universal-dev-standards 6.14.0-beta.6 → 6.14.0-beta.8
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +13 -3
- package/bin/uds.js +30 -6
- package/bundled/ai/options/code-review/pr-review.ai.yaml +20 -1
- package/bundled/ai/standards/code-review.ai.yaml +15 -2
- package/bundled/ai/standards/developer-memory.ai.yaml +12 -2
- package/bundled/ai/standards/full-coverage-testing.ai.yaml +44 -6
- package/bundled/ai/standards/open-work-tracking.ai.yaml +191 -11
- package/bundled/ai/standards/pipeline-security-gates.ai.yaml +48 -3
- package/bundled/ai/standards/project-context-memory.ai.yaml +12 -2
- package/bundled/core/code-review-checklist.md +20 -2
- package/bundled/core/developer-memory.md +12 -2
- package/bundled/core/full-coverage-testing.md +44 -4
- package/bundled/core/open-work-tracking.md +156 -16
- package/bundled/core/pipeline-security-gates.md +63 -0
- package/bundled/core/project-context-memory.md +10 -2
- package/bundled/locales/zh-CN/CHANGELOG.md +78 -3
- package/bundled/locales/zh-CN/README.md +1 -1
- package/bundled/locales/zh-CN/SECURITY.md +1 -1
- package/bundled/locales/zh-CN/adoption/ADOPTION-GUIDE.md +13 -4
- package/bundled/locales/zh-CN/adoption/STATIC-DYNAMIC-GUIDE.md +7 -1
- package/bundled/locales/zh-CN/adoption/checklists/enterprise.md +7 -1
- package/bundled/locales/zh-CN/adoption/checklists/minimal.md +13 -3
- package/bundled/locales/zh-CN/adoption/checklists/recommended.md +7 -1
- package/bundled/locales/zh-CN/core/code-review-checklist.md +17 -0
- package/bundled/locales/zh-CN/core/developer-memory.md +9 -0
- package/bundled/locales/zh-CN/core/full-coverage-testing.md +48 -8
- package/bundled/locales/zh-CN/core/pipeline-security-gates.md +64 -4
- package/bundled/locales/zh-CN/core/project-context-memory.md +5 -0
- package/bundled/locales/zh-CN/docs/AVAILABLE-STANDARDS.md +70 -0
- package/bundled/locales/zh-CN/docs/CHEATSHEET.md +4 -3
- package/bundled/locales/zh-CN/docs/CLI-INIT-OPTIONS.md +25 -23
- package/bundled/locales/zh-CN/docs/FEATURE-REFERENCE.md +13 -11
- package/bundled/locales/zh-CN/docs/USAGE-MODES-COMPARISON.md +6 -6
- package/bundled/locales/zh-CN/docs/WINDOWS-GUIDE.md +7 -1
- package/bundled/locales/zh-CN/docs/user/GETTING-STARTED.md +44 -0
- package/bundled/locales/zh-CN/options/code-review/pr-review.md +13 -2
- package/bundled/locales/zh-TW/CHANGELOG.md +78 -3
- package/bundled/locales/zh-TW/README.md +1 -1
- package/bundled/locales/zh-TW/SECURITY.md +1 -1
- package/bundled/locales/zh-TW/adoption/ADOPTION-GUIDE.md +13 -4
- package/bundled/locales/zh-TW/adoption/STATIC-DYNAMIC-GUIDE.md +7 -1
- package/bundled/locales/zh-TW/adoption/checklists/enterprise.md +7 -1
- package/bundled/locales/zh-TW/adoption/checklists/minimal.md +13 -3
- package/bundled/locales/zh-TW/adoption/checklists/recommended.md +7 -1
- package/bundled/locales/zh-TW/core/code-review-checklist.md +17 -0
- package/bundled/locales/zh-TW/core/developer-memory.md +9 -0
- package/bundled/locales/zh-TW/core/full-coverage-testing.md +48 -8
- package/bundled/locales/zh-TW/core/open-work-tracking.md +90 -14
- package/bundled/locales/zh-TW/core/pipeline-security-gates.md +64 -4
- package/bundled/locales/zh-TW/core/project-context-memory.md +5 -0
- package/bundled/locales/zh-TW/docs/AVAILABLE-STANDARDS.md +70 -0
- package/bundled/locales/zh-TW/docs/CHEATSHEET.md +4 -3
- package/bundled/locales/zh-TW/docs/CLI-INIT-OPTIONS.md +25 -23
