移除自造的待驗單機制,改由 Gitea 原生三格承接 #62
Closed
claude-code
wants to merge 0 commits from
fix/remove-claim-worksheet-mechanism into main
pull from: fix/remove-claim-worksheet-mechanism
merge into: inkstone:main
inkstone:main
inkstone:fix/roster-plugin-prefix
inkstone:chore/v0.16.0
inkstone:feat/nag-and-wiki-compress
inkstone:feat/named-workers-and-reply-gate
inkstone:feat/manifest-account-v2
inkstone:feat/debt-worklist
inkstone:feat/release-manifest-and-milestone-account
inkstone:chore/v0.13.0
inkstone:feat/handoff-writeback-loose
inkstone:feat/mainline-milestone
inkstone:chore/v0.12.0
inkstone:fix/cloud-wiring-isep90
inkstone:feat/pr-verdict-v2
inkstone:feat/countdown-stamp
inkstone:feat/gate-search-is-not-proof
inkstone:fix/stamp-proves-seen-not-run
inkstone:feat/gate-catches-recon-without-output
inkstone:docs/inventory-missing-rows
inkstone:fix/tests-leave-no-forged-pending
inkstone:integrate/isep59-wave2
inkstone:fix/gitea-write-search-guard
inkstone:feat/leo21c-mcp-guard
inkstone:fix/dispatch-guard-parens-and-wiring
inkstone:fix/dispatch-guard-noise-and-wiring
inkstone:feat/three-native-fields
inkstone:feat/ticket-carries-the-task
inkstone:feat/gate-fires-when-asking-not-after-stopping
inkstone:fix/stamp-release-stat-ordering-linux
inkstone:fix/beacon-says-which-copy
inkstone:fix/factory-idle-guard-blocks-actions-not-mentions
inkstone:fix/beacon-counts-registered-gates
inkstone:fix/push-guard-cross-repo-stamp
inkstone:feat/vendor-drift-guard
inkstone:fix/push-guard-target-not-substring
inkstone:feat/milestone-must-have-due
inkstone:fix/worksheet-dedup-by-content
inkstone:fix/b4-real-probe
inkstone:fix/cloud-setup-git-auth-verify
inkstone:feat/cloud-env-outdir
inkstone:fix/testing-b-section-discriminating
inkstone:fix/setup-self-verify
inkstone:fix/claim-worksheet-dedup
inkstone:fix/cloud-env-parity-14
inkstone:fix/sdd-guard-path-resolution
inkstone:feat/cloud-env-generator
inkstone:fix/one-copy-per-gate
inkstone:fix/group0-gates
inkstone:chore/withdraw-versions
inkstone:rule/milestone-formation
inkstone:rule/pr-only
inkstone:docs/three-source
inkstone:docs/plan
inkstone:release/v0.2.1
inkstone:fix/ticket-api-bypass
inkstone:leaf/24-descope
inkstone:release/v0.2.0
inkstone:leaf/5-cloud
inkstone:leaf/18-dogfood
inkstone:leaf/6-release
inkstone:leaf/2-docs
inkstone:leaf/3-wiki
inkstone:leaf/4-labels
No Reviewers
Labels
Clear labels
Human
hub
human/exec
👤 這件事需要 leo 親手批示/動手(花錢・不可逆・品味方向・物理動作)。與 s/* 正交
close
duplicate
重複——關新的、留舊的,並指向舊票
close
human-exec
人執票完成,由人手動關
close
merged
PR merge 自動關(正常路徑)
close
split
太大,已拆成數張子票
close
stale
逾期無資訊,自動關閉
close
transferred
走錯棚,已在正確的 repo 重開
close
wontfix
討論後決定不做(只有 leo 可下此判斷)
Tracking issue——聚合一批 leaf 的 scope 容器,本身不掛 milestone、不對應 PR
👤 這張票的『執行者』是人(GUI-only/實體權限/法定簽署),不是審核。交付物是環境狀態改變,不是 PR
p
high
高——擋住交付或有時間壓力
p
low
低——想做,但晚一點沒關係
s
backlog
驗過了、確定要做,但還沒排進任何 sprint
s
doing
進行中——已 self-assign,現在有人在做
s
pending
卡住——等外部/等依賴,不是沒人做
s
review
PR 已開(含 closes #n),等總管 review 並 merge
s
stage
已部署 stage,等 leo 實際驗收
s
todo
已排進 milestone,等人領票
s
triage
新進來的,還沒驗傷——還沒決定要不要做
type
bug
壞掉的東西
type
chore
雜務、整理、不改行為
type
feature
新能力
type
governance
規範、流程、閘本身
No Label
Milestone
No items
No Milestone
Projects
Clear projects
No project
