 marketplace.json、jsc-ask:ask、jsc-hooks:hooks-install |
| 完成條件 | 每一支偵測到的 CLI 都回報結束碼與 result 行，每個 skip、warn、compat 行都照實列出；安裝或更新還要拿到 hooks-install 對每支 CLI 的總結，兩份指引都印出 wrote，收尾印出重啟指示與兩份指引路徑 |
| 可驗證跡象 | 各 CLI 的外掛目錄多出或少掉 jsc-{domain}：claude 與 codex 在各自的 plugin 快取、copilot 在 installed-plugins、antigravity 與 kiro 走 $JSC_LOCAL_PLUGINS 的本地 clone 與 $JSC_KIRO_SKILLS 的複製。$JSC_HOME/restart-required.d/{cli} 出現這次的重啟狀態檔；$JSC_HOME/update-guide.md 與 $JSC_HOME/remove-guide.md 被重寫；各 CLI 的 hook 設定檔由 hooks-install 改寫 |
| 關鍵步驟 | 先跑一次 `readlink -f "$JSC_HOME/current"` 解出 `current` 這個目錄的絕對路徑，解到那一層就停，不再往下解成帶版本號的快取路徑——那種路徑放不進允許清單，版本號寫成萬用字元也對不上；同一步再跑 `[ -d "{剛印出來的路徑}" ]` 確認目錄存在，`JSC_HOME` 沒設時它印的是 `/current`、結束碼 0，非空又是絕對路徑，只看那兩項擋不下來。整輪只解這一次，之後每一次跨外掛腳本呼叫都填成那個字面絕對路徑，不留 `$JSC_HOME` 也不留波浪號；技能自己那四支 `tools/*.sh` 一律不走 `current`，即使那裡擺著 `jsc-cli` 那一條也一樣——第四步會為每個部署到的 domain 刷新連結，所以跑過一輪之後那一條通常在，但第一次建起它的正是這一輪，沒部署過的機器、或上一輪 link 回 fail 的機器，那一條不在或還停在舊版，四支腳本會解到不存在的路徑或自己的舊副本；根目錄取自 CLI 載入這支技能時講明的外掛基底目錄，原樣當字面絕對路徑用，一個指令都不跑，四支一律寫成 `{外掛根目錄}/tools/{腳本}`；那個基底目錄帶版本號，四次呼叫都會跳權限詢問，這一支有人在現場（第三步要問模式）所以按得掉，無人值守的技能不得照抄，叫用文字沒講明基底目錄就回報外掛根目錄不明並停手，不猜前綴——權限層靜態比對路徑，帶未展開變數的呼叫一律要人核准，無人看管的輪次會停在第一支腳本，補權限規則也擋不住，因為規則字面同樣是靜態比對；解不出來就停手回報。接著同時取得三項事實（detect-clis.sh 的 CLI 清單、version-guard.sh 的 report 版本表與 recommend 結論、marketplace.json 的 domain 清單）、把版本表原樣秀出並定出建議、依 jsc-ask 決策樹問出模式（呼叫端已帶模式就沿用並標明來源）、每支 CLI 各開一個 sub agent 同時跑 tools/deploy.sh {mode} {cli} {domain}...、讀 deploy.sh 收尾印出的 link 行確認 `current` 連結農場刷新到位（install 與 update 把每一條指到這次裝的版本目錄，uninstall 清掉指向已消失的那幾條；一台機器只有一組農場而五支 CLI 各有副本，所以只有基準 CLI 那一輪會動它，基準是 claude、codex、copilot、kiro 之中這台機器第一支裝得到的，其餘四支各印一行 link 標明基準是誰，平行覆寫會讓最後指到哪一份變成隨機；狀態 ok 要把指向抄進收尾報告，removed 不必處置，skip 帶 domain 是那個路徑上擺著非符號連結的東西、腳本刻意不覆寫而要人工處理，fail 是版本目錄取不到或連結建不起來、部署本身仍然成立但那條文件路徑可能還停在舊版，整輪判 degraded，連結失敗一律不記成部署失敗，否則會連重啟閘門與 result 行一起跳過，把一台裝好的機器講成失敗）、安裝或更新後把 CLI 清單交給 jsc-hooks:hooks-install、整台機器跑一次 tools/write-guides.sh、彙整每支 CLI 的結果並要求重新啟動工作階段，最後呼叫 jsc-hooks/tools/report-status.sh skill-end jsc-cli:deploy 寫下這一輪的結果。收尾那一筆接在彙整回報之後，不取代它；走每一條出口，連停在偵測不到 CLI 那一條也要寫。status 五選一，每支 CLI 都回 0、hooks-install 判定乾淨、兩份指引都寫成、基準 CLI 的 link 行全是 ok 或 removed 是 ok，偵測不到任何 CLI、整輪沒下過任何外掛命令是 blocked，marketplace 讀不到或每支 CLI 都失敗、冒煙結果出現 No such file 是 failed，部分 CLI 成功部分失敗、有 domain 被 skip、write-guides.sh 回 4 讓機器沒有最新指引、或有 link 行回 fail 與帶 domain 的 skip 是 degraded，使用者沒選模式或在第一支 CLI 開跑前停手是 aborted。detail 只放模式與各項筆數，cmd 與 exit 行留在回報裡 |
| 外部呼叫 | `readlink -f "$JSC_HOME/current"` 解出 `current` 這個目錄的絕對路徑，加上同一步的 `[ -d ]` 確認，是整輪唯一容許帶變數的兩個指令；跨外掛腳本一律用它組成的字面絕對路徑呼叫：{current 目錄}/jsc-hooks/hooks/version-guard.sh 的 report 與 recommend、{current 目錄}/jsc-gitea/tools/gitea.sh 讀 plugins/meta 的 marketplace.json、{current 目錄}/jsc-hooks/tools/report-status.sh skill-end。技能自己那四支腳本走外掛根目錄組成的字面絕對路徑：{外掛根目錄}/tools/detect-clis.sh、{外掛根目錄}/tools/deploy.sh、{外掛根目錄}/tools/write-guides.sh、{外掛根目錄}/tools/check-requires.sh（由 deploy.sh 在每個 domain 更新前轉呼叫）。第四步那段內文提到的 `jsc-hooks/hooks/version-guard.sh` 是在講擋人發生在哪一層，不是這支技能要下的呼叫，整輪只有第一步那一次真的跑它。另有 jsc-hooks/hooks/restart-gate.sh require（由 deploy.sh 收尾轉呼叫，install 與 update 一律呼叫，整輪判定成 fail 也照呼叫——外掛已經換了一部分，那時候更需要重啟；原本只在成功時呼叫，實測有一輪被一條與部署無關的 git pull 判成 fail，那一支 CLI 就沒有被掛上閘門）、jsc-ask:ask、jsc-hooks:hooks-install。`current` 連結農場的刷新不是另一支腳本，是 deploy.sh 自己收尾做的，技能只讀它印的 link 行，不自己下 ln 或 rm |
| 完成條件 | `current` 那個目錄在第一步就解出一條存在的絕對路徑（用 `[ -d ]` 查過，而且沒有再往下解成帶版本號的快取路徑），技能自己那四支腳本也有一條字面絕對的外掛根目錄可用，後續每一支腳本都用這兩條之一組成的字面絕對路徑呼叫；每一支偵測到的 CLI 都回報結束碼與 result 行，每個 skip、warn、compat、link 行都照實列出；基準 CLI 那一輪每個 domain 都有一條 link 行，狀態 ok 的把指向抄進收尾報告，狀態 fail 與帶 domain 的 skip 都點名外掛與原因並整輪判 degraded，非基準的那幾支各有一行標明基準是誰；安裝或更新還要拿到 hooks-install 對每支 CLI 的總結，兩份指引都印出 wrote，收尾印出重啟指示、兩份指引路徑，以及每個外掛的連結現在指到哪一個版本目錄。這一輪還要留下一筆 skill-end 事件，或是腳本不在而略過，兩者都算收好；略過不影響這支技能的結束碼 |
| 可驗證跡象 | 各 CLI 的外掛目錄多出或少掉 jsc-{domain}：claude 與 codex 在各自的 plugin 快取、copilot 在 installed-plugins、antigravity 與 kiro 走 $JSC_LOCAL_PLUGINS 的本地 clone 與 $JSC_KIRO_SKILLS 的複製；那一份本地 clone 一台機器只有一份而五支 CLI 平行跑，所以 deploy.sh 對每個 domain 取一把 mkdir 鎖再 pull，拿不到鎖或 pull 回非零時印一行 warn 並改用磁碟上現有的內容、不判整輪失敗（clone 不存在那一種照舊算失敗，磁碟上沒東西可裝）。$JSC_HOME/plugins/.lock-{domain} 在一輪跑完之後一個都不該留著。$JSC_HOME/current/ 底下每個外掛一條符號連結，安裝或更新之後拿 `readlink` 讀出來的指向，就是基準 CLI 這一輪裝到的那個帶版本號目錄，跟 link 行第五欄逐字相同，也跟 version-guard.sh report 那張表上該外掛的本機版本對得起來——這一項驗得到連結指向正確：連結上的版本號與剛裝上的版本號不一致，就是刷新沒做到；解除安裝之後，指向已消失的那幾條不再留在目錄裡，`readlink -e` 對每一條都解得出存在的目錄，沒有一條是斷的；那個目錄底下也不會出現實體目錄，非符號連結的項目腳本刻意不動並留下一行 skip。$JSC_HOME/restart-required.d/{cli} 出現這次的重啟狀態檔；$JSC_HOME/update-guide.md 與 $JSC_HOME/remove-guide.md 被重寫；各 CLI 的 hook 設定檔由 hooks-install 改寫；$JSC_HOME/usage/events.jsonl 會多一筆 {kind:skill,phase:end} 事件，name 是 jsc-cli:deploy，status 與 exit 就是這一輪的結果。回報與逐行紀錄裡出現的腳本路徑全是字面絕對路徑，找不到 `$JSC_HOME`、`$` 開頭或波浪號開頭的呼叫，唯一的例外是開頭那一次 `readlink -f "$JSC_HOME/current"` 與同一步的 `[ -d ]` 確認；跨外掛那幾支的路徑中段是 `current`，不是 `cache/jsc/{外掛}/{版本}`，技能自己那四支則一律是外掛根目錄接 `tools/`，沒有一支寫成裸的相對路徑 |
## doctor
| 項目 | 內容 |
| --- | --- |
| 觸發時機 | 裝完或更新完技能組、技能因設定或接線問題失敗、機器要交接前用。要動手修不用這支，那是 jsc-cli:setup |
| 關鍵步驟 | 同時開四個 sub agent 收版本、hook 接線、設定與未登錄變數，每個 sub agent 回傳原始輸出行、把四份輸出各存成檔、依 templates/check-page.md 印出五個區塊並寫明掃描的專案目錄、用 tools/build-todo.sh 把三份輸出合成待修項目表、透過 jsc-gitea:wiki 整頁覆寫 CHECK_{HASH} 並把本機那一列 upsert 進 CHECK_CONTENTS、報出四項計數並視情況建議 /jsc-cli:setup |
| 外部呼叫 | jsc-hooks/hooks/version-guard.sh report、jsc-cli/tools/detect-clis.sh、jsc-hooks/tools/wire-cli.sh status（一律帶 JSC_READONLY=1）、jsc-cli/tools/scan-config.sh 的 scan all 與 orphans、jsc-cli/tools/build-todo.sh、jsc-gitea/tools/gitea.sh 的 wiki-repo 與 hash-id、jsc-gitea:wiki |
| 完成條件 | 四項檢查各有結論，或明寫無法驗證與原因；五個區塊與待修項目表都在畫面上；wiki 頁寫成功，或寫入略過連同結束碼一起回報；必要項缺漏、設定錯誤、CLI 未接線、domain 落後四項計數都講出來 |
| 可驗證跡象 | wiki 的 CHECK_{HASH} 頁（雜湊來源是 {主機名}/{登入帳號}）被整頁覆寫成這次的結果，CHECK_CONTENTS 多出本機那一列，或該列的缺漏數與最後體檢時間被更新。機器本身的設定、接線與版本都不動：這支技能不寫任何設定 |
| 關鍵步驟 | 先跑一次 `readlink -f "$JSC_HOME/current"` 解出 `current` 這個目錄的絕對路徑，解到那一層就停，不再往下解成帶版本號的快取路徑——那種路徑放不進允許清單，版本號寫成萬用字元也對不上；同一步再跑 `[ -d "{剛印出來的路徑}" ]` 確認目錄存在，`JSC_HOME` 是有預設值的選擇性項目，沒設時它印的是 `/current`、結束碼 0，非空又是絕對路徑，只看那兩項擋不下來。這一支跟 deploy 不同，解不出來不停手：`JSC_HOME` 直接記成全域設定與待修項目上的一項發現，用得到跨外掛腳本的檢查記成無法驗證，其餘照跑到完——唯讀體檢中途停掉，操作者什麼都拿不到，而路徑解不開的機器正是最需要體檢的那一台。整輪只解這一次，之後每一次跨外掛腳本呼叫都填成那個字面絕對路徑，不留 `$JSC_HOME` 也不留波浪號，也不留裸的相對路徑——相對路徑會對著操作者的專案目錄解，那裡從來不是外掛根目錄，每個這種呼叫點都是前綴被猜出來的地方。技能自己的 `tools/` 與 `templates/` 一律不走 `current`，即使那裡擺著 `jsc-cli` 那一條也一樣：這一支正是機器可疑時才跑的技能，它的觸發時機自己就寫著裝完或更新完、技能因設定或接線失敗、機器要交接，而 deploy 判 degraded 那一輪留下的正是回 fail 或被 skip、還指著舊版的連結；從那條連結拿到的 scan-config.sh 與 config-spec.tsv 是這台機器沒在跑的版本，掃出來的每一列都像真的發現，舊掃描器跑得完，一句錯誤訊息都不會有。改走外掛根目錄就沒有這個問題，連結壞掉、過期或從來沒建起來的機器上，那三支腳本照樣跑得動，`JSC_HOME` 本身也才掃得到、報得出來。根目錄取自 CLI 載入這支技能時講明的外掛基底目錄，原樣當字面絕對路徑用，一個指令都不跑，寫成 `{外掛根目錄}/tools/{腳本}` 與 `{外掛根目錄}/templates/{檔案}`；那個基底目錄帶版本號，這幾次呼叫都會跳權限詢問，這一支有操作者在現場讀五個區塊與待修項目表所以按得掉，無人值守的技能不得照抄，叫用文字沒講明基底目錄就回報外掛根目錄不明並停手，不猜前綴。接著同時開四個 sub agent 收版本、hook 接線、設定與未登錄變數，每個 sub agent 回傳原始輸出行、把四份輸出各存成檔、依 templates/check-page.md 印出五個區塊並寫明掃描的專案目錄、用 tools/build-todo.sh 把三份輸出合成待修項目表、用程式取短主機名與登入帳號（主機名切掉第一個點之後的網域）交給 hash-id 算出 HASH、用 wiki-repo CHECK 解出的存取庫透過 jsc-gitea:wiki 整頁覆寫 CHECK_{HASH}、寫完再用 wiki-url 取該頁絕對網址、把要寫進兩頁的每個連結交給 jsc-gitea/tools/link-check.sh 驗證且只有結束碼 0 才往下寫、「體檢頁」那一條的連結寫成 `[CHECK_{HASH}]({絕對網址})`、用 wiki-contents.sh upsert CHECK 1 CHECK_{HASH} 以 H2 標題也就是體檢頁頁名當鍵，把本機那一個區塊寫進 CONTENTS 存取庫的 CHECK_CONTENTS，區塊檔是 `## CHECK_{HASH}` 那一行、一個空行，再照 templates/check-contents.md 的欄位順序每欄一條 `- {欄位名}：{值}`、報出四項計數並視情況建議 /jsc-cli:setup，最後呼叫 jsc-hooks/tools/report-status.sh skill-end jsc-cli:doctor 寫下這一輪的結果。這一筆是這支技能唯一的寫入動作，記的是查到什麼，不動設定、不動接線、不動版本，唯讀合約照樣成立；走每一條出口都要寫。status 五選一，四項檢查都有結論、五個區塊與待修項目表都在畫面上、兩頁都寫成是 ok，任何一項報成無法驗證、離線讓 Gitea 相關列變成 skipped、或 wiki-repo 回 3 而略過寫頁是 degraded——唯讀技能讀不到來源就是這一種，金鑰失效回 7 或其他 API 失敗回 8 讓紀錄寫不成是 failed，使用者在寫頁之前喊停是 aborted。blocked 這支用不到：沒 CLI、沒登錄檔、沒 wiki 存放庫的機器一樣查得出四項發現，報成 blocked 會把做完的一輪講成沒做事。detail 只放四項計數 |
| 外部呼叫 | `readlink -f "$JSC_HOME/current"` 解出 `current` 這個目錄的絕對路徑，加上同一步的 `[ -d ]` 確認，是整輪唯一容許帶變數的兩個指令；跨外掛腳本一律用它組成的字面絕對路徑呼叫：{current 目錄}/jsc-hooks/hooks/version-guard.sh report、{current 目錄}/jsc-hooks/tools/wire-cli.sh status（一律帶 JSC_READONLY=1）、{current 目錄}/jsc-gitea/tools/gitea.sh 的 wiki-repo、hash-id 與 wiki-url、{current 目錄}/jsc-gitea/tools/link-check.sh、{current 目錄}/jsc-gitea/tools/wiki-contents.sh upsert、{current 目錄}/jsc-hooks/tools/report-status.sh skill-end。技能自己的檔案走外掛根目錄組成的字面絕對路徑：{外掛根目錄}/tools/detect-clis.sh、{外掛根目錄}/tools/scan-config.sh 的 scan all 與 orphans、{外掛根目錄}/tools/build-todo.sh，比對用的 {外掛根目錄}/tools/config-spec.tsv 與兩份版型 {外掛根目錄}/templates/check-page.md、{外掛根目錄}/templates/check-contents.md 也一樣。另有 jsc-gitea:wiki |
| 完成條件 | `current` 那個目錄在第一步就解出一條存在的絕對路徑（用 `[ -d ]` 查過，而且沒有再往下解成帶版本號的快取路徑），解不出來就記成 `JSC_HOME` 那一項發現而不是停手；技能自己的 `tools/` 與 `templates/` 也有一條字面絕對的外掛根目錄可用，後續每一支腳本、每一份版型都用這兩條之一組成的字面絕對路徑呼叫。四項檢查各有結論，或明寫無法驗證與原因；五個區塊與待修項目表都在畫面上；每個要寫進頁面的連結都經 link-check.sh 驗過，結束碼 0 才寫，1 就兩頁都不寫並列出 DEAD 那幾筆，3 把 GITEA_HOST 排進待修項目最前面，7 停下來回報金鑰問題而不判成死連結；連結一律寫成文字加絕對網址的形式，H2 標題本身不放連結；兩頁各自寫成功，或寫入略過連同結束碼一起回報，wiki-contents.sh 宣告的 0、1、2、3、4、7、8 每一碼都有分流，結束碼 1 是組不出頁面內容或寫入失敗，頁上找不到本機那一個區塊不算錯、腳本改成附加，建不建新頁的判斷留在腳本裡，技能不自己建；必要項缺漏、設定錯誤、CLI 未接線、domain 落後四項計數都講出來。這一輪還要留下一筆 skill-end 事件，或是腳本不在而略過，兩者都算收好；略過不影響這支技能的結束碼 |