- package/bundled/locales/zh-TW/docs/FEATURE-REFERENCE.md +13 -11
- package/bundled/locales/zh-TW/docs/USAGE-MODES-COMPARISON.md +6 -6
- package/bundled/locales/zh-TW/docs/WINDOWS-GUIDE.md +7 -1
- package/bundled/locales/zh-TW/docs/user/GETTING-STARTED.md +44 -0
- package/bundled/locales/zh-TW/options/code-review/pr-review.md +13 -2
- package/bundled/skills/SKILL_NAMING.md +9 -0
- package/bundled/templates/gates/check-anti-fake-tests.mjs +305 -5
- package/package.json +3 -1
- package/src/commands/audit.js +9 -2
- package/src/commands/check.js +17 -0
- package/src/commands/hitl.js +3 -0
- package/src/commands/init.js +41 -21
- package/src/commands/list.js +12 -3
- package/src/commands/mcp.js +2 -3
- package/src/commands/open-work.js +65 -6
- package/src/commands/quickstart.js +29 -13
- package/src/commands/release.js +41 -17
- package/src/commands/skills.js +19 -33
- package/src/commands/spec-split.js +28 -16
- package/src/commands/spec.js +34 -13
- package/src/commands/uninstall.js +2 -2
- package/src/commands/update.js +61 -61
- package/src/core/open-work-config.js +103 -0
- package/src/flows/init-flow.js +10 -0
- package/src/hitl/manager.js +4 -2
- package/src/i18n/messages.js +240 -63
- package/src/installers/skills-installer.js +3 -1
- package/src/mcp/__tests__/server.test.js +0 -1
- package/src/mcp/server.js +16 -4
- package/src/prompts/init.js +148 -13
- package/src/reconciler/index.js +42 -5
- package/src/uninstallers/skills-uninstaller.js +83 -16
- package/src/uninstallers/standards-uninstaller.js +56 -7
- package/src/utils/agents-installer.js +10 -17
- package/src/utils/available-standards-report.js +144 -0
- package/src/utils/available-standards.js +216 -0
- package/src/utils/copier.js +5 -4
- package/src/utils/friction-detector.js +27 -0
- package/src/utils/open-work-tracking.mjs +919 -48
- package/src/utils/prompt-closed.js +18 -0
- package/src/utils/skill-frontmatter.js +102 -0
- package/src/utils/skill-name-collision.js +169 -0
- package/src/utils/skills-install-paths.js +114 -0
- package/src/utils/skills-installer.js +63 -105
- package/src/utils/transient-fs.js +113 -0
- package/standards-registry.json +34 -9
|
@@ -2,8 +2,8 @@
|
|
|
2
2
|
|
|
3
3
|
> **Language**: English | [繁體中文](../locales/zh-TW/core/open-work-tracking.md)
|
|
4
4
|
|
|
5
|
-
**Version**: 1.
|
|
6
|
-
**Last Updated**: 2026-
|
|
5
|
+
**Version**: 1.4.0
|
|
6
|
+
**Last Updated**: 2026-10-07
|
|
7
7
|
**Applicability**: Any project that carries work across more than one working session and risks losing an item between them
|
|
8
8
|
**Scope**: universal
|
|
9
9
|
|
|
@@ -35,10 +35,10 @@ Three distinct ways work goes missing between sessions are routinely folded into
|
|
|
35
35
|
| 某項目因等待別的事件而暫停 | 「等待中」若沒有記錄解除條件,與「被忘記」無法分辨 | 與等待一起記錄的解除條件 |
|
|
36
36
|
| 已規劃的項目還沒動工,時間過去 | 沒有時鐘的項目會無聲腐爛——沒有東西會再指向它 | 一個門檻,或一次被迫的定期檢視,讓它重新浮現 |
|
|
37
37
|
|
|
38
|
-
Each requirement below traces to one row of that table, to one of four failures observed the day this standard's design was drafted, or (OWT-017–OWT-019) to two gaps found in 1.1.0 (see [Evidence and calibration](#evidence-and-calibration)). **None of the mechanisms is prescribed** — per the same constraint [deferred-item-exit](deferred-item-exit.md) states for its own exits, and for the same reason (DEC-049: UDS defines relations that must hold, adoption layers choose what maintains them).