No Assignees
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: inkstone/ISEP#62
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Delete Branch "fix/remove-claim-worksheet-mechanism"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Closes
inkstone/ISEP#60.這個 PR 解什麼
自造的「待驗單」第四套機制(
subagent-claim-worksheet.sh+claim-verify-police.sh)整組移除。違反 D58:不要硬做平台不支援的機制——待驗單只是一張沒有狀態、沒有持有人的 markdown,
必然退化成雜訊(同一份重生三次、同內容兩個檔名)。改由
inkstone/ISEP#59/PR #58的三個 Gitea 原生欄位承接:子票相依(存在嗎/做完了嗎)、
s/*tag(卡在哪)、指派(誰該動)。內容
hooks/hooks.json:移除 Stop/SubagentStop 兩條註冊hooks/subagent-claim-worksheet.sh/hooks/claim-verify-police.sh:整組刪除hooks/lib/path-resolve.sh:拿掉已刪檔案的註解引用docs/hooks-inventory.md、README.md:更新閘數量(48→46 檔、59→57 條註冊)docs/governance/sdd-gitea-governance.md:E12/E14 標記現況,新增 §8.4 說明移交對象.claude-plugin/plugin.json:0.5.0 → 0.6.0實測(貼在票上,這裡只放結論)
scripts/test-*.sh全數 6 支通過(14/14、10/10、11/11、8/8、13/13、13/13)hooks/tests/*.test.sh:與 pristineorigin/main基準比對,失敗特徵完全一致(
gitea-arm-check/main-and-prod-push-guard-cross-repo/main-and-prod-push-guard/prod-write-guard/stage-before-prod-guard這幾支在環境裡本來就會這樣失敗,不是本次改動造成——已用兩個 worktree 對照過);其餘
ask-user-question-guard/dispatch-format-guard/factory-idle-guard/reply-identity/sdd-guard全綠claude plugin validate .→✔ Validation passedbash scripts/check-version-consistency.sh→ 目前會報不一致(0.6.0vs 最新 tagv0.5.0),這是預期的過渡態,等總管收斂 release 打
v0.6.0tag 時解決(同 PR #58 的先例)沒做的(誠實標記)
docs/governance/DIVERGENCE-v0.5.0-to-v0.6.0.md未動:那是 2026-08-20 的時間點快照(總管的意見書),修改它等於竄改歷史記錄,不在本票範圍
InkStoneCo/.claude/pending-verification/{done,done-20260826,verified}/三個歸檔目錄未動:它們是 InkStoneCo repo 裡已提交(tracked)的檔案,跨 repo 且需要 InkStoneCo 自己的
git 流程處理,超出本票(ISEP repo)範圍,判斷留給總管
【分診 @ inkstone/ISEP#64】狀態:✅ 該併
實測:
git merge-tree origin/main origin/fix/remove-claim-worksheet-mechanism無衝突;git merge-base --is-ancestor origin/main origin/該分支→ true(直接架在現在的 main 上,不必 rebase)。內容:整組移除自造的「待驗單」第四套機制(
subagent-claim-worksheet.sh+claim-verify-police.sh),理由是違反 D58(不硬做平台不支援的機制)——待驗單是一張沒有狀態、沒有持有人的 markdown,會退化成雜訊(PR 描述附證據:同一份重生三次、同內容兩個檔名)。改由ISEP#59/#58的 Gitea 原生三格承接。PR 自陳測試:6 支腳本測試全過(14/14、10/10、11/11、8/8、13/13、13/13);
hooks/tests/*.test.sh與 pristine main 對照,失敗特徵完全一致(main-and-prod-push-guard等幾支在這個環境本來就會這樣失敗,不是本次改動造成,用兩個 worktree 對照過);claude plugin validate .通過。這裡補一個判斷材料:本次分診查到
main-and-prod-push-guard.sh讀的CLAUDE_AGENT_NAME環境變數在 Claude Code 裡從來不存在,是那支閘自己編的假設——這正好解釋了為什麼 PR 描述裡那幾支 guard 測試「在這個環境裡本來就會這樣失敗」:它們的判斷條件本身建立在一個不存在的環境變數上,併這條分支不會讓情況變壞(本 PR 沒有修改那支閘本身),但那支閘的失敗特徵不會消失,這不是本 PR 該解的範圍。注意:本 PR 依賴
ISEP#59/#58描述的「三個原生欄位」已經到位(概念上),但程式碼層面它是直接架在 main 上、不含#58的任何 commit——也就是說如果#62先併、#58還沒併,會有一段時間「舊機制已拆、新機制的三個新動詞(subtask/handback/mine)還沒進 main」的空窗。建議併的順序:先併#58,再併#62,避免功能空窗期。[decision] 關閉:這個 PR 要做的事,在 main 上已經是既成事實
總管實查現在的 main(2026-08-28 12:45):
⇒ 這個 PR 要移除的兩支,已經不在 main 上了。內容在
v0.6.0就進去了(那一版的說明寫著「移除了自造的待驗單機制:它今天在同一個 session 裡重生三次
——同一批東西驗過了還會回來擋,因為那個機制沒有『已處理』這個狀態」)。
所以這不是「不做」,是「已經做了、只是這張 PR 忘了關」。
📌 順帶一個今天的實證:本 session 載入的是舊版 plugin(0.3.9),
那兩支已刪除的 hook 在它上面還活著,於是
.claude/pending-verification/今天在工作區裡又長回來一次——總管清掉一次,它重生一次。
那正是這個 PR 的判斷(「沒有『已處理』狀態的機制必然退化成雜訊」)在現場的樣子。
為什麼現在才關:今天出的
v0.11.0(inkstone/ISEP#81)新增了一支收工閘,會清點「還有哪些 PR 沒有結論」並點名。而它上線後第一批要抓的,就包含這一張。
與其讓那支閘去抓,不如現在給結論——我自己不遵守鐵律 7,那支閘就是裝飾。
Pull request closed