| 可驗證跡象 | wiki 的 CHECK_{HASH} 頁（雜湊來源是 {短主機名}/{登入帳號}）被整頁覆寫成這次的結果；另一個存取庫的 CHECK_CONTENTS 多出本機那一個 H2 區塊，或該區塊的缺漏數與最後體檢時間被更新；區塊標題是 `## CHECK_{HASH}`，也就是比對用的鍵，標題上沒有連結也沒有網址，「體檢頁」那一條是 `[CHECK_{HASH}]({絕對網址})` 這種文字加連結的寫法，點下去連得到體檢頁，頁面上找不到同 wiki 的雙括號連結；欄位一律是 `- {欄位名}：{值}` 的條列，頁上沒有 markdown 表格，同一台機器重跑幾次都只有這一個區塊，別台機器的區塊一個位元組都沒變；連結驗不過的那一輪，兩頁都維持上一輪的內容。$JSC_HOME/usage/events.jsonl 會多一筆 {kind:skill,phase:end} 事件，name 是 jsc-cli:doctor，status 與 exit 就是這一輪的結果，那也是這支技能唯一寫得出來的檔案痕跡。機器本身的設定、接線與版本都不動：這支技能不寫任何設定。回報與逐行紀錄裡出現的腳本路徑全是字面絕對路徑，找不到 `$JSC_HOME`、`$` 開頭或波浪號開頭的呼叫，也沒有一支寫成裸的相對路徑，唯一的例外是開頭那一次 `readlink -f "$JSC_HOME/current"` 與同一步的 `[ -d ]` 確認；跨外掛那幾支的路徑中段是 `current`，不是 `cache/jsc/{外掛}/{版本}`，技能自己那三支腳本與兩份版型則一律是外掛根目錄接 `tools/` 或 `templates/` |
## models
| 項目 | 內容 |
| --- | --- |
| 觸發時機 | 要盤點各 CLI 可用模型、確認模型合不合 SDLC 階段的能力要求、或檢視階段閘門設定時用。切換模型不用，改 .jsc/models 與 models.conf 也不用 |
| 關鍵步驟 | 同時起跑三個收集器（detect-clis.sh、list-models.sh 以 sub agent 執行、model-config.sh list）、對讀不到設定的 CLI 補上標「預設推定」的預設模型、依 references/model-tags.md 為每個模型掛能力標籤、印出 CLI、模型、標籤、使用中四欄表、跑 tools/model-tags.sh sync 把標籤表寫進 $JSC_HOME/model-tags.tsv、附上 SDLC 階段需求表與階段偏好模型表 |
| 外部呼叫 | jsc-cli/tools/detect-clis.sh、jsc-cli/tools/list-models.sh、jsc-cli/tools/model-config.sh list、jsc-cli/tools/model-tags.sh sync、references/model-tags.md；標籤表上查不到的模型改用 jsc-ask:ask 發問 |
| 完成條件 | 每支偵測到的 CLI 都有模型清單或一組預設推定；每個模型都掛到標籤，或已排進發問；sync 印出寫入路徑，失敗則連同結束碼回報；兩張階段表都列滿 plan、analyze、implement、maintain 四個階段 |
| 可驗證跡象 | $JSC_HOME/model-tags.tsv 被重寫，內容就是這次掛好的標籤表；jsc-hooks/hooks/sdlc-gate.sh 讀的正是這份檔，檔案不在，SDLC 階段閘門就判不出來。各 CLI 的模型設定檔不動 |
| 關鍵步驟 | 同時起跑三個收集器（detect-clis.sh、list-models.sh 以 sub agent 執行、model-config.sh list）、對讀不到設定的 CLI 補上標「預設推定」的預設模型、依 references/model-tags.md 為每個模型掛能力標籤、印出 CLI、模型、標籤、使用中四欄表、跑 tools/model-tags.sh sync 把標籤表寫進 $JSC_HOME/model-tags.tsv、附上 SDLC 階段需求表與階段偏好模型表，最後呼叫 jsc-hooks/tools/report-status.sh skill-end jsc-cli:models 寫下這一輪的結果。收尾那一筆走每一條出口，連停在偵測不到 CLI 那一條也要寫；status 五選一，每支 CLI 都有模型清單、每個模型都掛到標籤、sync 回 0 印出路徑、兩張階段表都在是 ok，偵測不到任何 CLI、盤點那半段整個沒開始是 blocked，sync 非零而 model-tags.tsv 沒寫成、SDLC 閘門判不出來是 failed，讀不到某支 CLI 的設定而改用預設推定、有模型查不到標籤只能排進發問、或 model-config.sh 失敗讓偏好表變成未取得是 degraded，使用者在 sync 寫檔之前停手是 aborted。detail 只放 CLI 與模型筆數 |
| 外部呼叫 | jsc-cli/tools/detect-clis.sh、jsc-cli/tools/list-models.sh、jsc-cli/tools/model-config.sh list、jsc-cli/tools/model-tags.sh sync、jsc-hooks/tools/report-status.sh skill-end、references/model-tags.md；標籤表上查不到的模型改用 jsc-ask:ask 發問 |
| 完成條件 | 每支偵測到的 CLI 都有模型清單或一組預設推定；每個模型都掛到標籤，或已排進發問；sync 印出寫入路徑，失敗則連同結束碼回報；兩張階段表都列滿 plan、analyze、implement、maintain 四個階段。這一輪還要留下一筆 skill-end 事件，或是腳本不在而略過，兩者都算收好；略過不影響這支技能的結束碼 |
| 可驗證跡象 | $JSC_HOME/model-tags.tsv 被重寫，內容就是這次掛好的標籤表；jsc-hooks/hooks/sdlc-gate.sh 讀的正是這份檔，檔案不在，SDLC 階段閘門就判不出來。$JSC_HOME/usage/events.jsonl 會多一筆 {kind:skill,phase:end} 事件，name 是 jsc-cli:models，status 與 exit 就是這一輪的結果。各 CLI 的模型設定檔不動 |
## setup
| 項目 | 內容 |
| --- | --- |
| 觸發時機 | doctor 報出待修項目、要實際動手修這台機器時用。只想做唯讀體檢不用這支，那是 jsc-cli:doctor |
| 關鍵步驟 | 從 wiki CHECK_{HASH} 讀待修項目表，讀不到就以 sub agent 同時重跑設定、接線、版本三個檢查器再用 build-todo.sh 合併、依 jsc-ask 決策樹逐項循序確認、依 fix 欄分流（auto 與 ask 走 apply-config.sh 的 set 或 mkdir、manual 印出步驟交給操作者、domain 落後轉呼叫 jsc-cli:deploy 並附上手上的版本報告、hook 未接線轉呼叫 jsc-hooks:hooks-install、缺 model-tags.tsv 轉呼叫 jsc-cli:models）、同時重驗每個已套用項目、重寫 CHECK_{HASH} 並 upsert CHECK_CONTENTS、報出已修、略過、轉呼叫、未修好四項計數 |
| 外部呼叫 | jsc-cli/tools/scan-config.sh、jsc-cli/tools/detect-clis.sh、jsc-hooks/tools/wire-cli.sh status（一律帶 JSC_READONLY=1）、jsc-hooks/hooks/version-guard.sh report、jsc-cli/tools/build-todo.sh、jsc-cli/tools/apply-config.sh 的 set、mkdir 與 show、jsc-gitea/tools/gitea.sh 的 wiki-repo 與 hash-id、jsc-ask:ask、jsc-gitea:wiki、jsc-cli:deploy、jsc-hooks:hooks-install、jsc-cli:models |
| 完成條件 | 每一項都有已修、略過、轉呼叫或未修好的結果；每個已套用項目都由自己那一列指定的檢查器重驗過；每個寫進去的環境變數都附上 export 那一行；頁面寫好或略過都有回報；四項計數都講出來 |
| 可驗證跡象 | 各 shell rc 檔的 `# jsc-config` 區塊被改寫，改寫前的備份落在 $JSC_HOME/backup/config/{時間戳}/；auto 路線建立的目錄實際出現在磁碟上；wiki CHECK_{HASH} 被改寫成修完後的狀態，CHECK_CONTENTS 的本機列跟著更新；轉呼叫出去的項目留下各自技能的跡象，也就是 deploy 的重啟狀態檔、hooks-install 改寫的接線設定、models 產生的 model-tags.tsv |
| 關鍵步驟 | 先跑一次 `readlink -f "$JSC_HOME/current"` 解出 `current` 這個目錄的絕對路徑，解到那一層就停，不再往下解成帶版本號的快取路徑——那種路徑放不進允許清單，版本號寫成萬用字元也對不上；同一步再跑 `[ -d "{剛印出來的路徑}" ]` 確認目錄存在。全技能組裡最常碰到 `JSC_HOME` 沒設的就是這一支：它是有預設值的選擇性項目，在待修項目表上是一列 auto，而把它寫進去正是這支技能要做的修復，所以那種機器本來就是叫用這支技能的常態理由；沒設時那道指令印的是 `/current`、結束碼 0，非空又是絕對路徑，兩項都擋不下來。解不出來既不停這一輪，也不擋那一項修復：寫 `JSC_HOME` 要用的 apply-config.sh、scan-config.sh、build-todo.sh 全掛在外掛根目錄底下，根本不靠 `JSC_HOME`，先把那一項修好、再重解一次 `current` 就能往下走，仍然構不到的（第五步的 wiki 紀錄、轉呼叫出去的修復）記成未修好並寫明原因，不用猜的。整輪只解這一次，之後每一次跨外掛腳本呼叫都填成那個字面絕對路徑，不留 `$JSC_HOME` 也不留波浪號，也不留裸的相對路徑——相對路徑會對著操作者的專案目錄解，那裡從來不是外掛根目錄。技能自己的 `tools/` 與 `templates/` 一律不走 `current`，即使那裡擺著 `jsc-cli` 那一條也一樣，理由有兩條：一是上面那條，要修的機器往往正是 `JSC_HOME` 壞掉的機器，修復工具走連結農場，就會被它們存在的理由本身擋住；二是這一支會拿舊腳本去寫檔，第三步把每個落後的 domain 交給 jsc-cli:deploy，正因為這台機器的版本可能落後，而連結農場受同一份落後管轄，從那裡拿 apply-config.sh，等於拿這台機器剛被判定已經跟不上的工具去修它，而且結果是寫進 rc 檔，不只是印在畫面上——舊腳本會照它那一版的鍵集重寫每個 rc 檔的 `# jsc-config` 區塊，把舊的當成正確版本備份起來，第四步再用同一份舊的 show 重驗，當然對得上，於是整輪報成已修。根目錄取自 CLI 載入這支技能時講明的外掛基底目錄，原樣當字面絕對路徑用，一個指令都不跑，寫成 `{外掛根目錄}/tools/{腳本}` 與 `{外掛根目錄}/templates/{檔案}`；那個基底目錄帶版本號，這幾次呼叫都會跳權限詢問，這一支有操作者在現場（第二步逐項問到答案才寫）所以按得掉，無人值守的技能不得照抄，叫用文字沒講明基底目錄就在寫任何東西之前回報外掛根目錄不明並停手，不猜前綴。接著用程式取短主機名與登入帳號算出 HASH，從 wiki-repo CHECK 解出的存取庫讀 CHECK_{HASH} 的待修項目表，讀不到就以 sub agent 同時重跑設定、接線、版本三個檢查器再用 build-todo.sh 合併、依 jsc-ask 決策樹逐項循序確認、依 fix 欄分流（auto 與 ask 走 apply-config.sh 的 set 或 mkdir、manual 印出步驟交給操作者、domain 落後轉呼叫 jsc-cli:deploy 並附上手上的版本報告、hook 未接線轉呼叫 jsc-hooks:hooks-install、缺 model-tags.tsv 轉呼叫 jsc-cli:models）、同時重驗每個已套用項目、重寫 CHECK_{HASH}、再用 wiki-url 取它的絕對網址、把要寫進兩頁的每個連結交給 jsc-gitea/tools/link-check.sh 驗證且只有結束碼 0 才往下寫、「體檢頁」那一條的連結寫成 `[CHECK_{HASH}]({絕對網址})`、以 wiki-contents.sh upsert CHECK 1 CHECK_{HASH} 用 H2 標題也就是體檢頁頁名當鍵，更新 CONTENTS 存取庫的 CHECK_CONTENTS 上本機那一個區塊，區塊檔是 `## CHECK_{HASH}` 那一行、一個空行，再照 templates/check-contents.md 的欄位順序每欄一條 `- {欄位名}：{值}`、報出已修、略過、轉呼叫、未修好四項計數，最後呼叫 jsc-hooks/tools/report-status.sh skill-end jsc-cli:setup 寫下這一輪的結果。收尾那一筆走每一條出口，連停在讀不到待修項目那一條也要寫；status 五選一，每一項都修好也重驗過、沒有略過、兩頁都寫成是 ok，環境不允許寫入、apply-config.sh 每一項都回 4 而一個 rc 檔都沒動到是 blocked，已套用的項目重驗仍失敗、apply-config.sh 回 2 是這支技能自己下錯命令、或金鑰失效回 7 與其他 API 失敗回 8 讓紀錄改寫不成是 failed，使用者否決某一項而那一項記成略過、或轉呼叫出去的修正沒收尾是 degraded，使用者在逐項確認到一半喊停、剩下的項目沒問到是 aborted。detail 只放四項計數，不放使用者輸入的值 |
| 外部呼叫 | `readlink -f "$JSC_HOME/current"` 解出 `current` 這個目錄的絕對路徑，加上同一步的 `[ -d ]` 確認，是整輪唯一容許帶變數的兩個指令；跨外掛腳本一律用它組成的字面絕對路徑呼叫：{current 目錄}/jsc-hooks/tools/wire-cli.sh status（一律帶 JSC_READONLY=1）、{current 目錄}/jsc-hooks/hooks/version-guard.sh report、{current 目錄}/jsc-gitea/tools/gitea.sh 的 wiki-repo、hash-id 與 wiki-url、{current 目錄}/jsc-gitea/tools/link-check.sh、{current 目錄}/jsc-gitea/tools/wiki-contents.sh upsert、{current 目錄}/jsc-hooks/tools/report-status.sh skill-end。技能自己的檔案走外掛根目錄組成的字面絕對路徑：{外掛根目錄}/tools/scan-config.sh、{外掛根目錄}/tools/detect-clis.sh、{外掛根目錄}/tools/build-todo.sh、{外掛根目錄}/tools/apply-config.sh 的 set、mkdir 與 show，比對用的 {外掛根目錄}/tools/config-spec.tsv 與兩份版型 {外掛根目錄}/templates/check-page.md、{外掛根目錄}/templates/check-contents.md 也一樣。另有 jsc-ask:ask、jsc-gitea:wiki、jsc-cli:deploy、jsc-hooks:hooks-install、jsc-cli:models |
| 完成條件 | `current` 那個目錄在第一步就解出一條存在的絕對路徑（用 `[ -d ]` 查過，而且沒有再往下解成帶版本號的快取路徑），解不出來就把它記成 `JSC_HOME` 那一項待修並照修，不停手也不擋住其他項；技能自己的 `tools/` 與 `templates/` 也有一條字面絕對的外掛根目錄可用，後續每一支腳本、每一份版型都用這兩條之一組成的字面絕對路徑呼叫。每一項都有已修、略過、轉呼叫或未修好的結果；每個已套用項目都由自己那一列指定的檢查器重驗過；每個寫進去的環境變數都附上 export 那一行；每個要寫進頁面的連結都經 link-check.sh 驗過，結束碼 0 才寫，1 就兩頁都不寫並列出 DEAD 那幾筆，3 回報 GITEA_HOST 仍未修好，7 停下來回報金鑰問題而不判成死連結；連結一律寫成文字加絕對網址的形式，H2 標題本身不放連結；兩頁各自寫好或略過都有回報，wiki-contents.sh 宣告的 0、1、2、3、4、7、8 每一碼都有分流，結束碼 1 是組不出頁面內容或寫入失敗，頁上找不到本機那一個區塊不算錯、腳本改成附加，建不建新頁的判斷留在腳本裡，技能不自己建；四項計數都講出來。這一輪還要留下一筆 skill-end 事件，或是腳本不在而略過，兩者都算收好；略過不影響這支技能的結束碼 |
| 可驗證跡象 | 各 shell rc 檔的 `# jsc-config` 區塊被改寫，改寫前的備份落在 $JSC_HOME/backup/config/{時間戳}/；auto 路線建立的目錄實際出現在磁碟上；wiki CHECK_{HASH} 被改寫成修完後的狀態，另一個存取庫的 CHECK_CONTENTS 只有本機那一個 H2 區塊跟著更新，區塊標題是當鍵用的 `## CHECK_{HASH}`，標題上沒有連結也沒有網址，「體檢頁」那一條是 `[CHECK_{HASH}]({絕對網址})` 這種文字加連結的寫法、點下去連得到體檢頁，頁面上找不到同 wiki 的雙括號連結，欄位一律是 `- {欄位名}：{值}` 的條列、頁上沒有 markdown 表格；連結驗不過的那一輪，兩頁都維持上一輪的內容；轉呼叫出去的項目留下各自技能的跡象，也就是 deploy 的重啟狀態檔、hooks-install 改寫的接線設定、models 產生的 model-tags.tsv；$JSC_HOME/usage/events.jsonl 會多一筆 {kind:skill,phase:end} 事件，name 是 jsc-cli:setup，status 與 exit 就是這一輪的結果。回報與逐行紀錄裡出現的腳本路徑全是字面絕對路徑，找不到 `$JSC_HOME`、`$` 開頭或波浪號開頭的呼叫，也沒有一支寫成裸的相對路徑，唯一的例外是開頭那一次 `readlink -f "$JSC_HOME/current"` 與同一步的 `[ -d ]` 確認；跨外掛那幾支的路徑中段是 `current`，不是 `cache/jsc/{外掛}/{版本}`，技能自己那四支腳本與兩份版型則一律是外掛根目錄接 `tools/` 或 `templates/`；備份目錄 $JSC_HOME/backup/config/{時間戳}/ 底下那幾份 rc 檔，是由外掛根目錄那一份 apply-config.sh 寫出來的，跟這台機器目前跑的 jsc-cli 版本同一份 |
skills/delegate/SKILL.md
+16
View File
@@ -97,6 +97,22 @@ Detection and tag filtering are decided in code, not in prose: `jsc-cli/tools/de
Done when the CLI, the model, the requirement and the per-criterion verdicts are all in the report, and every failure recorded in step 6 appears in the failure section.