|
|
38
|
+
Each requirement below traces to one row of that table, to one of four failures observed the day this standard's design was drafted, or (OWT-017–OWT-019) to two gaps found in 1.1.0, or (OWT-020–OWT-026) to two states 1.1.0 left unnamed, or (OWT-027–OWT-029) to work that waits on another project (see [Evidence and calibration](#evidence-and-calibration)). **None of the mechanisms is prescribed** — per the same constraint [deferred-item-exit](deferred-item-exit.md) states for its own exits, and for the same reason (DEC-049: UDS defines relations that must hold, adoption layers choose what maintains them).
|
|
39
39
|
|
|
40
40
|
下面每一條要求都對應這張表的一列、對應本標準設計當天觀察到的四個失效之一,
|
|
41
|
-
或(OWT-017–OWT-019)對應 1.1.0
|
|
41
|
+
或(OWT-017–OWT-019)對應 1.1.0 補上的兩個缺口、或(OWT-020–OWT-026)對應 1.1.0 沒有命名的兩個狀態、或(OWT-027–OWT-029)對應等著另一個專案的工作
|
|
42
42
|
(見〈[證據與校準](#evidence-and-calibration)〉)。**沒有任何一個機制被規定**——
|
|
43
43
|
理由與 [deferred-item-exit](deferred-item-exit.md) 對自己出口的約束相同(DEC-049:
|
|
44
44
|
UDS 定義必須成立的關係,維持它的機制由採用層選擇)。
|
|
@@ -83,13 +83,13 @@ UDS 定義**活動**,採用層負責**編排**(DEC-049)。一份寫成工
|
|
|
83
83
|
|
|
84
84
|
## The invariant
|
|
85
85
|
|
|
86
|
-
**A carrier of open work must (1) accept a new item without demanding classification, (2) record a release condition for every item it marks waiting, (3) generate any field a reliable source already determines, (4) disclose what it cannot see whenever it reports what remains, (5) be checked at the moment control returns from agent to human — by something that cannot fail the turn, (6) keep what the work is *for* apart from how far it has got, account for every edit to the former,
|
|
86
|
+
**A carrier of open work must (1) accept a new item without demanding classification, (2) record a release condition for every item it marks waiting, (3) generate any field a reliable source already determines, (4) disclose what it cannot see whenever it reports what remains, (5) be checked at the moment control returns from agent to human — by something that cannot fail the turn, (6) keep what the work is *for* apart from how far it has got, account for every edit to the former, (7) make every "next action" name something a reader can go and find (or, for an item already asked and awaiting a reply, say when it was asked, what it waits for and what releases it), and (8) tell a waiting item nobody has asked about from one awaiting a reply, and give a fact it cannot see a stamped place to live — `unknown` a real value, every observation's age shown, and (9) let a wait name an object in another project by its logical name, say whether this machine can see it, and keep that record where the sessions of every project involved can read it.**
|
|
87
87
|
|
|
88
88
|
**一個承載開放工作的地方,必須:(1)不要求分類就能收下新項目、(2)為每一個標為等待中的項目記下解除條件、
|
|
89
89
|
(3)對任何有可靠來源可推導的欄位改用生成、(4)回報還剩什麼時同時揭露看不到什麼、
|
|
90
90
|
(5)在控制權從 agent 交回人的那一刻被檢視——而且那個檢視不能讓回合失敗、
|
|
91
91
|
(6)把「這份工作為了什麼」與「做到哪了」分開存放,並替前者的每一次修改留下交代、
|
|
92
|
-
(7
|
|
92
|
+
(7)讓每一個「下一步」都點名一個讀的人找得到的東西(對已問、等回覆的項目,則寫明何時問的、在等什麼、什麼事件解除它)、(8)分得出「還沒問」與「已問、等回覆」,並讓看不見的事實有一個帶戳記的存放處——`unknown` 是正式的值、每筆觀察的年齡一律顯示、(9)讓一個等待可以用邏輯名稱點名另一個專案裡的東西、說明這台機器看不看得到它,並把那份紀錄放在每個相關專案的 session 都讀得到的地方。**
|
|
93
93
|
|
|
94
94
|
---
|
|
95
95
|
|
|
@@ -115,7 +115,17 @@ UDS 定義**活動**,採用層負責**編排**(DEC-049)。一份寫成工
|
|
|
115
115
|
| **OWT-016** | Any window or threshold this standard's requirements reference carries its provenance, or is marked uncalibrated | warning |
|
|
116
116
|
| **OWT-017** | A carrier that holds a piece of work's goal, acceptance criteria, or constraints holds none of its progress or next action, and the reverse. This is decided by walking each carrier's structural fields (sections, columns, typed markers), never by file names. A progress update therefore never requires touching the goal | warning |
|
|
117
117
|
| **OWT-018** | Every change to a piece of work's goal, acceptance criteria, or constraints leaves a revision record stating what changed, who approved it, and why. A change with no approver is listed when control returns to a human (OWT-007); it is never silent | error |
|
|
118
|
-
| **OWT-019** | A "next action" field names at least one concrete object: a file path, a test name, a command, or a requirement identifier. A verb alone ("continue", "handle the rest") does not. This judges whether an object is named, never how well the sentence is worded | warning |
|
|
118
|
+
| **OWT-019** | A "next action" field names at least one concrete object: a file path, a test name, a command, or a requirement identifier. A verb alone ("continue", "handle the rest") does not. This judges whether an object is named, never how well the sentence is worded. An item in `asked-awaiting` that satisfies OWT-022 is not judged here: its next step may be the reply it waits for, and OWT-022 judges it (when it was asked, what it waits for, what releases it) | warning |
|
|
119
|
+