8. **Record how the run ended.** This is the last thing this skill does, and it runs on every path out of the skill, the ones that stop at step 1 or step 3 included. Call
`jsc-hooks/tools/report-status.sh skill-end jsc-cli:delegate {status} {exit code} [detail]`
`{exit code}` is the exit code of whatever decided the outcome — the subagent's own status, or the `detect-clis.sh` or `model-tags.sh` call that ruled the run — and `0` when nothing failed. `{detail}` is one short line, no more than 200 characters: the target CLI and the model id fit there, the subagent's output does not. **If the script is not on this machine, skip this step in silence and finish the run as it stood** — missing infrastructure is not a failure, and a reporting call may never change what this skill returns or reports.
| status | When this skill uses it |
| --- | --- |
| `ok` | The subagent exited 0, its output held one `result` and one `summary` line, every acceptance criterion came back `pass`, and every `wrote` path sat inside the write scope |
| `blocked` | The capability gate stopped the delegation before it started, so no subagent ran: every candidate model returned `FAIL` or `UNKNOWN-MODEL` from `model-tags.sh gate`, `detect-clis.sh` exited 0 with no row, the CLI the user named is not on this machine, or the target CLI has no model row and no forced model |
| `failed` | The delegation ran and broke: the subagent exited non-zero, its exit status could not be obtained at all, its output was missing the `result` or `summary` line, a criterion came back `fail`, or a `wrote` path landed outside the write scope. `detect-clis.sh` or `list-models.sh` exiting non-zero sits here too |
| `degraded` | The subagent returned `result needs-input`: part of the goal is done and the rest waits on the user, so the report has a 需要使用者補充 section that is not empty. The delegation happened, the goal did not close |
| `aborted` | The premise did not hold, so the skill stopped on its own: step 1 could give the goal no pass-or-fail acceptance criterion, or the request carried two or more independent goals and has to be split. Also used when the user stops the run before step 5 spawns the subagent |
Done when exactly one `skill-end` line was recorded for this run, or the script was absent and the run finished without it.
## Do not use this skill
- Do not use it for model inventory. That is `/jsc-cli:models`.
skills/deploy/SKILL.md
+82 -13
View File
@@ -5,6 +5,38 @@ description: Batch install, update, or uninstall the whole jsc skill set on ever
# deploy — batch install, update, or uninstall the skill set
## Path rule — every script call is a literal absolute path
Write every script call in this skill as a literal absolute path. Never hand the shell a path that still holds a variable or a tilde — `$JSC_HOME/...`, `~/.jsc/...`, or anything like them. The permission layer matches paths statically. It never expands a variable or a tilde, so such a path matches no allow rule, and the call falls through to an approval prompt. An unattended round has nobody to approve it. The run then dies at its first script, before it deploys anything.
Measured on a real machine, not assumed. `$JSC_HOME/current/jsc-assist/tools/patrol.sh` was blocked and never ran. `~/.jsc/current/...` was blocked and never ran. `/root/.jsc/current/...` ran. Adding an allow rule that itself starts with `$JSC_HOME` to `~/.claude/settings.json` changed nothing: the call stayed blocked. A wider allow list is not the fix, because the rule text is matched statically too.
Portability is no reason to put the variable back. Step 0 resolves the root once, at run time, on whatever machine this runs on — that is where portability comes from. Rewriting `{JSC_ROOT}/jsc-hooks/...` back to `$JSC_HOME/current/jsc-hooks/...` for tidiness re-breaks every unattended round.
## Step 0 — resolve the two roots, once
Before step 1, run this one command:
`readlink -f "$JSC_HOME/current"`
It prints one absolute directory: the absolute path of the `current` directory itself. Call it `{JSC_ROOT}` for the rest of this document. **Stop at that directory — never resolve one level further.** `current` is an ordinary directory, and the symbolic links are its entries, one per plugin; resolving one of those entries lands on the versioned plugin cache (`/root/.claude/plugins/cache/jsc/jsc-hooks/0.4.2`, say), and a versioned path is exactly the kind no allow rule can hold — a rule with `*` where the version segment goes matches nothing, measured. `{JSC_ROOT}` is the version-free root, and staying at it is the whole point. This is the only place a variable may appear. The shell expands it inside the command itself, so no unexpanded path ever reaches the permission layer.
Substitute `{JSC_ROOT}` with that directory in every later call, so what runs is a literal absolute path. `{JSC_ROOT}/jsc-hooks/hooks/version-guard.sh` becomes, for example, `/root/.jsc/current/jsc-hooks/hooks/version-guard.sh`.
Resolve it once, at the start of the run. Do not re-resolve it per call. Do not add a tool that prints it.
Empty output, a non-zero exit, or a path that is not an existing directory → stop and report that `$JSC_HOME/current` does not resolve. **The third item is the one the first two wave through**, so check it: with `JSC_HOME` unset the command prints `/current` and exits 0 — non-empty, absolute, and nowhere — and every literal path built from it then names a place that is not there. Run `[ -d "{the path just printed}" ]` in the same approved step as the resolve, and treat only an existing directory as a root. Every cross-plugin script this skill calls lives under it.
### The second root — this skill's own `tools/`
**Do not reach this skill's own `tools/` through `{JSC_ROOT}`, even when a `jsc-cli` link is sitting there.** Step 4 refreshes the farm for every deployed domain, so after a round has run, `{JSC_ROOT}/jsc-cli` usually exists — but this skill cannot rely on it, because the round that first creates it is this very one. On a machine that has never deployed, and on any machine where the last round's link write came back `fail`, `{JSC_ROOT}/jsc-cli` is absent or stale, and `tools/detect-clis.sh`, `tools/deploy.sh`, `tools/check-requires.sh` and `tools/write-guides.sh` would resolve to nothing or to a superseded copy of themselves. None of the four may be written as a bare relative path either — the rule above wants a literal absolute path at every call site, and a call site with no way to build one is where a prefix gets guessed.
They sit at `{plugin root}/tools/`, and the plugin root is the base directory the CLI states when it loads this skill. Take that literal path verbatim, call it `{CLI_ROOT}`, and write all four calls as `{CLI_ROOT}/tools/{script}` — `/root/.claude/plugins/cache/jsc/jsc-cli/0.3.2/tools/deploy.sh`, for example. No command runs for this one, and it is taken once, like `{JSC_ROOT}`.
That base directory carries a version segment, so no allow rule covers it and each of those four calls raises an approval prompt. **That is acceptable in this skill and in no unattended one**: `deploy` runs with a person in front of it — step 3 asks them for the mode, step 4 rewrites every CLI's plugin set — so there is somebody to approve. Never carry this branch into a skill that runs from a scheduler, and never guess a prefix when the invocation states no base directory: report that this skill's own plugin root is unknown and stop, because a guessed prefix runs some other version's copy of these scripts, or nothing at all.
Done when `{JSC_ROOT}` holds one existing absolute directory and `{CLI_ROOT}` holds one literal absolute path.
## Inputs a caller may pass
`jsc-cli:setup` already confirmed the mode with the user and already holds a fresh version report. Re-asking and re-querying would put a second decision tree in front of someone who just answered it.
@@ -20,13 +52,13 @@ Nothing passed in → run every step as written below.
1. Collect the three facts the rest of the run needs. They are independent, so start all three at once and wait for all three.
1. **Installed CLIs** — `tools/detect-clis.sh`, printing `{name}<TAB>{path}<TAB>{version}`. Exit 0 with at least one row → take the CLI list from it. Exit 0 with no row → stop, and report that none of claude, codex, copilot, antigravity, kiro is installed. Any non-zero exit → stop and report the exit code and stderr; never guess a CLI list.
2. **Version evidence and recommendation** — two subcommands of `jsc-hooks/hooks/version-guard.sh`, both needed, run together: `report` prints the per-plugin rows `{domain}<TAB>{本機}<TAB>{遠端}<TAB>{落後|最新|超前|查詢失敗}` closing with `behind<TAB>{count}`, and `recommend` prints one single line and nothing else — `recommend<TAB>{update|none|unverifiable}`. `recommend` deliberately never reprints the table, so its second column stays readable by `cut`; the version table that steps 2, 3 and 7 show comes from `report`, and the conclusion comes from `recommend`. Skip this collector when the caller passed a version report.
1. **Installed CLIs** — `{CLI_ROOT}/tools/detect-clis.sh`, printing `{name}<TAB>{path}<TAB>{version}`. Exit 0 with at least one row → take the CLI list from it. Exit 0 with no row → stop, and report that none of claude, codex, copilot, antigravity, kiro is installed. Any non-zero exit → stop and report the exit code and stderr; never guess a CLI list.
2. **Version evidence and recommendation** — two subcommands of `{JSC_ROOT}/jsc-hooks/hooks/version-guard.sh`, both needed, run together: `report` prints the per-plugin rows `{domain}<TAB>{本機}<TAB>{遠端}<TAB>{落後|最新|超前|查詢失敗}` closing with `behind<TAB>{count}`, and `recommend` prints one single line and nothing else — `recommend<TAB>{update|none|unverifiable}`. `recommend` deliberately never reprints the table, so its second column stays readable by `cut`; the version table that steps 2, 3 and 7 show comes from `report`, and the conclusion comes from `recommend`. Skip this collector when the caller passed a version report.
3. **Domain list** — read `plugins[].name` from the unified marketplace (**never hardcode it**; this skill then follows automatically when domains are added or removed):
`jsc-gitea/tools/gitea.sh api GET /repos/plugins/meta/raw/.claude-plugin/marketplace.json`.
`{JSC_ROOT}/jsc-gitea/tools/gitea.sh api GET /repos/plugins/meta/raw/.claude-plugin/marketplace.json`.
Exit 0 with at least one `plugins[].name` → use that list. Exit 0 with an empty or unparseable list → stop and report that the marketplace holds no plugin entry. Any non-zero exit → stop and report the exit code and stderr; a partial domain list would install a partial skill set and look successful.
The marketplace is unified as `jsc`; the install token is `jsc-{domain}@jsc`. Each `plugins[].name` already carries the `jsc-` prefix (e.g. `jsc-ask`) — pass it to `tools/deploy.sh` as-is, prefixed or not; the script normalizes it.
The marketplace is unified as `jsc`; the install token is `jsc-{domain}@jsc`. Each `plugins[].name` already carries the `jsc-` prefix (e.g. `jsc-ask`) — pass it to `{CLI_ROOT}/tools/deploy.sh` as-is, prefixed or not; the script normalizes it.
Done when the CLI list holds at least one CLI, the domain list holds at least one name, and the recommendation is either in hand or explicitly inherited from the caller.
@@ -50,9 +82,9 @@ Nothing passed in → run every step as written below.
Done when the user has named exactly one of `install`, `update` or `uninstall`, or the inherited mode is named with its source.
4. Run `tools/deploy.sh {mode} {cli} {domain}...` once per detected CLI, passing the whole domain list in one call so the marketplace command runs only once. This step **MUST run as a sub agent**, one sub agent per CLI, and **all of them start together** — the CLIs write to separate plugin directories, so serialising them only adds up their install times.
4. Run `{CLI_ROOT}/tools/deploy.sh {mode} {cli} {domain}...` once per detected CLI, passing the whole domain list in one call so the marketplace command runs only once. This step **MUST run as a sub agent**, one sub agent per CLI, and **all of them start together** — the CLIs write to separate plugin directories, so serialising them only adds up their install times.
The script prints `cmd` and `exit` lines for every command, one `requires` line before each domain update, optional `compat` lines for Codex cache links, and one `result` line at the end; `-n` prints the commands without running them.
The script prints `cmd` and `exit` lines for every command, one `requires` line before each domain update, optional `compat` lines for Codex cache links, `link` lines for the `current` link farm (see 4a), and one `result` line at the end; `-n` prints the commands without running them.
| Exit | Action |
| --- | --- |
@@ -61,17 +93,34 @@ Nothing passed in → run every step as written below.
| 2 | Usage error — the mode, the CLI name or the domain list is wrong. Report it as a defect in this skill, and do not retry with a guessed argument |
| other | Record that CLI as failed with the exit code and stderr |
On update, `tools/check-requires.sh {cli} {manifest}` checks each domain's `jsc.requires` before that domain is updated. Exit 0 updates the domain as usual. Exit 1 — a missing or too-old required jsc plugin — prints a `warn` line and the domain **is still updated**: skipping it would leave a behind domain permanently unable to reach the version its dependency needs. The block lives one layer up, at skill invocation time, where `jsc-hooks/hooks/version-guard.sh` stops that domain's skills. Exit 4 — the manifest is unreadable, is not valid JSON, or python3 is missing — prints a `note` line and also still updates the domain: no verdict is not the same fact as behind, so it gets its own line rather than a `warn` that would send the operator hunting for a version problem that is not there. Exit 2 or any other code — a `check-requires.sh` usage error or a broken script — prints a `skip` line and leaves that domain untouched, because a checker that failed outright is not a pass. Codex update preserves old `jsc-cli` and `jsc-hooks` cache version paths as symlinks to the newest installed version, so a still-running Codex deploy can keep using its helper scripts and a still-running Codex session whose hook_run_id points at the old cache can finish without `No such file`. Antigravity cannot install from a Gitea URL, so the script clones each domain into the local plugin directory (`JSC_LOCAL_PLUGINS`, default `$JSC_HOME/plugins`) and installs from that path — keep that clone, because update pulls the same one. That default deliberately avoids a development checkout: when the directory holds uncommitted changes or unpushed commits, the script prints a `skip` line, leaves the tree untouched, and installs the on-disk content.
On update, `{CLI_ROOT}/tools/check-requires.sh {cli} {manifest}` checks each domain's `jsc.requires` before that domain is updated. Exit 0 updates the domain as usual. Exit 1 — a missing or too-old required jsc plugin — prints a `warn` line and the domain **is still updated**: skipping it would leave a behind domain permanently unable to reach the version its dependency needs. The block lives one layer up, at skill invocation time, where `jsc-hooks/hooks/version-guard.sh` stops that domain's skills. **That last mention is a description of where the block happens, not a call this skill makes** — nothing here runs `version-guard.sh` except step 1.2, which is written as `{JSC_ROOT}/jsc-hooks/hooks/version-guard.sh`, so do not read it as a call site that was left un-prefixed. Exit 4 — the manifest is unreadable, is not valid JSON, or python3 is missing — prints a `note` line and also still updates the domain: no verdict is not the same fact as behind, so it gets its own line rather than a `warn` that would send the operator hunting for a version problem that is not there. Exit 2 or any other code — a `check-requires.sh` usage error or a broken script — prints a `skip` line and leaves that domain untouched, because a checker that failed outright is not a pass. Codex update preserves old `jsc-cli` and `jsc-hooks` cache version paths as symlinks to the newest installed version, so a still-running Codex deploy can keep using its helper scripts and a still-running Codex session whose hook_run_id points at the old cache can finish without `No such file`. Antigravity cannot install from a Gitea URL, so the script clones each domain into the local plugin directory (`JSC_LOCAL_PLUGINS`, default `$JSC_HOME/plugins`) and installs from that path — keep that clone, because update pulls the same one. That default deliberately avoids a development checkout: when the directory holds uncommitted changes or unpushed commits, the script prints a `skip` line, leaves the tree untouched, and installs the on-disk content.
Done when every detected CLI has reported an exit code and a `result` line, every skipped domain has a checker-failure reason or a local-tree reason, and every `warn` domain is named with the version it still has to catch up to.
**That one clone is shared by all five CLIs, and this step runs them in parallel, so the script takes a per-domain lock around its `git pull` and a `warn` line is what a contended or failed pull looks like.** Every CLI reaches that clone: the four that install from it, and `check-requires.sh`, which reads each domain's manifest there. Measured: two CLIs collided inside one round, one could not create `ORIG_HEAD.lock` and the other could not move its remote refs, and **both runs came back `fail` while every plugin had in fact installed** — a report saying failure over a machine that succeeded, for a reason that has nothing to do with deploying. So a pull that cannot get the lock within thirty seconds, or that exits non-zero on a clone already on disk, prints `warn` and **does not fail the run**: the content is there, it may simply be older than the remote. Read every such `warn` line into the report as "this CLI installed what was already on disk" — it is the one line between that and a silent install of a stale version. A `git clone` that fails is different and still fails the run: nothing is on disk to install.
**The `link` lines say where `{JSC_ROOT}` now points, and they are the reason step 0's literal paths keep working.** `{JSC_ROOT}` is a farm of version-free symbolic links, one per plugin, each pointing at that plugin's versioned directory in a CLI's plugin cache. Every literal absolute path this skill builds rests on it, so a link left on an old version silently runs an old script — and a script that old version never shipped is simply absent, which stops a heartbeat without printing anything. `deploy.sh` refreshes the whole farm at the end of its run: install and update repoint every link at the version just installed, uninstall clears the ones whose target is gone.
Only one CLI's run touches the farm. The machine has one farm and five CLIs hold five copies, so the script picks the first of claude, codex, copilot, kiro that is installed and lets only that run write; every other CLI prints one `link<TAB>-<TAB>skip` line naming the base CLI, which is how a deliberate skip is told apart from a farm nobody refreshed. Read the base CLI's sub agent output for the real result. Each line is `link<TAB>{domain}<TAB>{status}<TAB>{link path}<TAB>{target or reason}`.
| `link` status | What it means and what to do |
| --- | --- |
| `ok` | The link now points at column 5, the version directory this round installed. Carry the target into step 7 for at least the plugins this skill itself reaches through `{JSC_ROOT}` |
| `removed` | Uninstall left the target gone, so the dangling link was cleared. Nothing to do |
| `skip` with a domain in column 2 | Something that is not a symbolic link already sits at that path, and the script deliberately refused to overwrite it. **Report it as an operator action** — until it is cleared by hand, that plugin's documented path keeps resolving to whatever is sitting there |
| `skip` with `-` in column 2 | This run is not the base CLI, or no base CLI is installed at all. The second case means no documented path got refreshed this round, so say so |
| `fail` | The version directory could not be found, or the link could not be written. The deploy still stands, but that plugin's documented path may still be on an old version. Name the plugin and the reason, and judge the round `degraded` |
| `dryrun` | `-n` only: column 5 is where the link would point. Nothing was written |
A failed link never fails the deploy. By the time the farm is refreshed the plugins are installed and working, and marking the round failed would skip the restart gate and the `result` line too, handing the operator a fully deployed machine described as a failure. The stale link is a real defect, but its remedy is a line the operator can see and act on.
Done when every detected CLI has reported an exit code and a `result` line, every skipped domain has a checker-failure reason or a local-tree reason, every `warn` domain is named with either the version it still has to catch up to or the reason its local clone was not refreshed this round, and every `link` line has been read — with each `ok` target in hand for step 7 and each `fail` or domain-level `skip` named as an operator action.
5. After install or update, call `jsc-hooks:hooks-install` and **hand it the CLI list from step 1.1**, so it does not probe the same five executables a second time. `hooks-install` still detects for itself when it receives no list — that fallback is what keeps it usable on its own.
Take its aggregate result rather than re-reading each CLI's smoke detail; the installer already judged purge, wiring, smoke and scan per CLI, and refreshes `$JSC_HOME/current/jsc-hooks` on the way — the path `deploy.sh` follows to reach `restart-gate.sh`. Keep exactly one extra judgement here, because it is a deploy-side fact the installer does not rule on: **a smoke result containing `No such file` is a failed update**, since it means a rewritten hook path cannot execute. Report it and do not let the deploy finish as successful.
Take its aggregate result rather than re-reading each CLI's smoke detail; the installer already judged purge, wiring, smoke and scan per CLI, and refreshes `{JSC_ROOT}/jsc-hooks` on the way — the path `deploy.sh` follows to reach `restart-gate.sh`. Keep exactly one extra judgement here, because it is a deploy-side fact the installer does not rule on: **a smoke result containing `No such file` is a failed update**, since it means a rewritten hook path cannot execute. Report it and do not let the deploy finish as successful.
Done when hooks-install has returned an aggregate verdict for every CLI in the list, and every `No such file` in it is reported as an update failure.
6. After install or update, run `tools/write-guides.sh {mode} {domain}...` **once for the whole machine**, after every CLI in step 4 has finished. It rewrites `$JSC_HOME/update-guide.md` and `$JSC_HOME/remove-guide.md` from the live detection result, so the later update and removal runs have the real commands for this machine. Skip it for `uninstall`: the guides describe an installed skill set.
6. After install or update, run `{CLI_ROOT}/tools/write-guides.sh {mode} {domain}...` **once for the whole machine**, after every CLI in step 4 has finished. It rewrites `$JSC_HOME/update-guide.md` and `$JSC_HOME/remove-guide.md` from the live detection result, so the later update and removal runs have the real commands for this machine. Skip it for `uninstall`: the guides describe an installed skill set.
| Exit | Action |
| --- | --- |
@@ -82,8 +131,28 @@ Nothing passed in → run every step as written below.
Done when both `wrote` lines are printed, or the failure is reported with the exit code and the paths involved.
7. Report the run and close it, in one block. The result and any failure reason for every CLI × mode, plus every `skip` line, every `warn` line, every Codex `compat` line, and every CLI that could not be version-checked in step 2.
7. Report the run and close it, in one block. The result and any failure reason for every CLI × mode, plus every `skip` line, every `warn` line, every Codex `compat` line, every `link` line that is not a plain base-CLI skip, and every CLI that could not be version-checked in step 2.
For install or update, the same block ends with the restart instruction, in these words: 「請關閉目前的工作階段並重新啟動，新的技能內容才會載入」. `deploy.sh` recorded this round in `$JSC_HOME/restart-required.d/{cli}` — one file per CLI — and prints its path on a `restart` line; `jsc-hooks` reads only that CLI's own file and keeps reminding until that CLI restarts, with `JSC_RESTART_GATE=off` as the escape hatch. Restarting one CLI clears its own file and leaves the others' gates standing. Name the two guide paths from step 6 in that same closing block, so the operator knows where this machine's update and removal commands now live.
State the farm explicitly: name `{JSC_ROOT}` and, per plugin, the version directory its link now points at, taken from the `ok` targets. That one list is what lets the next round's operator check by eye that a documented path leads to the version just deployed, instead of finding out through a heartbeat that quietly stopped.
Done when every detected CLI appears in the report with its `result` status, and — for install or update — the restart instruction is printed with both guide paths named, or step 6's failure is repeated in their place.
For install or update, the same block ends with the restart instruction, in these words: 「請關閉目前的工作階段並重新啟動，新的技能內容才會載入」. `deploy.sh` recorded this round in `$JSC_HOME/restart-required.d/{cli}` — one file per CLI — and prints its path on a `restart` line. **It writes that file on every install and update, including a run it judged `fail`**: a partial failure means part of what is on disk changed, which makes a restart more necessary, not less. It used to write it only on success, and one round proved the cost — an unrelated `git pull` failure turned the run into a `fail`, that branch never reached the gate, and the one CLI running the freshly replaced code was the only one never told to restart. Report a `restart` line missing from a `fail` run as a defect in that script rather than raising the gate by hand; `jsc-hooks` reads only that CLI's own file and keeps reminding until that CLI restarts, with `JSC_RESTART_GATE=off` as the escape hatch. Restarting one CLI clears its own file and leaves the others' gates standing. Name the two guide paths from step 6 in that same closing block, so the operator knows where this machine's update and removal commands now live.
Done when every detected CLI appears in the report with its `result` status, every plugin's refreshed link target is named, and — for install or update — the restart instruction is printed with both guide paths named, or step 6's failure is repeated in their place.
8. **Record how the run ended.** This is the last thing this skill does, and it runs on every path out of the skill, the ones that stop at step 1 included. Call
`{JSC_ROOT}/jsc-hooks/tools/report-status.sh skill-end jsc-cli:deploy {status} {exit code} [detail]`
`{exit code}` is the exit code of whatever decided the outcome — the worst `deploy.sh` exit of the run, or the collector that stopped step 1 — and `0` when nothing failed. `{detail}` is one short line, no more than 200 characters: the mode and the per-CLI counts fit there, the `cmd` and `exit` lines do not. **If the script is not on this machine, skip this step in silence and finish the run as it stood** — missing infrastructure is not a failure, and a reporting call may never change what this skill returns or reports.
Record it after step 7's block, never instead of it. The event stream carries one status; the operator still needs the per-CLI report on screen.
| status | When this skill uses it |
| --- | --- |
| `ok` | Every detected CLI's `deploy.sh` exited 0, hooks-install returned a clean verdict for each of them with no `No such file` in any smoke result, both guides printed their `wrote` line, and the base CLI's `link` lines are all `ok` or `removed` |
| `blocked` | Nothing was deployed because there was nothing to deploy to: `detect-clis.sh` exited 0 with no row, so none of claude, codex, copilot, antigravity, kiro is installed and the run stops before any plugin command |
| `failed` | The run broke: the marketplace read in step 1.3 exited non-zero or returned no plugin entry, or every detected CLI's `deploy.sh` came back non-zero. Also used when a smoke result carries `No such file`, which this skill judges as a failed update even when hooks-install did not |
| `degraded` | The deploy landed on part of the machine only: some CLIs exited 0 while others failed, a domain was left untouched by a `skip` line from `check-requires.sh`, `write-guides.sh` exited 4 so the plugins are installed but this machine has no up-to-date update and remove guide, or a `link` line came back `fail` or skipped a plugin whose link path is not a symbolic link — the plugins are installed, but a documented path may still lead to the old version |
| `aborted` | The user chose none of `install`, `update` or `uninstall` at step 3, or stopped the run before step 4 launched the first CLI, so no plugin command ran |
Done when exactly one `skill-end` line was recorded for this run, or the script was absent and the run finished without it.
skills/doctor/SKILL.md
+121 -17
View File
@@ -1,13 +1,47 @@
---
name: doctor
description: Health-check the execution environment in one pass and record the result, changing nothing. Four checks - plugin versions from jsc-hooks/hooks/version-guard.sh report, hook wiring from jsc-hooks/tools/wire-cli.sh status, settings from tools/scan-config.sh against tools/config-spec.tsv, and orphan variables. Report one findings table per check, build the 待修項目 table with tools/build-todo.sh, then write the whole run to wiki CHECK_{HASH} where HASH comes from {hostname}/{user}; the page keeps only the latest run. Use after installing or updating the skill set, when a skill fails on a settings or wiring problem, or before handing a machine over; not for applying fixes, which is jsc-cli:setup.
description: Health-check the execution environment in one pass and record the result, changing nothing. Four checks - plugin versions from jsc-hooks/hooks/version-guard.sh report, hook wiring from jsc-hooks/tools/wire-cli.sh status, settings from tools/scan-config.sh against tools/config-spec.tsv, and orphan variables. Report one findings table per check, build the 待修項目 table with tools/build-todo.sh, then write the whole run to wiki CHECK_{HASH} where HASH comes from {short hostname}/{user}; the page keeps only the latest run, while its CHECK_CONTENTS block is upserted through wiki-contents.sh into the contents repo. Use after installing or updating the skill set, when a skill fails on a settings or wiring problem, or before handing a machine over; not for applying fixes, which is jsc-cli:setup.
---
# doctor — execution environment health check
Read-only. Every command below either reads a file or asks Gitea; none of them writes a setting. That is the contract with `jsc-cli:setup`: doctor states the facts, setup changes things.
The read-only contract is enforced in code, not by prose: every `jsc-hooks/tools/wire-cli.sh` call in this skill runs with `JSC_READONLY=1` in its environment. A mistyped subcommand then refuses to write instead of rewiring the machine that was only supposed to be measured.
The read-only contract is enforced in code, not by prose: every `{JSC_ROOT}/jsc-hooks/tools/wire-cli.sh` call in this skill runs with `JSC_READONLY=1` in its environment. A mistyped subcommand then refuses to write instead of rewiring the machine that was only supposed to be measured.
## Path rule — every script call is a literal absolute path
Write every script call in this skill as a literal absolute path. Never hand the shell a path that still holds a variable or a tilde — `$JSC_HOME/...`, `~/.jsc/...`, or anything like them — and never a bare relative one either. The permission layer matches paths statically: it expands no variable and no tilde, so such a path matches no allow rule and the call falls through to an approval prompt. A bare relative path is the worse form, because it resolves against whatever directory the CLI happens to be in — and this skill deliberately runs from the operator's project directory, which is never a plugin root — so every bare call site is a place where a prefix gets guessed.
A guessed prefix is the specific danger here, and it is quiet. This skill measures a machine and writes what it found onto a page other people act on, so a call that reaches the wrong copy of a script does not fail — it reports. An old `tools/scan-config.sh` reached through some other version's directory grades this machine against that version's `tools/config-spec.tsv`, and every row it prints looks exactly like a real finding. A checkup that cannot say which script produced its rows is worth less than no checkup, because nobody doubts it.
Portability is no reason to put the variable back. Step 0 resolves the roots once, at run time, on whatever machine this runs on — that is where portability comes from.
## Step 0 — resolve the two roots, once
Before step 1, run this one command:
`readlink -f "$JSC_HOME/current"`
It prints one absolute directory: the `current` directory itself, a farm of version-free symbolic links with one entry per plugin. Call it `{JSC_ROOT}` for the rest of this document. **Stop at that directory — never resolve one level further.** Resolving one of those entries lands on the versioned plugin cache (`/root/.claude/plugins/cache/jsc/jsc-hooks/0.4.2`, say), and a versioned path is exactly the kind no allow rule can hold: a rule with `*` where the version segment goes matches nothing, measured. `{JSC_ROOT}` is the version-free root, and staying at it is the whole point. This is the only place a variable may appear; the shell expands it inside the command itself, so no unexpanded path ever reaches the permission layer.
Confirm the directory exists, in the same approved step: `[ -d "{the path just printed}" ]`. **That is the check the other two wave through.** `JSC_HOME` is an optional variable with a default, so an unset one is an ordinary machine state rather than an exotic one — it has its own row in `{CLI_ROOT}/tools/config-spec.tsv`, and finding it unset is a normal outcome of this very checkup. With it unset the command prints `/current` and exits 0 — non-empty, absolute, and nowhere — and every literal path built from it then names a place that is not there.
**Unlike `jsc-cli:deploy`, an unresolvable `{JSC_ROOT}` never stops this run.** It is a finding, and findings are what this skill produces: report `JSC_HOME` in the 全域設定 block and at the top of the 待修項目 table. Mark every check that needed a cross-plugin script as 無法驗證 with that as its reason, and let the checks that do not need one finish. A read-only checkup that aborts tells the operator nothing, and the machine most in need of a checkup is the machine whose paths do not resolve.
Substitute `{JSC_ROOT}` in every cross-plugin call, so what runs is a literal absolute path. `{JSC_ROOT}/jsc-hooks/hooks/version-guard.sh` becomes, for example, `/root/.jsc/current/jsc-hooks/hooks/version-guard.sh`. Resolve it once, at the start of the run. Do not re-resolve it per call, and do not add a tool that prints it.
### The second root — this skill's own `tools/` and `templates/`
**Do not reach this skill's own files through `{JSC_ROOT}`, even though a `jsc-cli` link is normally sitting there.** `jsc-cli:deploy` refreshes the whole farm at the end of every round, so on a machine that has deployed, that link exists. This skill still may not lean on it, for a reason of its own: **doctor is the skill that gets run when the machine is suspect.** Its own triggers say so — after an install or update, after a skill failed on a settings or wiring problem, before a handover. A deploy that ended `degraded` is one whose `link` lines came back `fail`, or `skip` on a path holding something that is not a symbolic link at all, leaving a plugin's documented path on an old version — and doctor is what runs next. Reached through that link, `tools/scan-config.sh` and its `tools/config-spec.tsv` come from a version this machine is not running, and the findings table then describes a machine that does not exist. Nothing errors: an old scanner runs perfectly well.
The second root is free of that. `tools/scan-config.sh`, `tools/detect-clis.sh` and `tools/build-todo.sh` reached through it keep working on a machine whose farm is stale, broken, or never built — which is precisely the machine being measured — so `JSC_HOME` itself still gets scanned and still gets reported.
They sit at `{plugin root}/tools/` and `{plugin root}/templates/`, and the plugin root is the base directory the CLI states when it loads this skill. Take that literal path verbatim, call it `{CLI_ROOT}`, and write every own-plugin path as `{CLI_ROOT}/tools/{script}` or `{CLI_ROOT}/templates/{file}` — `/root/.claude/plugins/cache/jsc/jsc-cli/0.3.3/tools/scan-config.sh`, for example. No command runs for this one, and it is taken once, like `{JSC_ROOT}`.
That base directory carries a version segment, so no allow rule covers it and each of those calls raises an approval prompt. **That is acceptable in this skill and in no unattended one**: doctor runs with the operator in front of it — the five blocks and the 待修項目 table are printed for a person to read and act on — so there is somebody to approve. Never carry this branch into a skill that runs from a scheduler, and never guess a prefix when the invocation states no base directory: report that this skill's own plugin root is unknown and stop, because a guessed prefix reports some other version's idea of this machine as if it were this machine.
Done when `{JSC_ROOT}` holds one existing absolute directory or its failure is recorded as a `JSC_HOME` finding, and `{CLI_ROOT}` holds one literal absolute path.
## 1. Collect, four checks in parallel
@@ -17,7 +51,7 @@ Done when all four sub agents have returned, and each has either its raw lines o
### 1.1 Skill versions
Run `jsc-hooks/hooks/version-guard.sh report`. It prints `{domain}<TAB>{本機}<TAB>{遠端}<TAB>{落後|最新|超前|查詢失敗}` per plugin, then `behind<TAB>{count}`.
Run `{JSC_ROOT}/jsc-hooks/hooks/version-guard.sh report`. It prints `{domain}<TAB>{本機}<TAB>{遠端}<TAB>{落後|最新|超前|查詢失敗}` per plugin, then `behind<TAB>{count}`.
A report with no `{domain}` row, or one carrying `noregistry<TAB>{path}`, means this CLI has no local plugin registry. Report it as 無法驗證 — never as 最新. `behind<TAB>0` proves nothing when no domain row precedes it.
@@ -27,9 +61,9 @@ Done when every installed domain has a status literal, or the check is reported
### 1.2 Hook wiring
First run `jsc-cli/tools/detect-clis.sh`. Exit 0 with at least one row → those are the CLIs to check. Exit 0 with no row → report the wiring table as empty and say no AI agent CLI was detected; that is a finding, not a pass. Any non-zero exit → report the wiring check as 無法驗證 with the exit code and stderr.
First run `{CLI_ROOT}/tools/detect-clis.sh`. Exit 0 with at least one row → those are the CLIs to check. Exit 0 with no row → report the wiring table as empty and say no AI agent CLI was detected; that is a finding, not a pass. Any non-zero exit → report the wiring check as 無法驗證 with the exit code and stderr.
Then run `JSC_READONLY=1 jsc-hooks/tools/wire-cli.sh status {cli}` for every detected CLI. Use `status` and nothing else: `wire-cli.sh` without a subcommand rewires, `purge` deletes, and `smoke` executes hooks — all three break the read-only contract.
Then run `JSC_READONLY=1 {JSC_ROOT}/jsc-hooks/tools/wire-cli.sh status {cli}` for every detected CLI. Use `status` and nothing else: `wire-cli.sh` without a subcommand rewires, `purge` deletes, and `smoke` executes hooks — all three break the read-only contract.
| Exit | Meaning | Action |
| --- | --- | --- |
@@ -46,7 +80,7 @@ Done when every detected CLI carries one of those verdicts and its missing items
### 1.3 Settings, global and project in one scan
Run `jsc-cli/tools/scan-config.sh scan all` from the current working directory. One call covers both scopes: the spec table is read once and the `scope` column separates the rows. It prints `{項目}<TAB>{範圍}<TAB>{必要}<TAB>{現況}<TAB>{說明}<TAB>{修法}<TAB>{判定}`, closing with `summary<TAB>{missing}<TAB>{invalid}<TAB>{unset}<TAB>{skipped}`.