| **OWT-020** | An item marked waiting states which of two states it is in: `not-yet-asked` (the request is drafted or decided and has not been sent) or `asked-awaiting` (it has been sent and nothing has come back). A waiting item in neither is named individually in the open-work summary — never folded into a total, never counted as done | warning |
|
|
120
|
+
| **OWT-021** | An item in `not-yet-asked` names the draft or action whose sending releases it: a file path, a command, a test name or a requirement identifier (the recognition OWT-019 uses) | warning |
|
|
121
|
+
| **OWT-022** | An item in `asked-awaiting` carries `asked-at` (a calendar day with a year, not later than today) and satisfies OWT-002: what it waits for and what event releases it | warning |
|
|
122
|
+
| **OWT-023** | A hand-written row is allowed only for a fact the carrier cannot derive from version control, a spec marker or a CI result — for example whether a message was sent, whether the other side replied, whether something was approved. Each such row carries `observed-by` (who or what saw it), `observed-at` (a calendar day with a year, not later than today) and a value of `yes`, `no` or `unknown`; a row missing any of the three is a violation. A check decides that the fields exist and are well formed — never that the observation is true | warning |
|
|
123
|
+
| **OWT-024** | `unknown` is a value, not a blank: rows valued `unknown` are counted and shown apart in any summary and are never counted as complete. A row marked done whose value is `unknown` is a violation | warning |
|
|
124
|
+
| **OWT-025** | A summary of observations shows each observation's age (days since `observed-at`). An observation older than the declared threshold is reported as stale, counted apart, and never counted as confirmed. A stamp says who saw it and when — never that it still holds | warning |
|
|
125
|
+
| **OWT-026** | The exception in OWT-023 does not reach a field that version control, a spec marker or a CI result determines; OWT-003 stands. A check can only decide this against a word list of such subjects, so its coverage is unknown (OWT-011) | warning |
|
|
126
|
+
| **OWT-027** | The release condition of a waiting item (OWT-002) may name an object in another project, written `<project>:<object>`. The project is a **logical name** (`uds`, `vibeops`), never a directory; the object is a file path, a test name, a command or a requirement identifier (the kinds OWT-019 recognises) or a version-control tag; and the carrier holds **no machine-specific absolute path**. Where a name lives on a given machine is declared outside the carrier. An item that waits for such an object has no request that could have been asked, so the two states of OWT-020 do not apply to it | warning |
|
|
127
|
+
| **OWT-028** | For each such release condition a check says one of four things, counted apart and never folded into one total: **released** (the object exists in the other project now), **not yet released** (the project is visible and the object is not there), **not visible from here** (this machine cannot see the project: not released, and not zero) or **needs a person** (the project is visible but this kind of object, such as a test name, a command or a requirement identifier, has nothing a check can read). The look-up reads this machine's files and version control only and opens no network connection | warning |
|
|
128
|
+
| **OWT-029** | The record of work that waits on another project is a file tracked by version control in a repository the sessions of every project involved can read. It is not held only in a personal AI-tool memory, which is bound to one user and one machine and which another session, or CI, cannot read. Which repository holds it is the adopting project's choice. The reference check does not decide where a carrier is kept | warning |
|
|
119
129
|
|
|
120
130
|
---
|
|
121
131
|
|
|
@@ -207,6 +217,24 @@ This is the same failure DEX-006 names for a different artefact, one layer remov
|
|
|
207
217
|
| 為什麼行為相反 | 它在看的事件本身夠稀少,擋在它上面不會把耐性用完 | TCI 自己的規則已經寫出這裡不能做成閘門的理由:**「一個在每個回合都為真的閘門會被關掉,關掉之後它什麼都不保護」**(TCI R4)。開放工作非空幾乎永遠為真,所以這個確認點被設計成永不保留控制權 |
|
|
208
218
|
| 兩者掛同一事件時的順序 | — | 先回報(OWT-009),所以即使那個回合隨後被 TCI 擋下,它的輸出仍然可見 |
|
|
209
219
|
|
|
220
|
+
### The summary can be printed by a program; the full list is fetched on demand
|
|
221
|
+
|
|
222
|
+
OWT-007 says a summary occurs at the point control returns from agent to human. It does **not** say who produces it or how long it is, and the two obvious ways of meeting it differ a great deal in price. **A program prints a short summary when the turn ends**: it is the output of a command run by whatever already sits on the turn-end event, so it costs the model no tokens, and the **full version is produced only when the reader asks for it**. **Or the model writes a complete table into every reply**: that satisfies OWT-007 and is a valid choice, but it is the expensive one, and it is not what this standard requires. Everything OWT-008 and OWT-009 say still holds for the first way: the program reports and never blocks, and its output comes before any blocking check's verdict.
|
|
223
|
+
|
|
224
|
+
A suggested vocabulary for what a short summary says about each open item (a suggestion, not a requirement): waiting on a person's decision; waiting on an outside party's reply; waiting on CI; due on a fixed date; nobody is looking at it. These line up with what OWT-002 and OWT-020–OWT-022 already let a carrier record (what an item waits for, and whether it was asked); this standard fixes no vocabulary for the summary itself.
|
|
225
|
+
|
|
226
|
+
One adopting project does it this way: a program prints the short summary at the end of a turn, and the full list is produced when the user calls for it. That project's files are private, so they are not linked here. It is an external example, **not part of the UDS package and not guaranteed to be compatible with it**, and it shows only that the approach is workable.
|
|
227
|
+
|
|
228
|
+
**An honest note:** UDS ships no program that produces an OWT-007 summary, and nothing in this standard checks that one exists (see [What enforces this standard](#what-enforces-this-standard)). This subsection describes a cheaper way to satisfy a requirement that already exists. It adds no requirement number and no check.
|
|
229
|
+
|
|
230
|
+
OWT-007 說的是:控制權從 agent 交回人的那一刻要有一份摘要。它**沒有**規定誰產生、多長,而滿足它的兩種明顯做法價格差很多。**由程式在回合結束時印出一份簡短摘要**:那是掛在回合結束事件上的既有機制所執行的一個指令的輸出,所以不消耗模型的 token,而**完整版在讀的人要看時才產生**。**或者由模型在每則回覆裡手工附上一張完整的表**:它滿足 OWT-007,是可行的選擇,但那是昂貴的做法,不是本標準要求的做法。OWT-008 與 OWT-009 對第一種做法照樣成立:程式只回報、永不阻斷,它的輸出排在任何阻斷式檢查的判決之前。
|
|
231
|
+
|
|
232
|
+
簡短摘要對每個開放項目怎麼分類,這裡建議一組詞彙(建議,不是要求):等人決策;等外部(對方回覆);等 CI;固定日期到期;無人看管。它們與 OWT-002、OWT-020–OWT-022 已讓承載庫記錄的東西(在等什麼、有沒有問過)相呼應;本標準不替摘要本身規定任何詞彙。
|
|
233
|
+
|
|
234
|
+
有一個採用專案是這樣做的:回合結束時由程式印出簡短摘要,完整清單在使用者呼叫時才產生。那個專案的檔案是私有的,所以這裡不放連結。它是一個外部範例,**不屬於 UDS 套件,也不保證與它相容**,只用來說明這個做法可行。
|
|
235
|
+
|
|
236
|
+
**誠實註記:** UDS 沒有出貨任何產生 OWT-007 摘要的程式,本標準也沒有任何東西檢查這種程式存在(見〈[什麼在執行本標準](#what-enforces-this-standard)〉)。本小節描述的是滿足一條既有要求的較便宜做法,不新增要求編號,也不新增檢查。
|
|
237
|
+
|
|
210
238
|
---
|
|
211
239
|
|
|
212
240
|
## Anchors: structure, not wording
|
|
@@ -270,31 +298,99 @@ agent 正當地會提出修改,禁止只會教它學會靜默地改——但
|
|
|
270
298
|
它是**結構**判準(有沒有點名對象),不是措辭好壞的判斷;措辭漂亮但什麼都沒點名的句子照樣不過,
|
|
271
299
|
簡短但點了一個測試名稱的句子照樣過。這讓它留在 OWT-010 與 OWT-014 的範圍之內。
|
|
272
300
|
|
|
273
|
-
A check reports three outcomes, never one green: **named and resolved** (the object was found — for instance the path exists), **named, unresolved** (an object is named but could not be found — legitimate when the next action is to create it), and **unnamed** (a violation). Recognising *that* a string is a path, a command, a test name, or an identifier is itself a pattern match, so per OWT-011 its coverage is declared unknown: an unrecognised format is reported as unnamed, and a clean pass never means "every next action is specific".