Run `{CLI_ROOT}/tools/scan-config.sh scan all` from the current working directory. One call covers both scopes: the spec table is read once and the `scope` column separates the rows. It prints `{項目}<TAB>{範圍}<TAB>{必要}<TAB>{現況}<TAB>{說明}<TAB>{修法}<TAB>{判定}`, closing with `summary<TAB>{missing}<TAB>{invalid}<TAB>{unset}<TAB>{skipped}`.
| Exit | Action |
| --- | --- |
@@ -63,7 +97,7 @@ Done when the summary line is read, the rows are split into the two scopes, and
### 1.4 Orphan variables
Run `jsc-cli/tools/scan-config.sh orphans` — variables used in the source but absent from the spec table. Same exit codes as 1.3; exit 3 here also covers a missing plugins root, so name `JSC_PLUGINS_ROOT` in that case.
Run `{CLI_ROOT}/tools/scan-config.sh orphans` — variables used in the source but absent from the spec table. Same exit codes as 1.3; exit 3 here also covers a missing plugins root, so name `JSC_PLUGINS_ROOT` in that case.
They are a maintenance note for the skill set, not a fault on this machine, so they never enter the 待修項目 table.
@@ -73,7 +107,7 @@ Done when the orphan list is returned or the check is reported as skipped with i
### 2.1 The five blocks
Report all five blocks per `templates/check-page.md`: 技能版本 from 1.1, Hook 接線 from 1.2, 全域設定 and 自我設定 from 1.3's two scopes, and 未登錄變數 from 1.4. Step 1.4 is the only place the orphan list is collected, so leaving it out here drops it from the run entirely — it never enters the 待修項目 table of 2.2, which is exactly why it needs its own block.
Report all five blocks per `{CLI_ROOT}/templates/check-page.md`: 技能版本 from 1.1, Hook 接線 from 1.2, 全域設定 and 自我設定 from 1.3's two scopes, and 未登錄變數 from 1.4. Step 1.4 is the only place the orphan list is collected, so leaving it out here drops it from the run entirely — it never enters the 待修項目 table of 2.2, which is exactly why it needs its own block.
The project rows carry the working directory in their heading. A project-scope result is meaningless without it, because the answer changes with every `cd` — so name the directory that step 1.3 scanned, even when the project table is empty.
@@ -83,7 +117,7 @@ When `.env` or `.envrc` exists in that directory, name the spec-table variables
Save each collector's raw lines to a file, then run
`jsc-cli/tools/build-todo.sh --config {scan-all 輸出} --wiring {cli}={status 輸出} --version {report 輸出}`
`{CLI_ROOT}/tools/build-todo.sh --config {scan-all 輸出} --wiring {cli}={status 輸出} --version {report 輸出}`
one `--wiring` per detected CLI. The script merges the three sources and orders them `missing` → `invalid` → `unwired` → `落後`. Nothing wrong → it prints one row whose class reads 無.
@@ -102,18 +136,88 @@ Done when all five blocks of 2.1 are on screen with the scanned directory stated
### 3.1 Record
Write the page through `jsc-gitea:wiki`:
This run writes two pages, and they live in **two different wiki repos**. Resolve each repo on its own and never reuse one for the other.
- Wiki repo: `jsc-gitea/tools/gitea.sh wiki-repo CHECK`.
- Page name: `CHECK_` plus `gitea.sh hash-id "{hostname}/{user}"` — the host and the login account, not `{owner}/{repo}`. Doctor checks a machine, and it has to work in directories that are not repositories at all.
- `CHECK_{HASH}` is a **content page**: overwrite the whole page, because this page type keeps only the latest run of this one machine.
- `CHECK_CONTENTS` is a **contents page** and follows the opposite rule: read it back first, then upsert this machine's row from `templates/check-contents.md` in the same pass — add the row if missing, otherwise refresh its 必要項缺漏、設定錯誤、最後體檢 columns. Never overwrite the whole page, and never touch a row belonging to another machine: those rows are other people's records, and this run never read them from anywhere else.
- The `CHECK_CONTENTS` read branches by exit code, and only exit 4 opens the create path. Exit 0 means the page is there, so upsert into what came back. Exit 4 means the page really does not exist yet, so build it from the template. Exit 7 (key invalid or no permission) and exit 8 (any other API failure) both mean the old rows are unknown, never that the page is missing: skip the `CHECK_CONTENTS` write, name the exit code in the report, and create nothing. Writing a fresh template over a directory whose rows were never read wipes every other machine's row, and the write carries no merge and no backup.
**The HASH — take it from the machine, do not compose it by hand.** `jsc-assist`'s `MONITOR_{HASH}` claims the same hash source and takes it in code, so the two pages only ever line up when this skill takes it the same way:
`wiki-repo` exiting 3 means no wiki repo is configured for CHECK. Print the tables, skip the wiki write, and put `JSC_WIKI_REPO_CHECK` at the top of 待修項目 — that unset variable is itself a finding, so a failed write never fails the health check. Any other non-zero exit from `wiki-repo`, `hash-id` or the wiki write is reported the same way: tables on screen, write skipped, exit code named.
```sh
host=$(hostname 2>/dev/null || uname -n 2>/dev/null || printf 'unknown'); host=${host%%.*}
user=${USER:-$(id -un 2>/dev/null || printf 'unknown')}
```
`${host%%.*}` is the point of the snippet: `hostname` prints the FQDN on some machines and the short name on others, so a hand-written value gives one machine two pages that never merge again. Pass `"{host}/{user}"` to `{JSC_ROOT}/jsc-gitea/tools/gitea.sh hash-id` and use exactly what it prints. It is the host and the login account, not `{owner}/{repo}` — doctor checks a machine, and it has to work in directories that are not repositories at all.
**`CHECK_{HASH}` — content page.** Repo: `{JSC_ROOT}/jsc-gitea/tools/gitea.sh wiki-repo CHECK`. Write it through `jsc-gitea:wiki` and overwrite the whole page, because this page type keeps only the latest run of this one machine. Whole-page overwrite is correct here and forbidden on the page below.
**`CHECK_CONTENTS` — contents page, in the contents repo.** Repo: `{JSC_ROOT}/jsc-gitea/tools/gitea.sh wiki-repo CONTENTS`. Every contents page lives there now; it never falls back to `JSC_WIKI_REPO_CHECK`. It is an H1, a `>` preamble and one H2 block per machine — no markdown table anywhere on it. Do not hand-edit it — write the block with
`{JSC_ROOT}/jsc-gitea/tools/wiki-contents.sh upsert CHECK 1 "CHECK_{HASH}" {block file} {CLI_ROOT}/templates/check-contents.md`
The block file holds the whole H2 block: the `## CHECK_{HASH}` line, a blank line, then one bullet per field in the order `{CLI_ROOT}/templates/check-contents.md` lists them, written as `- {欄位名}：{值}` with a full-width colon. Every field of the template gets a bullet, `HASH` included — the heading is the key, and a field only in the heading is a field the next reader cannot read.
The script reads the page back, replaces this machine's block or appends it, then writes the whole page. That keeps the rule in one place: one block per run, never a whole-page overwrite, never another machine's block — those blocks are other people's records, and this run read them from nowhere else.
**The key is the H2 heading, `CHECK_{HASH}`.** It is the name of the content page this block points at: the literal `CHECK_` plus exactly what `hash-id` printed — 40 uppercase hex characters, not shortened, not otherwise prefixed, not wrapped in a link, no date appended. The script compares the whole heading text after trimming, so the key and that heading must match character for character.
The page name is the key because it is the one value that does not move. It is decided by `{host}/{user}` alone, so it survives a changed `GITEA_HOST`, a `JSC_WIKI_REPO_CHECK` moved to another repo, and a different Gitea encoding of the page name — all of which change the URL. Key on anything holding a URL and the comparison never matches, so every run appends a second block for the same machine — silently, because the page still looks right.
`1` is `<key-col>`, and it only matters while a page is still the old markdown table: it is the 1-based index of the column that held the identity, the `[CHECK_{HASH}]({URL})` cell in column 1, whose text becomes the H2 heading when the script converts that table to blocks. On a page already in block form the script ignores it.
The `體檢頁` bullet stays the human-facing link and is never the key; the heading itself carries no link and no URL. Write the bullet as `[CHECK_{HASH}]({absolute URL})` — text plus link, the one link form this skill set uses. The URL comes from `{JSC_ROOT}/jsc-gitea/tools/gitea.sh wiki-url {CHECK repo} CHECK_{HASH}` and is never composed by hand. Fetch it after `CHECK_{HASH}` is written: `wiki-url` exits 4 on a page that does not exist yet.
A same-wiki link form resolves only inside its own wiki, and the two pages are no longer in the same one. It fails without an error, reading on screen as plain text or a dead link, so nobody finds it and nobody fixes it.
**Verify every link before writing it.** Collect every URL heading into `CHECK_{HASH}` or into the `CHECK_CONTENTS` block, then hand the whole list to `{JSC_ROOT}/jsc-gitea/tools/link-check.sh`. It prints `{OK|DEAD|SKIP}<TAB>{URL}<TAB>{reason}` per line. Only exit 0 may be written.
| Exit | Meaning | Action |
| --- | --- | --- |
| 0 | Every link answers | Write the page |
| 1 | At least one link is unreachable | Write nothing, on either page. Report the `DEAD` lines to the caller |
| 2 | Usage error: no URL was given | Report it as a defect in this skill and pass the URLs |
| 3 | The list holds a Gitea URL but `GITEA_HOST` is unset | Skip both writes and put `GITEA_HOST` at the top of 待修項目. Never write without verifying |
| 7 | Gitea authentication failed | Stop and report the key problem. This is not a dead link |
Exit 7 stays apart from exit 1 on purpose: with a dead key, Gitea's answer for a private repo looks the same as "page absent". Merge the two and one expired key marks every live page dead, and the blocks pointing at them get rewritten or dropped.
The script asks the API and never reads a web status code. A private repo's web URL answers 404 to a request with no credentials, so status codes turn good links into dead ones.
| Exit | Meaning | Action |
| --- | --- | --- |
| 0 | `updated` or `added` | Report which one it printed, with the repo and page it named |
| 1 | The page content could not be built, or the write failed | Nothing landed. Report it with the stderr, and keep 2.1's tables on screen. A page with no block to replace is not this case: the script appends instead |
| 2 | Usage error | Report it as a defect in this skill. Do not retry with guessed arguments. A `{CLI_ROOT}/templates/check-contents.md` that is not on disk also lands here — then name the path the script looked for, confirm the plugin install is complete, and rerun |
| 3 | No contents repo configured | Skip this write and put `JSC_WIKI_REPO_CONTENTS` at the top of 待修項目 |
| 4 | Page absent and no template given | Unreachable the way this skill calls the script — the command above always passes `{CLI_ROOT}/templates/check-contents.md`. A template that is not on disk comes back as exit 2, not 4. So treat a 4 as a malformed call: report it as a defect in this skill, name the command that produced it, and do not retry with guessed arguments |
| 7 | Key invalid or no permission | Nothing was read and nothing written. Name the exit code and create nothing |
| 8 | Any other API failure | Same as 7: the old blocks are unknown, so name the exit code and create nothing |
Exits 7 and 8 never mean the page is missing. Writing a fresh template over a directory whose blocks were never read wipes every other machine's block, with no merge and no backup behind it — which is exactly why the script creates a page only when its own read reported that page absent, and why it owns that branch instead of this prose.
`wiki-repo` exiting 3 means that page type has no wiki repo configured: `JSC_WIKI_REPO_CHECK` for the content page, `JSC_WIKI_REPO_CONTENTS` for the contents page. Print the tables, skip that one write, and put the unset variable at the top of 待修項目 — it is itself a finding, so a failed write never fails the health check. Any other non-zero exit from `wiki-repo`, `wiki-url`, `hash-id` or the wiki write is reported the same way: tables on screen, write skipped, exit code named.
### 3.2 Hand off
State the four counts from 2.2's `summary` line: required items missing, settings invalid, CLIs unwired, domains behind. Recommend `/jsc-cli:setup` when any of those is above zero. Never fix anything here.
Done when either the wiki page URL is reported or the skipped write is reported together with its reason, **and** the four counts are stated with the recommendation given or explicitly withheld.
Done when every link written into either page passed `link-check.sh` first — or the `DEAD` list is on screen and that write was skipped — each of the two pages is reported with its URL, or its skipped write is reported together with its reason, **and** the four counts are stated with the recommendation given or explicitly withheld.
### 3.3 Record how the run ended
This is the last thing this skill does, and it runs on every path out of the skill. Call
`{JSC_ROOT}/jsc-hooks/tools/report-status.sh skill-end jsc-cli:doctor {status} {exit code} [detail]`
`{exit code}` is the exit code of whatever decided the outcome, and `0` when nothing failed. `{detail}` is one short line, no more than 200 characters: the four counts fit there, the five blocks do not. **If the script is not on this machine, skip this step in silence and finish the run as it stood** — missing infrastructure is not a failure, and a reporting call may never change what this skill returns or reports.
This one call writes, and it is the only write this skill makes. It records what the run found; it changes no setting, no wiring and no version, so the read-only contract of the opening paragraph still holds.
| status | When this skill uses it |
| --- | --- |
| `ok` | All four checks reached a conclusion, the five blocks and the 待修項目 table are on screen, and both pages were written |
| `degraded` | The checkup ran but part of it has no conclusion, and this is the common outcome for a read-only skill that cannot reach a source. Any check reported as 無法驗證 lands here — `version-guard.sh report` exiting non-zero, `scan-config.sh` exiting 3 on a missing spec table, `wire-cli.sh status` returning an unexpected code, a Gitea-dependent row coming back `skipped` in offline mode — and so does a `wiki-repo` exit 3 that skipped a page write, which this skill treats as a finding rather than a fault |
| `failed` | Reading the machine worked, then recording it broke on an error: `link-check.sh`, `gitea.sh` or `wiki-contents.sh` returned 7 on an invalid key, or 8 on any other API failure. Both are errors, never an absent page, and neither leaves a usable record |
| `aborted` | The user stopped the run before the record was written, for example by declining to supply `GITEA_HOST` and asking to end the checkup there |
`blocked` has no place in this skill. Nothing gates a read-only checkup: a machine with no CLI installed, no plugin registry and no wiki repo still produces four findings, and reporting that as `blocked` would hide a run that did its whole job.
Done when exactly one `skill-end` line was recorded for this run, or the script was absent and the run finished without it.
skills/models/SKILL.md
+17 -1
View File
@@ -1,6 +1,6 @@
---
name: models
description: List every model usable by each installed AI CLI (claude, codex, copilot, antigravity, kiro) and attach capability tags from references/model-tags.md. Syncs the tag table to $JSC_HOME/model-tags.tsv via tools/model-tags.sh, so jsc-sdlc gates are enforced in code. States each SDLC stage's required tags: plan and analyze need reasoning-max, implement needs coding, maintain any. Resolves each stage's preferred model chain via tools/model-config.sh (project .jsc/models overrides $JSC_HOME/models.conf), for switch suggestions only. Use when checking model fitness, inventorying models, or reviewing stage gating; not for switching models or editing the config files.
description: 'List every model usable by each installed AI CLI (claude, codex, copilot, antigravity, kiro) and attach capability tags from references/model-tags.md. Syncs the tag table to $JSC_HOME/model-tags.tsv via tools/model-tags.sh, so jsc-sdlc gates are enforced in code. States each SDLC stage''s required tags: plan and analyze need reasoning-max, implement needs coding, maintain any. Resolves each stage''s preferred model chain via tools/model-config.sh (project .jsc/models overrides $JSC_HOME/models.conf), for switch suggestions only. Use when checking model fitness, inventorying models, or reviewing stage gating; not for switching models or editing the config files.'
---
# models — list CLI models with capability tags
@@ -38,3 +38,19 @@ description: List every model usable by each installed AI CLI (claude, codex, co
2. A 「階段偏好模型」 table right after it, built from collector 1.3's rows, with three columns: stage, chain, source (`project` / `global`). State below the table that the chain does **not** grant passage: it only names the model to suggest switching to when the gate blocks, and expresses preference among models that already satisfy the required tags.
Done when the requirement table shows all four stages with their required tags, and the 階段偏好模型 table shows the same four stages with `-` for unconfigured ones.
7. **Record how the run ended.** This is the last thing this skill does, and it runs on every path out of the skill, the ones that stop at step 1 included. Call
`jsc-hooks/tools/report-status.sh skill-end jsc-cli:models {status} {exit code} [detail]`
`{exit code}` is the exit code of whatever decided the outcome — usually `model-tags.sh sync` — and `0` when nothing failed. `{detail}` is one short line, no more than 200 characters: the CLI and model counts fit there, the four-column table does not. **If the script is not on this machine, skip this step in silence and finish the run as it stood** — missing infrastructure is not a failure, and a reporting call may never change what this skill returns or reports.
| status | When this skill uses it |
| --- | --- |
| `ok` | Every detected CLI has a model list, every model carries a tag, `sync` exited 0 and printed the path, and both stage tables are on screen |
| `blocked` | Nothing could be inventoried because nothing is installed: `detect-clis.sh` exited 0 with no row, so steps 2 to 4 have no CLI to work on. The tag table and the stage requirements are still printed, so say in `{detail}` that the inventory half of the run never started |
| `failed` | `model-tags.sh sync` returned 1, 2 or any other non-zero code, so `$JSC_HOME/model-tags.tsv` was not written and the SDLC gate stays broken until it is. `detect-clis.sh` exiting non-zero sits here too |
| `degraded` | The tag table was written but the picture is incomplete: `list-models.sh` stayed silent for a CLI so step 2 fell back to 「預設推定」 defaults, a model is missing from `references/model-tags.md` and was queued as a `jsc-ask:ask` question instead of tagged, or `model-config.sh list` failed so the 階段偏好模型 table shows 未取得 |
| `aborted` | The user stopped the run before `sync` wrote the file, so the gate reads whatever the previous run left behind |
Done when exactly one `skill-end` line was recorded for this run, or the script was absent and the run finished without it.
skills/setup/SKILL.md
+119 -14
View File
@@ -7,19 +7,62 @@ description: Fix what jsc-cli:doctor found, one confirmed item at a time. Read t
This skill writes. Every write is confirmed first, backed up, and verified afterwards.
## Path rule — every script call is a literal absolute path
Write every script call in this skill as a literal absolute path. Never hand the shell a path that still holds a variable or a tilde — `$JSC_HOME/...`, `~/.jsc/...`, or anything like them — and never a bare relative one either. The permission layer matches paths statically: it expands no variable and no tilde, so such a path matches no allow rule and the call falls through to an approval prompt. A bare relative path is the worse form, because it resolves against whatever directory the CLI happens to be in — the operator's project directory, which is never a plugin root — so every bare call site is a place where a prefix gets guessed.
**This skill writes, and that is what turns a guessed prefix from an annoyance into a wrong machine.** An old `tools/apply-config.sh` reached through some other version's directory rewrites the `# jsc-config` block of every rc file here to that version's idea of the key set, backs the old block up as though that were correct, and then step 4 re-verifies it with `show` from the same wrong copy — which agrees, because it is the same copy. The run reports that item as 已修 and the operator believes it, and nothing in this document catches it. Only the path does.
Portability is no reason to put the variable back. Step 0 resolves the roots once, at run time, on whatever machine this runs on — that is where portability comes from.
## Step 0 — resolve the two roots, once
Before step 1, run this one command:
`readlink -f "$JSC_HOME/current"`
It prints one absolute directory: the `current` directory itself, a farm of version-free symbolic links with one entry per plugin. Call it `{JSC_ROOT}` for the rest of this document. **Stop at that directory — never resolve one level further.** Resolving one of those entries lands on the versioned plugin cache (`/root/.claude/plugins/cache/jsc/jsc-gitea/0.4.2`, say), and a versioned path is exactly the kind no allow rule can hold: a rule with `*` where the version segment goes matches nothing, measured. `{JSC_ROOT}` is the version-free root, and staying at it is the whole point. This is the only place a variable may appear; the shell expands it inside the command itself, so no unexpanded path ever reaches the permission layer.
Confirm the directory exists, in the same approved step: `[ -d "{the path just printed}" ]`. **No skill meets an unset `JSC_HOME` as often as this one.** It is an optional variable with a default, it has an `auto` row in `{CLI_ROOT}/tools/config-spec.tsv`, and writing it is one of the repairs this skill performs — so a machine whose `JSC_HOME` is unset or wrong is the ordinary reason this skill was called. With it unset the command prints `/current` and exits 0: non-empty, absolute, and nowhere. The emptiness check and the exit code both wave that through, and every literal path built from it names a place that is not there.
**An unresolvable `{JSC_ROOT}` stops neither this run nor the repair.** Everything needed to write `JSC_HOME` — `{CLI_ROOT}/tools/apply-config.sh`, `{CLI_ROOT}/tools/scan-config.sh`, `{CLI_ROOT}/tools/build-todo.sh` — hangs off the second root below, which does not depend on `JSC_HOME` at all. Fix that item first, re-resolve `{JSC_ROOT}` once afterwards, and carry on; whatever is still unreachable — the wiki record of step 5, a delegated repair — is recorded as 未修好 with that as its reason, never guessed at.
Substitute `{JSC_ROOT}` in every cross-plugin call, so what runs is a literal absolute path. `{JSC_ROOT}/jsc-gitea/tools/gitea.sh` becomes, for example, `/root/.jsc/current/jsc-gitea/tools/gitea.sh`. Resolve it once. Do not re-resolve it per call, and do not add a tool that prints it.
### The second root — this skill's own `tools/` and `templates/`
**Do not reach this skill's own files through `{JSC_ROOT}`, even though a `jsc-cli` link is normally sitting there.** `jsc-cli:deploy` refreshes the whole farm at the end of every round, so on a machine that has deployed, that link exists. This skill still may not lean on it, for two reasons of its own.
The first is the paragraph above: the machine this skill is called to repair is often the machine whose `JSC_HOME` is unset or wrong, and on it the farm cannot be reached while the repair itself must still run. Route the repair tools through the farm and the one fault they exist to fix becomes the fault that stops them.
The second is what this skill does with a stale script. Step 3 hands every 落後 domain to `jsc-cli:deploy` precisely because the versions on this machine may be behind — and the farm is governed by that same staleness. Reaching `apply-config.sh` through it would repair the machine with the very tools that machine has just been judged to have outgrown, and the result is written into rc files rather than merely printed.
They sit at `{plugin root}/tools/` and `{plugin root}/templates/`, and the plugin root is the base directory the CLI states when it loads this skill. Take that literal path verbatim, call it `{CLI_ROOT}`, and write every own-plugin path as `{CLI_ROOT}/tools/{script}` or `{CLI_ROOT}/templates/{file}` — `/root/.claude/plugins/cache/jsc/jsc-cli/0.3.3/tools/apply-config.sh`, for example. No command runs for this one, and it is taken once, like `{JSC_ROOT}`.
That base directory carries a version segment, so no allow rule covers it and each of those calls raises an approval prompt. **That is acceptable in this skill and in no unattended one**: setup runs with the operator in front of it — step 2 puts every item to them one at a time, and nothing is written before they answer — so there is somebody to approve. Never carry this branch into a skill that runs from a scheduler, and never guess a prefix when the invocation states no base directory: report that this skill's own plugin root is unknown and stop **before writing anything**, because a guessed prefix writes this machine with some other version's script.
Done when `{JSC_ROOT}` holds one existing absolute directory or its failure is recorded as the `JSC_HOME` item, and `{CLI_ROOT}` holds one literal absolute path.
## 1. Get the work list
Read the 待修項目 table from wiki `CHECK_{HASH}` — repo from `jsc-gitea/tools/gitea.sh wiki-repo CHECK`, page name from `gitea.sh hash-id "{hostname}/{user}"`.
Read the 待修項目 table from wiki `CHECK_{HASH}` — repo from `{JSC_ROOT}/jsc-gitea/tools/gitea.sh wiki-repo CHECK`, page name from `{JSC_ROOT}/jsc-gitea/tools/gitea.sh hash-id "{host}/{user}"`, where `host` is the **short hostname** and `user` the login account, both taken from the machine:
```sh
host=$(hostname 2>/dev/null || uname -n 2>/dev/null || printf 'unknown'); host=${host%%.*}
user=${USER:-$(id -un 2>/dev/null || printf 'unknown')}
```
`${host%%.*}` matters: `hostname` prints the FQDN on some machines, and a hash built on the long name reads a page doctor never wrote. This is the same value doctor hashes, so it must be taken the same way.
No page, or `wiki-repo` exits 3, or any other non-zero exit from `wiki-repo`, `hash-id` or the wiki read → rebuild the list here. Rebuilding **MUST run as a sub agent**, and its three checkers **start together**: they read different files and share no state, so serialising them only triples the wait.
| Checker | Command | Exit branching |
| --- | --- | --- |
| Settings | `tools/scan-config.sh scan all` | 0 → use the rows; 2 → usage error, report it as a defect in this skill; 3 → spec table missing, name the path and `JSC_CONFIG_SPEC`; other → report settings as 無法驗證 |
| Wiring | `tools/detect-clis.sh`, then `JSC_READONLY=1 jsc-hooks/tools/wire-cli.sh status {cli}` per detected CLI — this step only takes stock, and `wire-cli.sh` without a subcommand rewires, so the read-only contract is carried in the environment rather than trusted to a correctly typed subcommand | detect-clis exit 0 with no row → no CLI to wire, say so and skip; detect-clis non-zero → report wiring as 無法驗證 with the exit code. Per CLI: 0 wired and 1 degraded → nothing to fix; 2 → usage error, defect in this skill; 3 → CLI not installed, drop it; 5 → collect its `missing` items; 6 → readonly refused the call, which means the subcommand was mistyped into a writing one — nothing on the machine changed; fix the command and rerun that CLI; other → report that CLI as 無法驗證 |
| Versions | `jsc-hooks/hooks/version-guard.sh report` | 0 → use the rows, and treat a report with no `{domain}` row or a `noregistry` line as 無法驗證 — other exits → report versions as 無法驗證 with the exit code |
| Settings | `{CLI_ROOT}/tools/scan-config.sh scan all` | 0 → use the rows; 2 → usage error, report it as a defect in this skill; 3 → spec table missing, name the path and `JSC_CONFIG_SPEC`; other → report settings as 無法驗證 |
| Wiring | `{CLI_ROOT}/tools/detect-clis.sh`, then `JSC_READONLY=1 {JSC_ROOT}/jsc-hooks/tools/wire-cli.sh status {cli}` per detected CLI — this step only takes stock, and `wire-cli.sh` without a subcommand rewires, so the read-only contract is carried in the environment rather than trusted to a correctly typed subcommand | detect-clis exit 0 with no row → no CLI to wire, say so and skip; detect-clis non-zero → report wiring as 無法驗證 with the exit code. Per CLI: 0 wired and 1 degraded → nothing to fix; 2 → usage error, defect in this skill; 3 → CLI not installed, drop it; 5 → collect its `missing` items; 6 → readonly refused the call, which means the subcommand was mistyped into a writing one — nothing on the machine changed; fix the command and rerun that CLI; other → report that CLI as 無法驗證 |
| Versions | `{JSC_ROOT}/jsc-hooks/hooks/version-guard.sh report` | 0 → use the rows, and treat a report with no `{domain}` row or a `noregistry` line as 無法驗證 — other exits → report versions as 無法驗證 with the exit code |
Merge the three with `tools/build-todo.sh --config {設定輸出} --wiring {cli}={接線輸出} --version {版本輸出}` so the ordering rule lives in one place. Exit 0 → the `todo` rows are the work list; exit 2 → usage error, report it as a defect in this skill; exit 3 → name the unreadable input and rerun that one checker; any other exit → stop and report, because a half-merged list would silently drop a whole class of items.
Merge the three with `{CLI_ROOT}/tools/build-todo.sh --config {設定輸出} --wiring {cli}={接線輸出} --version {版本輸出}` so the ordering rule lives in one place. Exit 0 → the `todo` rows are the work list; exit 2 → usage error, report it as a defect in this skill; exit 3 → name the unreadable input and rerun that one checker; any other exit → stop and report, because a half-merged list would silently drop a whole class of items.
Keep the version report from that run. Step 3 hands it to `jsc-cli:deploy` instead of making it query again.
@@ -43,8 +86,8 @@ Done when every item is either confirmed with a value or recorded as skipped.
| Route | Action |
| --- | --- |
| `auto` on a variable | `tools/apply-config.sh set {KEY} {VALUE}` |
| `auto` on a directory | `tools/apply-config.sh mkdir {PATH}` |
| `auto` on a variable | `{CLI_ROOT}/tools/apply-config.sh set {KEY} {VALUE}` |
| `auto` on a directory | `{CLI_ROOT}/tools/apply-config.sh mkdir {PATH}` |
| `ask` | same two commands, with the value the user just gave |
| `manual` | print the exact steps and the file to edit; the operator does it |
| domain 落後 | call `jsc-cli:deploy` with mode `update` **and the version report from step 1**, so it neither re-asks the mode nor re-queries the versions |
@@ -74,8 +117,8 @@ Pick the check by what was actually written, because the two kinds of write beco
| What was written | Re-verify with | Why this check |
| --- | --- | --- |
| An environment variable in a shell rc file | `tools/apply-config.sh show`, confirming the `KEY<TAB>VALUE` line is in the `# jsc-config` block | The block is a fact that is already true. The variable reaching the environment is not — a rc file does not touch the running shell |
| A directory | `tools/scan-config.sh scan {scope}`, confirming the row is no longer `missing` or `invalid` | The directory exists the moment it is created |
| An environment variable in a shell rc file | `{CLI_ROOT}/tools/apply-config.sh show`, confirming the `KEY<TAB>VALUE` line is in the `# jsc-config` block | The block is a fact that is already true. The variable reaching the environment is not — a rc file does not touch the running shell |
| A directory | `{CLI_ROOT}/tools/scan-config.sh scan {scope}`, confirming the row is no longer `missing` or `invalid` | The directory exists the moment it is created |
| Wiring, versions, model tags (delegated) | The owner skill's own returned result | The owner already ran its own verification |
The same exit branching as step 1 applies to `scan-config.sh` and to `apply-config.sh`.
@@ -88,14 +131,76 @@ Done when every applied item has a fresh verdict from the checker its own row na
## 5. Record
Rewrite `CHECK_{HASH}` through `jsc-gitea:wiki` with the post-fix state, per `templates/check-page.md`. That page is a **content page** and keeps only the latest run, so this overwrites the pre-fix picture on purpose.
The two pages live in **two different wiki repos**. Resolve each one on its own.
`CHECK_CONTENTS` is a **contents page** and gets the opposite treatment: read it back first, then upsert this machine's row per `templates/check-contents.md` — add the row if missing, otherwise refresh its counts and 最後體檢. Never overwrite the whole page, and never touch another machine's row.
Rewrite `CHECK_{HASH}` through `jsc-gitea:wiki` with the post-fix state, per `{CLI_ROOT}/templates/check-page.md` — repo from `{JSC_ROOT}/jsc-gitea/tools/gitea.sh wiki-repo CHECK`. That page is a **content page** and keeps only the latest run, so this overwrites the pre-fix picture on purpose.
The `CHECK_CONTENTS` read branches by exit code, and only exit 4 opens the create path. Exit 0 means upsert into the content that came back. Exit 4 means the page really is not there yet, so build it from the template. Exit 7 (key invalid or no permission) and exit 8 (any other API failure) mean the old rows are unknown, not that the page is missing: skip the `CHECK_CONTENTS` write, name the exit code, and create nothing — the overwrite that is correct for `CHECK_{HASH}` would here destroy every other machine's row, unread and unrecoverable.
`CHECK_CONTENTS` is a **contents page**, it lives in the contents repo (`{JSC_ROOT}/jsc-gitea/tools/gitea.sh wiki-repo CONTENTS`, never a fallback to `JSC_WIKI_REPO_CHECK`), and it gets the opposite treatment. It is an H1, a `>` preamble and one H2 block per machine — no markdown table anywhere on it. Write the block with
No wiki repo configured, or any non-zero exit from the wiki write → report the tables on screen, say the record was skipped, and name the exit code.
`{JSC_ROOT}/jsc-gitea/tools/wiki-contents.sh upsert CHECK 1 "CHECK_{HASH}" {block file} {CLI_ROOT}/templates/check-contents.md`
which reads the page back and refreshes this machine's block, or appends it when missing. Never overwrite the whole page, and never touch another machine's block.
The block file holds the whole H2 block: the `## CHECK_{HASH}` line, a blank line, then one bullet per field in the order `{CLI_ROOT}/templates/check-contents.md` lists them, written as `- {欄位名}：{值}` with a full-width colon. Every field of the template gets a bullet, `HASH` included — the heading is the key, and a field only in the heading is a field the next reader cannot read.
**The key is the H2 heading, `CHECK_{HASH}`.** It is the name of the content page this block points at: the literal `CHECK_` plus exactly what `hash-id` printed for `{host}/{user}` in step 1 — 40 uppercase hex characters, not shortened, not otherwise prefixed, not wrapped in a link, no date appended. The script compares the whole heading text after trimming, so the key and that heading must match character for character.
A key holding a URL would be a moving key: the URL changes with `GITEA_HOST`, with a move of `JSC_WIKI_REPO_CHECK` to another repo, and with Gitea's encoding of the page name. The page name moves with none of them — `{host}/{user}` alone decides it. Key on the URL and the comparison never matches, so every run appends a second block for the same machine instead of updating it.
`1` is `<key-col>`, and it only matters while a page is still the old markdown table: it is the 1-based index of the column that held the identity, the `[CHECK_{HASH}]({URL})` cell in column 1, whose text becomes the H2 heading when the script converts that table to blocks. On a page already in block form the script ignores it.
The `體檢頁` bullet stays the human-facing link and is never the key; the heading itself carries no link and no URL. Write the bullet as `[CHECK_{HASH}]({absolute URL})` — text plus link, the one link form this skill set uses. The URL comes from `{JSC_ROOT}/jsc-gitea/tools/gitea.sh wiki-url {CHECK repo} CHECK_{HASH}`, fetched after `CHECK_{HASH}` is rewritten, and is never composed by hand.
A same-wiki link form resolves only inside its own wiki, and the two pages are no longer in the same one. It fails without an error, reading on screen as plain text or a dead link, so nobody finds it and nobody fixes it.
**Verify every link before writing it.** Collect every URL heading into `CHECK_{HASH}` or into the `CHECK_CONTENTS` block, then hand the whole list to `{JSC_ROOT}/jsc-gitea/tools/link-check.sh`. It prints `{OK|DEAD|SKIP}<TAB>{URL}<TAB>{reason}` per line. Only exit 0 may be written.
| Exit | Meaning | Action |
| --- | --- | --- |
| 0 | Every link answers | Write the page |
| 1 | At least one link is unreachable | Write nothing, on either page. Report the `DEAD` lines and leave both pages as they are |
| 2 | Usage error: no URL was given | Report it as a defect in this skill and pass the URLs |
| 3 | The list holds a Gitea URL but `GITEA_HOST` is unset | Skip both writes and report `GITEA_HOST` as still unfixed. Never write without verifying |
| 7 | Gitea authentication failed | Stop and report the key problem. This is not a dead link |
Exit 7 stays apart from exit 1 on purpose: with a dead key, Gitea's answer for a private repo looks the same as "page absent". Merge the two and one expired key marks every live page dead, and the blocks pointing at them get rewritten or dropped.
The script asks the API and never reads a web status code. A private repo's web URL answers 404 to a request with no credentials, so status codes turn good links into dead ones.
| Exit | Action |
| --- | --- |
| 0 | Report the `updated` or `added` result with the repo and page it named |
| 1 | The page content could not be built, or the write failed, and nothing landed. Report it with the stderr. A page with no block to replace is not this case: the script appends instead |
| 2 | Usage error. Report it as a defect in this skill; do not retry with guessed arguments. A `{CLI_ROOT}/templates/check-contents.md` that is not on disk also lands here — then name the path the script looked for, confirm the plugin install is complete, and rerun |
| 3 | No contents repo configured. Skip this write and report `JSC_WIKI_REPO_CONTENTS` as still unfixed |
| 4 | Unreachable the way this skill calls the script — the command above always passes `{CLI_ROOT}/templates/check-contents.md`, and a template that is not on disk comes back as exit 2. So treat a 4 as a malformed call: report it as a defect in this skill, name the command that produced it, and do not retry with guessed arguments |
| 7 | Key invalid or no permission. Nothing was read or written; name the exit code and create nothing |
| 8 | Any other API failure. Same as 7 |
Exits 7 and 8 never mean the page is missing: the whole-page overwrite that is correct for `CHECK_{HASH}` would here destroy every other machine's block, unread and unrecoverable. The script creates a page only when its own read reported that page absent, and it owns that branch.
`{JSC_ROOT}/jsc-gitea/tools/gitea.sh wiki-url` has its own exits, and they are read before the upsert runs. Exit 4 means `CHECK_{HASH}` is not on the wiki yet, so rewrite that page first and fetch the URL again. Any other non-zero exit: name the exit code and stop — never hand-build the URL, because a guessed link goes into the block and points nowhere.
No wiki repo configured, or any non-zero exit from `wiki-repo`, `hash-id`, `wiki-url` or the wiki write → report the tables on screen, say the record was skipped, and name the exit code.
Then state the counts: fixed, skipped, delegated, and 未修好. Recommend `/jsc-cli:doctor` for a clean re-check when anything was delegated.
Done when the page is written or the skip is reported, and the four counts are stated.
Done when every link written into either page passed `link-check.sh` first — or the `DEAD` list is on screen and that write was skipped — each of the two pages is written or its skip is reported, and the four counts are stated.
## 6. Record how the run ended
This is the last thing this skill does, and it runs on every path out of the skill, the ones that stop at step 1 included. Call
`{JSC_ROOT}/jsc-hooks/tools/report-status.sh skill-end jsc-cli:setup {status} {exit code} [detail]`
`{exit code}` is the exit code of whatever decided the outcome — usually the `apply-config.sh` call that ruled the run — and `0` when nothing failed. `{detail}` is one short line, no more than 200 characters: the four counts fit there, the item table does not, and no value the user typed goes in it. **If the script is not on this machine, skip this step in silence and finish the run as it stood** — missing infrastructure is not a failure, and a reporting call may never change what this skill returns or reports.
| status | When this skill uses it |
| --- | --- |
| `ok` | Every item on the list was confirmed and applied, each one re-verified by the checker its own row names, nothing was skipped, and both pages were written |
| `blocked` | The environment refused every write, so nothing on the machine changed: `apply-config.sh` returned 4 on each item because the backup or the write failed. A run that could not touch a single rc file did no work, so it is never reported as `failed` half-done |
| `failed` | Writes landed but the run broke: an applied item still fails its re-verification in step 4, `apply-config.sh` returned 2 on a malformed call this skill made, or `link-check.sh`, `gitea.sh` or `wiki-contents.sh` returned 7 or 8 and the record could not be rewritten |
| `degraded` | The run finished with part of the list untouched. The usual case is the user turning an item down at step 2 — a skip is recorded as skipped and never as fixed — and a delegated repair that its owner skill did not close counts the same way. The machine is better than it was, and the 未修好 and 略過 counts are above zero |
| `aborted` | The user stopped the sequential confirmation partway and asked to end the run, so the remaining items were never put to them |
Done when exactly one `skill-end` line was recorded for this run, or the script was absent and the run finished without it.
templates/check-contents.md
+25 -5
View File
@@ -1,9 +1,29 @@
# 體檢目錄
> 由 `jsc-cli:doctor` 維護。每台執行環境一列；`HASH` 取 `{主機名}/{登入帳號}`，算法與其他頁面共用。
> 由 `jsc-cli:doctor` 維護。每台執行環境一個區塊；`HASH` 取 `{短主機名}/{登入帳號}`，算法與其他頁面共用。主機名取不含網域的短名，FQDN 要先切掉第一個點之後的部分，否則同一台機器會多出第二頁。
>
> 寫入語意：一列代表一台執行環境，也就是一組主機加帳號。寫入前先讀回整頁，該執行環境已經有列就更新那一列，沒有才在文末附加一列，最後整頁寫回。禁止整頁覆蓋，也不得改動別人的列。體檢頁 `CHECK_{HASH}` 只留最新一次結果、可以整頁改寫，這份目錄頁不行。
> 本頁落在 `JSC_WIKI_REPO_CONTENTS` 解出的專用存取庫，與體檢頁 `CHECK_{HASH}` 不同庫。目錄頁全部住這裡，不退回 `JSC_WIKI_REPO_CHECK`。
>
> 路徑寫法：底下每一支腳本都以字面絕對路徑呼叫，不留 `$JSC_HOME`、不留波浪號，也不留裸的相對路徑——權限層靜態比對路徑，帶變數的比不中任何規則，相對路徑則會對著操作者當下的工作目錄解，而那裡從來不是外掛根目錄。`{JSC_ROOT}` 是 `readlink -f "$JSC_HOME/current"` 解出的那個目錄，解到那一層就停，不再往下解成帶版本號的快取路徑；`{CLI_ROOT}` 是 CLI 載入技能時講明的 jsc-cli 外掛基底目錄。兩者都在技能開跑時各解一次，之後逐字代入。
>
> 寫入語意：一個區塊代表一台執行環境，也就是一組主機加帳號。一律用 `{JSC_ROOT}/jsc-gitea/tools/wiki-contents.sh upsert CHECK 1 CHECK_{HASH} {區塊檔} {CLI_ROOT}/templates/check-contents.md` 寫，它先讀回整頁，該執行環境已經有區塊就整塊換掉，沒有才在頁尾附加一個區塊。禁止整頁覆蓋，也不得改動別人的區塊。體檢頁 `CHECK_{HASH}` 只留最新一次結果、可以整頁改寫，這份目錄頁不行。
>
> 參數說明：第二個參數 `1` 是 `<key-col>`，只在這一頁還留著舊的 markdown 表格時用得到，指舊表格裡持有身分那一欄的序號，也就是持有 `[CHECK_{HASH}](網址)` 的第 1 欄，自動轉檔時取那一格的文字當 H2 標題；頁面已經是條列格式就完全忽略它。第三個參數 `<key>` 是 H2 標題文字，也就是體檢頁頁名 `CHECK_{HASH}`。第四個參數是區塊檔，不是列檔：內容為 `## CHECK_{HASH}` 那一行、一個空行，再接各條欄位。
>
> 鍵是 H2 標題 `CHECK_{HASH}`：`{JSC_ROOT}/jsc-gitea/tools/hash-id` 印出什麼就接在 `CHECK_` 後面，完整 40 碼大寫十六進位，不截短、不加別的前後綴、不包成連結、不加日期。腳本比對的是去掉頭尾空白後的整段標題文字，鍵一定要跟標題一字不差。
>
> 鍵用頁名才穩。頁名只由 `{短主機名}/{登入帳號}` 決定；`GITEA_HOST` 換掉、`JSC_WIKI_REPO_CHECK` 換過存取庫、Gitea 對頁名的網址編碼有差，網址就跟著變，頁名一個字都不動。拿含網址的值當鍵，比對就永遠比不中，同一台機器每體檢一次就多附一個區塊，畫面上還看不出來。
>
> 連結寫法：「體檢頁」那一條的連結只給人點，不當鍵用；H2 標題本身不放連結、不放網址。連結一律寫成 `[{文字}]({連結})`，也就是 `[CHECK_{HASH}]({wiki-url 印出的絕對網址})`，網址取 `{JSC_ROOT}/jsc-gitea/tools/gitea.sh wiki-url` 印出的那一串，不自己組路徑。同 wiki 的雙括號寫法一概不用：兩頁分屬不同存取庫，連不過去，畫面上還看不出壞掉。
>
> 連結驗證：這個區塊要寫進去的每一個連結，先交給 `{JSC_ROOT}/jsc-gitea/tools/link-check.sh`，結束碼 0 才寫。有任何一筆 DEAD 就整個區塊都不寫，把連不到的清單回報給呼叫端。結束碼 3 代表 `GITEA_HOST` 沒設定，先設好再寫，不得跳過驗證；結束碼 7 代表金鑰失效，停下來回報金鑰問題，不要當成死連結。驗證一律走 API，不看網頁狀態碼：私有存取庫的網頁網址對未登入請求一律回 404，拿狀態碼判會把好連結判成壞的，整批砍掉還在的頁。
| 體檢頁 | 主機 | 帳號 | 必要項缺漏 | 設定錯誤 | 最後體檢 |
| --- | --- | --- | --- | --- | --- |
| [[CHECK_{HASH}]] | {hostname} | {使用者帳號} | {n} | {n} | {yyyy-MM-dd HH:mm} |
## CHECK_{HASH}
- 體檢頁：[CHECK_{HASH}]({wiki-url 印出的絕對網址})
- 主機：{短主機名}
- 帳號：{使用者帳號}
- HASH：{HASH}
- 必要項缺漏：{n}
- 設定錯誤：{n}
- 最後體檢：{yyyy-MM-dd HH:mm}
templates/check-page.md
+3 -2
View File
@@ -2,6 +2,7 @@
> 由 `jsc-cli:doctor` 維護。這是體檢頁 `CHECK_{HASH}`，只保留最新一次結果，重跑就整頁覆寫。
> 修復請執行 `/jsc-cli:setup`，它讀這頁的「待修項目」逐項處理。
> 底下提到的腳本都以字面絕對路徑呼叫，不留 `$JSC_HOME`、不留波浪號，也不留裸的相對路徑；`{CLI_ROOT}` 是 CLI 載入技能時講明的 jsc-cli 外掛基底目錄，技能開跑時取一次，之後逐字代入。
- 體檢時間：{yyyy-MM-dd HH:mm}
- 工作目錄：{絕對路徑}
@@ -44,7 +45,7 @@
## 待修項目
> 前六欄直接來自 `jsc-cli/tools/build-todo.sh` 的 `todo` 列，順序與類別由那支腳本決定，這裡不另行排序。
> 前六欄直接來自 `{CLI_ROOT}/tools/build-todo.sh` 的 `todo` 列，順序與類別由那支腳本決定，這裡不另行排序。
> 「影響」欄由 `jsc-cli:doctor` 補上。`/jsc-cli:setup` 從這張表接手；沒有待修項目時腳本會印一列「無」。
| 順序 | 類別 | 項目 | 範圍 | 現況 | 修法 | 影響 |
@@ -53,7 +54,7 @@
## 未登錄變數
> 原始碼有用到、`config-spec.tsv` 沒登錄的變數。體檢不判定它們，只提醒維護者補登錄。
> 原始碼有用到、`{CLI_ROOT}/tools/config-spec.tsv` 沒登錄的變數。體檢不判定它們，只提醒維護者補登錄。
| 變數 | 出現次數 |
| --- | --- |
tools/config-spec.tsv
+18 -13
View File
@@ -14,26 +14,31 @@
# fix auto=工具算得出，可直接寫入 ask=值要人給，問完才寫 manual=只能人手動處理 -=不需修
# desc 一句繁中說明，直接印給使用者看
#
# wiki 存取庫分兩路：目錄頁（{TYPE}_CONTENTS）全部住 JSC_WIKI_REPO_CONTENTS 解出的專用存取庫，
# 內容頁（{TYPE}_{HASH}）才看自己的型別變數。目錄頁不退回型別變數，型別變數也管不到目錄頁。
#
GITEA_HOST env global yes - gitea-api ask Gitea 站台位址，所有 wiki 與 PR 操作的去處
GITEA_TOKEN env global yes - gitea-auth ask Gitea API token；未設定時退回 tea CLI 的登入金鑰
JSC_HOME env global no ~/.jsc dir auto hook 資料目錄，放工作階段計時、用量統計、版本快取
JSC_WIKI_REPO env global no - wiki-repo ask 未逐類設定時的共用 wiki {owner}/{repo}
JSC_WIKI_REPO_QUESTION env global no JSC_WIKI_REPO wiki-repo ask QUESTION_CONTENTS、QUESTION_{HASH} 所在存取庫
JSC_WIKI_REPO_PLAN env global no JSC_WIKI_REPO wiki-repo ask PLAN_CONTENTS、PLAN_{HASH} 所在存取庫
JSC_WIKI_REPO_ANALYZE env global no JSC_WIKI_REPO wiki-repo ask ANALYZE_CONTENTS、ANALYZE_{HASH} 所在存取庫
JSC_WIKI_REPO_DELIVER env global no JSC_WIKI_REPO wiki-repo ask DELIVER_CONTENTS、DELIVER_{HASH} 所在存取庫
JSC_WIKI_REPO_MAINTAIN env global no JSC_WIKI_REPO wiki-repo ask MAINTAIN_CONTENTS、MAINTAIN_{HASH} 所在存取庫
JSC_WIKI_REPO_REPO env global no JSC_WIKI_REPO wiki-repo ask REPO_CONTENTS、REPO_{HASH} 所在存取庫
JSC_WIKI_REPO_LOG env global no JSC_WIKI_REPO wiki-repo ask LOG_CONTENTS、LOG_{HASH} 所在存取庫
JSC_WIKI_REPO_LEARN env global no JSC_WIKI_REPO wiki-repo ask LEARN_CONTENTS、LEARN_{HASH} 所在存取庫
JSC_WIKI_REPO_ERROR env global no JSC_WIKI_REPO wiki-repo ask ERROR_CONTENTS、ERROR_{HASH} 所在存取庫
JSC_WIKI_REPO_CHECK env global no JSC_WIKI_REPO wiki-repo ask CHECK_CONTENTS、CHECK_{HASH} 所在存取庫，體檢紀錄寫在這裡
JSC_WIKI_REPO_REPORT env global no JSC_WIKI_REPO wiki-repo ask REPORT_CONTENTS、REPORT_{HASH} 所在存取庫，年月週日報表寫在這裡
JSC_WIKI_REPO_CONTENTS env global no JSC_WIKI_REPO wiki-repo ask 全部目錄頁（{TYPE}_CONTENTS）所在存取庫；沒有型別變數可退，只退 JSC_WIKI_REPO
JSC_WIKI_REPO_QUESTION env global no JSC_WIKI_REPO wiki-repo ask QUESTION_{HASH} 內容頁所在存取庫
JSC_WIKI_REPO_PLAN env global no JSC_WIKI_REPO wiki-repo ask PLAN_{HASH} 內容頁所在存取庫
JSC_WIKI_REPO_ANALYZE env global no JSC_WIKI_REPO wiki-repo ask ANALYZE_{HASH} 內容頁所在存取庫
JSC_WIKI_REPO_DELIVER env global no JSC_WIKI_REPO wiki-repo ask DELIVER_{HASH} 內容頁所在存取庫
JSC_WIKI_REPO_MAINTAIN env global no JSC_WIKI_REPO none - 保留待用，不必設定：MAINTAIN 只有目錄頁 MAINTAIN_CONTENTS，沒有內容頁，目錄頁又已改看 JSC_WIKI_REPO_CONTENTS，這一列目前沒有頁面可管；留著這一列是為了讓孤兒掃描認得這個變數
JSC_WIKI_REPO_REPO env global no JSC_WIKI_REPO wiki-repo ask REPO_{HASH} 內容頁所在存取庫
JSC_WIKI_REPO_LOG env global no JSC_WIKI_REPO wiki-repo ask LOG_{HASH} 內容頁所在存取庫
JSC_WIKI_REPO_LEARN env global no JSC_WIKI_REPO wiki-repo ask LEARN_{HASH} 內容頁所在存取庫
JSC_WIKI_REPO_ERROR env global no JSC_WIKI_REPO wiki-repo ask ERROR_{HASH} 內容頁所在存取庫
JSC_WIKI_REPO_CHECK env global no JSC_WIKI_REPO wiki-repo ask CHECK_{HASH} 內容頁所在存取庫，體檢紀錄寫在這裡
JSC_WIKI_REPO_REPORT env global no JSC_WIKI_REPO wiki-repo ask REPORT_{HASH} 內容頁所在存取庫，年月週日報表寫在這裡
JSC_VERSION_GUARD env global no on set ask 設成 off 可完全略過版本前置檢查，離線工作時用
JSC_VERSION_TTL env global no 600 set ask 版本查詢快取秒數
JSC_RESTART_GATE env global no on set ask 設成 off 可略過部署後的重啟提示閘門，判讀在 jsc-hooks
JSC_WIKI_REPO_SKILLSET env global no JSC_WIKI_REPO wiki-repo ask SKILLSET_CONTENTS、SKILLSET_{HASH} 所在的 {owner}/{repo}，技能組異動報告寫在這裡
JSC_WIKI_REPO_TOOLING env global no JSC_WIKI_REPO wiki-repo ask TOOLING_CONTENTS、TOOLING_{HASH} 所在的 {owner}/{repo}，技能盤點寫在這裡
JSC_WIKI_REPO_SKILLSET env global no JSC_WIKI_REPO wiki-repo ask SKILLSET_{HASH} 內容頁所在的 {owner}/{repo}，技能組異動報告寫在這裡
JSC_WIKI_REPO_TOOLING env global no JSC_WIKI_REPO wiki-repo ask TOOLING_{HASH} 內容頁所在的 {owner}/{repo}，技能盤點寫在這裡
JSC_WIKI_REPO_MONITOR env global no JSC_WIKI_REPO wiki-repo ask MONITOR_{HASH} 內容頁所在存取庫，助理巡檢紀錄寫在這裡
JSC_LANG_GUARD env global no on set ask 設成 off 可關閉繁中編碼與簡體字守門，誤判時用
JSC_COMMENT_SCOPE env global no on set ask 設成 off 可關閉註解夾帶文件編號的守門，誤判時用
JSC_CHANGED_FILE internal runtime no - none - 非 Claude CLI 傳入的變更檔路徑，註解範圍與繁中編碼守門讀它
1 # config-spec.tsv — jsc 技能組的設定規格表。體檢（/jsc-cli:doctor）與設定（/jsc-cli:setup）共用這一份。
14 # fix auto=工具算得出，可直接寫入 ask=值要人給，問完才寫 manual=只能人手動處理 -=不需修
15 # desc 一句繁中說明，直接印給使用者看
16 #
17 # wiki 存取庫分兩路：目錄頁（{TYPE}_CONTENTS）全部住 JSC_WIKI_REPO_CONTENTS 解出的專用存取庫，
18 # 內容頁（{TYPE}_{HASH}）才看自己的型別變數。目錄頁不退回型別變數，型別變數也管不到目錄頁。
19 #
20 GITEA_HOST
21 GITEA_TOKEN
22 JSC_HOME
23 JSC_WIKI_REPO
24 JSC_WIKI_REPO_QUESTION JSC_WIKI_REPO_CONTENTS
25 JSC_WIKI_REPO_PLAN JSC_WIKI_REPO_QUESTION
26 JSC_WIKI_REPO_ANALYZE JSC_WIKI_REPO_PLAN
27 JSC_WIKI_REPO_DELIVER JSC_WIKI_REPO_ANALYZE
28 JSC_WIKI_REPO_MAINTAIN JSC_WIKI_REPO_DELIVER
29 JSC_WIKI_REPO_REPO JSC_WIKI_REPO_MAINTAIN
30 JSC_WIKI_REPO_LOG JSC_WIKI_REPO_REPO
31 JSC_WIKI_REPO_LEARN JSC_WIKI_REPO_LOG
32 JSC_WIKI_REPO_ERROR JSC_WIKI_REPO_LEARN
33 JSC_WIKI_REPO_CHECK JSC_WIKI_REPO_ERROR
34 JSC_WIKI_REPO_REPORT JSC_WIKI_REPO_CHECK
35 JSC_WIKI_REPO_REPORT
36 JSC_VERSION_GUARD
37 JSC_VERSION_TTL
38 JSC_RESTART_GATE
39 JSC_WIKI_REPO_SKILLSET
40 JSC_WIKI_REPO_TOOLING
41 JSC_WIKI_REPO_MONITOR
42 JSC_LANG_GUARD
43 JSC_COMMENT_SCOPE
44 JSC_CHANGED_FILE
tools/deploy.sh
+240 -7
View File
@@ -9,13 +9,29 @@
# cmd<TAB>{指令} 即將執行的指令
# exit<TAB>{結束碼}<TAB>{指令} 該指令的結束碼；dry-run 時結束碼印「-」
# skip<TAB>{domain}<TAB>{原因} 本地 clone 是開發中的樹，略過 git pull
# warn<TAB>{domain}<TAB>{原因} 相依版本不符，仍照樣更新的提醒
# warn<TAB>{domain}<TAB>{原因} 照樣裝下去、但要人知道的事：相依版本不符，
# 或本機複本這一輪沒有重新拉取（拿不到
# 更新鎖，或 git pull 回非零），那時候
# 裝的是磁碟上現有的內容，可能不是最新版
# compat<TAB>codex<TAB>{舊路徑}<TAB>{新路徑} codex 舊版快取路徑補成指向新版的相容連結
# note<TAB>{cli}<TAB>{原因} 非逐指令的說明（例：kiro 整批改走複製退路的理由）
# restart<TAB>{路徑} 這次寫下的重啟狀態檔
# link<TAB>{domain}<TAB>{狀態}<TAB>{連結路徑}<TAB>{指向或原因} current 連結農場這一條的刷新結果
# restart<TAB>{路徑} 這次寫下的重啟狀態檔。install 與 update
# 一律寫，整輪判定成 fail 也照寫：外掛
# 已經換了一部分，這時候更需要重啟
# requires<TAB>{domain}<TAB>{檢查結果} update 前的 jsc.requires 檢查
# result<TAB>{cli}<TAB>{mode}<TAB>{domain 清單}<TAB>{ok|fail}
# 結束碼：全部指令成功 0；任一指令失敗 1；參數錯誤 2。skip、warn、note 不算失敗，但呼叫端要據實回報。
# 結束碼：全部指令成功 0；任一指令失敗 1；參數錯誤 2。skip、warn、note、link 不算失敗，
# 但呼叫端要據實回報。
# link 那幾行講的是 $JSC_HOME/current/ 這一組不帶版本號的符號連結：一個 domain 一條，
# 指向該外掛在快取裡帶版本號的實體目錄。技能文件裡所有跨外掛的腳本呼叫都以那一層當根，
# 因為它不帶版本號、寫得進權限允許清單。install 與 update 收尾會把整組連結刷新一次，
# uninstall 收尾會清掉指向已消失的那幾條。狀態欄的五個值：
# ok 連結已經指到這次實際安裝的版本目錄
# removed 指向已消失，這一條清掉了
# skip 這一輪不動它，原因寫在最後一欄
# fail 該指的目標取不到或連結建不起來，這一條維持原樣，指向可能還是舊版本
# dryrun 乾跑，只算出會指到哪裡，沒有真的寫
# update 前每個 domain 都先跑一次同目錄的 check-requires.sh，它的四種結束碼分流如下：
# 0 相依符合，或沒有宣告相依 → 照常更新這個 domain
# 1 相依版本不符或缺相依 plugin
@@ -57,6 +73,7 @@ MKT="$HOST/$OWNER/meta.git"
REPO_BASE="$HOST/$OWNER"
LOCAL_DIR="${JSC_LOCAL_PLUGINS:-${JSC_HOME:-$HOME/.jsc}/plugins}"
KIRO_SKILLS="${JSC_KIRO_SKILLS:-$HOME/.kiro/skills}"
CURRENT_DIR="${JSC_HOME:-$HOME/.jsc}/current"
DRYRUN="${JSC_DEPLOY_DRYRUN:-0}"
FAILED=0
@@ -113,6 +130,141 @@ codex_preserve_old_plugin_cache() { # $1=plugin name $2=required file under vers
done
}
# 連結農場一台機器只有一組，五支 CLI 卻各裝各的副本，所以只能挑一支當基準。
# 挑法是固定的偏好順序，取這台機器上第一支裝得到的：
# claude 擺第一——現有那幾條連結本來就指向它的快取，換基準等於把每一條寫進技能文件的
# 路徑悄悄換到另一份副本上，而文件不會跟著改。
# codex 次之，同樣是 CLI 自己的外掛快取寫下來的帶版本目錄，版本號是它裝到什麼就是什麼。
# copilot 與 kiro 只當退路：它們的路徑不帶版本號，連結指得到，卻查不出指的是哪一版。
# antigravity 一律不當基準。它的來源是本地 clone，那份 clone 允許是維護者的開發樹，
# sync_local 碰到未提交的變更還會刻意不 pull；把全機器的路徑指到一份做到一半的樹，
# 正好就是這次要修掉的那種無聲跑錯腳本。
# 用 command -v 判斷裝沒裝，跟 detect-clis.sh 的認定一致，所以每一支平行跑的 deploy.sh
# 都會算出同一個基準，整輪只有一支真的去寫連結。五支都寫的話，平行行程會互相覆寫，
# 最後指到哪一份是隨機的，出了事也重現不出來。
link_base_cli() {
for _b in claude codex copilot kiro; do
if command -v "$(cli_bin "$_b")" >/dev/null 2>&1; then
printf '%s' "$_b"
return 0
fi
done
return 1
}
# 某個外掛在快取根目錄底下的實體版本目錄。claude 與 codex 的快取一個版本一層，
# 取版本排序最大、而且真的有 plugin.json 的那一層：裝到一半的目錄沒有 plugin.json，
# 挑到它會讓連結指向一份不完整的外掛，而且照樣不會報錯。
# 本身就是符號連結的項目也要跳過——舊版相容連結就是那樣，指過去只是多繞一層，
# 原目標一被清掉還會跟著一起斷。
latest_version_dir() { # $1=某個外掛的快取根目錄
[ -d "$1" ] || return 1
_vfound=$(
for _vp in "$1"/*; do
[ -d "$_vp" ] || continue
[ -L "$_vp" ] && continue
[ -f "$_vp/plugin.json" ] || continue
printf '%s\n' "$_vp"
done | sort -V | tail -n1
)
[ -n "$_vfound" ] || return 1
printf '%s' "$_vfound"
}
# 基準 CLI 底下這個 domain 的實體目錄。這裡只收得起當基準的那四支，
# antigravity 不列：它從來不會被選成基準，補一條沒人走的分支只會讓人以為那也是選項。
plugin_dir() { # $1=CLI 代號 $2=domain；印出實體目錄，取不到回 1
case "$1" in
claude) latest_version_dir "$HOME/.claude/plugins/cache/jsc/jsc-$2" ;;
codex) latest_version_dir "$(codex_plugin_cache_root "jsc-$2")" ;;
copilot)
_pd="${COPILOT_HOME:-$HOME/.copilot}/installed-plugins/jsc/jsc-$2"
[ -d "$_pd" ] || return 1
printf '%s' "$_pd"
;;
kiro)
_pd="$KIRO_SKILLS/jsc-$2"
[ -d "$_pd" ] || return 1
printf '%s' "$_pd"
;;
*) return 1 ;;
esac
}
link_line() { # $1=domain 或 - $2=狀態 $3=連結路徑 $4=指向或原因
printf 'link\t%s\t%s\t%s\t%s\n' "$1" "$2" "$3" "$4"
}
# 刷新一條連結。
# 失敗只記一行、不中止整輪：這一段跑在外掛都裝好之後，部署本身已經成功了。把整輪記成
# 失敗會連帶跳過重啟閘門與 result 行，操作者拿到的是一台明明裝好、卻被說成失敗的機器，
# 反而更難處置。連結沒刷新確實是缺陷，但補救方式是一行看得見的紀錄加上人來判斷，
# 不是把一次成功的部署丟掉。
refresh_one() { # $1=基準 CLI $2=domain
_rl="$CURRENT_DIR/jsc-$2"
# 目標已經在、又不是符號連結時一律不覆寫。ln -sfn 對著一個實體目錄下手，會把連結建進
# 那個目錄裡面（變成 current/jsc-{domain}/jsc-{domain}），農場當場就壞了還不會報錯；
# 改成 rm -rf 清掉更糟，那是這支腳本沒建過、也可能沒有第二份的內容。
if [ -e "$_rl" ] && [ ! -L "$_rl" ]; then
link_line "$2" skip "$_rl" '目標已存在而且不是符號連結，這一條不覆寫，請人工確認'
return 0
fi
if ! _rt=$(plugin_dir "$1" "$2"); then
if [ "$DRYRUN" = 1 ]; then
link_line "$2" dryrun "$_rl" "乾跑沒有真的安裝，$1 底下還取不到 jsc-$2 的版本目錄"
else
link_line "$2" fail "$_rl" "$1 底下找不到 jsc-$2 的版本目錄，這一條維持原樣"
fi
return 0
fi
if [ "$DRYRUN" = 1 ]; then
link_line "$2" dryrun "$_rl" "$_rt"
return 0
fi
if mkdir -p "$CURRENT_DIR" 2>/dev/null && ln -sfn "$_rt" "$_rl" 2>/dev/null; then
link_line "$2" ok "$_rl" "$_rt"
else
link_line "$2" fail "$_rl" '連結建不起來，這一條維持原樣，指向可能還是舊版本'
fi
}
# 解除安裝之後清掉指向已經消失的那一條。判準是連結還在、指向卻沒了：
# 這一輪只解除安裝其中一支 CLI 時，連結可能還指著另一支 CLI 手上完好的副本，那一條要留著。
# 斷掉的連結不清，等於對外宣稱一支已經不在機器上的外掛還裝著，下一輪體檢會照著它報錯狀態。
prune_one() { # $1=domain
_ql="$CURRENT_DIR/jsc-$1"
[ -L "$_ql" ] || return 0
[ -e "$_ql" ] && return 0
if [ "$DRYRUN" = 1 ]; then
link_line "$1" dryrun "$_ql" '指向已消失，這一條會被清掉'
return 0
fi
if rm -f "$_ql" 2>/dev/null; then
link_line "$1" removed "$_ql" '-'
else
link_line "$1" fail "$_ql" '指向已消失卻清不掉，這一條還是斷的'
fi
}
# 整組連結的刷新收尾。跑在各 CLI 的部署動作全部結束之後，不夾在每個 domain 的安裝指令
# 中間：連結要指的是寫完的內容，中途去指有機會指到只寫了一半的目錄。
refresh_links() {
if ! _lbase=$(link_base_cli); then
link_line - skip "$CURRENT_DIR" '這台機器一支基準 CLI 都沒裝到，沒有版本目錄可指'
return 0
fi
if [ "$CLI" != "$_lbase" ]; then
link_line - skip "$CURRENT_DIR" "這一輪的基準 CLI 是 $_lbase，$CLI 這一支不動連結"
return 0
fi
for _ldom in $DOMAINS; do
case "$MODE" in
install|update) refresh_one "$_lbase" "$_ldom" ;;
uninstall) prune_one "$_ldom" ;;
esac
done
}
usage() {
echo "用法：deploy.sh [-n] {install|update|uninstall} {claude|codex|copilot|antigravity|kiro} {domain} [domain...]" >&2
exit 2
@@ -234,21 +386,93 @@ local_hold() { # $1=存取庫路徑
[ "${ahead:-0}" -eq 0 ] || printf '有 %s 個未推送的 commit' "$ahead"
}
# 取一個 domain 的更新鎖。
#
# 這一份本機複本一台機器只有一份，而五支 CLI 的部署是平行跑的——平行是對的，它們寫的是
# 不同的外掛目錄。但這裡不是：五支都會來 pull 同一個目錄，而 git 對同一個存取庫的併發寫入
# 沒有保護。
# 2026-09-07 實測踩到：同一輪裡兩支 CLI 撞在一起，一支拿不到 ORIG_HEAD.lock、一支的遠端
# refs 換不上去，兩支的整輪判定都變成 fail——而外掛其實全部裝好了。那是最難查的一種失效：
# 報告說失敗，實際成功，而真正的原因跟部署無關。
# 一個 domain 一把鎖，用 mkdir 取：那是檔案系統這一層唯一原子的建立動作，兩支同時 mkdir
# 只有一支成功。等不到就改用磁碟上的內容——別人正在拉同一份，硬等下去只是排隊。
clone_lock() { # $1=鎖目錄 $2=最多試幾次（每次之間睡一秒，所以約等於秒數）；0=拿到了
_try=0
while :; do
mkdir "$1" 2>/dev/null && return 0
_try=$((_try + 1))
[ "$_try" -ge "$2" ] && break
sleep 1
done
# 等不到還要分一種情形：上一輪中途死掉會留下一個沒有人放的鎖，那種鎖永遠等不到。
# 判準取鎖的年紀，門檻放寬到等待秒數的四倍——比任何一次正常的 pull 都久。
if [ -d "$1" ]; then
_age=$(( $(date +%s) - $(date -r "$1" +%s 2>/dev/null || date +%s) ))
if [ "$_age" -gt $(( $2 * 4 )) ]; then
rmdir "$1" 2>/dev/null || true
mkdir "$1" 2>/dev/null && return 0
fi
fi
return 1
}
# 截一段訊息到指定長度。cut -c 數的是位元組不是字元，一個多位元組字剛好被切成兩半時
# 尾巴會留一個替代字元，而亂碼不影響結束碼、沒有人會來報。截完再過一次 iconv -c 丟掉
# 那個不完整的序列；iconv 不在這台機器上就退回原樣。
clip1() { # $1=位元組上限；讀標準輸入
_raw=$(tr '\n' ' ' | cut -c"1-$1")
_cln=$(printf '%s' "$_raw" | iconv -c -f UTF-8 -t UTF-8 2>/dev/null)
if [ -n "$_cln" ]; then printf '%s' "$_cln"; else printf '%s' "$_raw"; fi
}
# 把某 domain 的存取庫抓到本地：有 .git 就 pull，沒有就 clone。
# 目標是開發中的樹時只印 skip，改用現地內容安裝，不 pull：這支腳本可以被指到任何
# 目錄，蓋掉維護者未提交或未推送的工作救不回來，安裝一份舊內容還能重跑。
sync_local() { # $1=domain
dir="$LOCAL_DIR/$1"
lock="$LOCAL_DIR/.lock-$1"
if [ -d "$dir/.git" ]; then
hold=$(local_hold "$dir")
if [ -n "$hold" ]; then
printf 'skip\t%s\t%s %s，未執行 git pull\n' "$1" "$dir" "$hold"
return 0
fi
run git -C "$dir" pull
else
run git clone "$REPO_BASE/$1.git" "$dir"
# 乾跑一步都不能動，連鎖都不取：取鎖是建目錄，那已經是寫入了。
if [ "$DRYRUN" = 1 ]; then
printf 'cmd\t%s\n' "git -C $dir pull"
printf 'exit\t-\t%s\n' "git -C $dir pull"
return 0
fi
if ! clone_lock "$lock" 30; then
printf 'warn\t%s\t%s\n' "$1" "$dir 的更新鎖等了 30 秒還拿不到，別的 CLI 正在拉同一份；這一輪改用磁碟上現有的內容，沒有重新拉取"
return 0
fi
printf 'cmd\t%s\n' "git -C $dir pull"
_out=$(git -C "$dir" pull 2>&1); _rc=$?
rmdir "$lock" 2>/dev/null || true
printf 'exit\t%s\t%s\n' "$_rc" "git -C $dir pull"
if [ "$_rc" -ne 0 ]; then
# 複本已經在磁碟上，拉不動不等於裝不了：安裝改用現有內容，那可能是舊版。
# 這裡刻意不記 FAILED。記了整輪會判成 fail，而 fail 那條路徑會連重啟閘門一起跳過
# ——外掛換了一半而沒有人被告知要重啟，比一句「拉不動」嚴重得多。
# 但一定要印出來：安靜地裝一份舊內容，是這一組工具最怕的那種失效。
printf 'warn\t%s\t%s\n' "$1" "git pull 回 $_rc，這一輪用磁碟上現有的內容安裝，可能不是最新版：$(printf '%s' "$_out" | clip1 160)"
fi
return 0
fi
if [ "$DRYRUN" = 1 ]; then
run git clone "$REPO_BASE/$1.git" "$dir"
return 0
fi
if ! clone_lock "$lock" 60; then
FAILED=1
printf 'warn\t%s\t%s\n' "$1" "$dir 還不存在，而 clone 鎖等了 60 秒拿不到，這個 domain 這一輪沒有本機複本可以裝"
return 1
fi
run git clone "$REPO_BASE/$1.git" "$dir"
_rc=$?
rmdir "$lock" 2>/dev/null || true
return "$_rc"
}
# claude、copilot、kiro-cli 共用的 plugin 指令組。
@@ -432,8 +656,17 @@ case "$CLI" in
*) usage ;;
esac
refresh_links
# 重啟閘門一律先掛，不等整輪判定。
#
# 部分失敗代表磁碟上的外掛已經換了一部分，這時候更需要重啟，不是更不需要。
# 2026-09-07 實測踩到：一條與部署無關的 git pull 失敗把整輪判成 fail，而原本的寫法是
# 「判定成功才掛閘門」——於是那一支 CLI 沒有被掛上閘門，沒有人被告知要重啟，而它跑的
# 是舊版程式碼。一道只在成功時才生效的提醒，在最需要它的那一次不會出現。
mark_restart
if [ "$FAILED" -eq 0 ]; then
mark_restart
printf 'result\t%s\t%s\t%s\tok\n' "$CLI" "$MODE" "$DOMAINS"
exit 0
fi
tools/scan-config.sh
+2
View File
@@ -97,6 +97,8 @@ if [ "$cmd" = orphans ]; then
# 去掉再比對，否則每次體檢都會多報一個不存在的變數。
name=${name%_}
printf '%s\n' "$known" | grep -qx "$name" && continue
# 只排掉樣板佔位字那一個名字，不做前綴比對：CONTENTS、MONITOR 這些是真的頁面類型，
# 排寬了就等於把整族存取庫變數從孤兒掃描裡挖掉，漏登錄再也看不出來。
case "$name" in JSC_WIKI_REPO_TYPE) continue ;; esac
printf 'orphan\t%s\t%s\n' "$name" "$count"
done