|
|
301
|
+
A check reports three outcomes, never one green: **named and resolved** (the object was found — for instance the path exists), **named, unresolved** (an object is named but could not be found — legitimate when the next action is to create it), and **unnamed** (a violation). Since 1.4.0 there is a fourth outcome for one kind of row, described below. Recognising *that* a string is a path, a command, a test name, or an identifier is itself a pattern match, so per OWT-011 its coverage is declared unknown: an unrecognised format is reported as unnamed, and a clean pass never means "every next action is specific". The words that mark a field as a next action (`Next action`, `下一步`, `下一個動作`, and so on) are **one list**, read three ways (a heading, a table header, an inline label), and it is uncalibrated (OWT-016): a carrier that heads its column `待辦` or `TODO` is read only once the adopting project declares that word, because the wide words also head columns that are not the next action; and when no field is found at all, the check exits 2 and prints what each carrier showed and which words it knows.
|
|
274
302
|
|
|
275
303
|
檢查回報三種結果,而不是一個綠燈:**點名且已找到**(對象被找到——例如路徑存在)、**點名但未找到**
|
|
276
|
-
|
|
304
|
+
(有點名對象但找不到——當下一步就是要建立它時是正當的)、**未點名**(違反)。自 1.4.0 起,有一種列另有第四種結果,見下面。
|
|
277
305
|
辨認「這串字是路徑、指令、測試名稱還是編號」本身是樣式比對,所以依 OWT-011 其涵蓋率明示為未知:
|
|
278
|
-
|
|
306
|
+
認不出的格式會被回報為未點名,而乾淨的通過絕不表示「每個下一步都夠具體」。標示「下一步」欄位的詞(`Next action`、`下一步`、`下一個動作` 等)是**一份清單**,以三種方式讀(標題、表頭、行內標籤),且屬未校準(OWT-016):欄位標題寫 `待辦` 或 `TODO` 的載體,要等採用專案宣告那個詞才會被讀到,因為範圍太寬的詞也會標在不是下一步的欄位上;而完全找不到欄位時,檢查以 2 結束,並印出每個載體看到了什麼、認得哪些詞。
|
|
307
|
+
|
|
308
|
+
**A fourth outcome, for a reply (1.4.0).** A next action that is *the reply itself* ("wait for the vendor to answer", "wait for them to propose a time") names no file, command, test or identifier, and used to be reported as unnamed. That judged the wrong thing. For an item that was already asked, the starting point is not a verb with an object; it is the three facts **OWT-022** requires: when it was asked (`asked-at`), what it waits for, and what event releases it. An item in `asked-awaiting` that satisfies OWT-022 is therefore judged by OWT-022 and not by OWT-019, and a check reports it as a fourth outcome, **waiting on reply**. It is not a violation. It is **not counted as named** and **not counted as complete**. It is listed on its own with its **age**, and one older than the declared threshold is marked **stale** (the threshold OWT-025 uses, uncalibrated). The exemption is not a hiding place: it applies only when the item is `asked-awaiting` **and** OWT-022 finds nothing missing. A row with no `asked-at` (or one in the future), nothing stated that it waits for, or no release event is still judged by OWT-019; so is a `not-yet-asked` item (its next step is sending the draft, which OWT-021 requires it to name), an item in any other state, and a next action written as a heading section rather than as a row that has a status. A carrier with no status field is read exactly as before. The two checks read the item through one function, so they cannot give opposite answers about it: a row `uds open-work waiting` accepts is never an OWT-019 violation under `next-action`, and a row `waiting` reports under OWT-022 is never waived. **What this cannot decide:** that the request was really sent, or is still unanswered. The check decides only that the fields exist and are well formed, as OWT-023 does for an observation (OWT-014).
|
|
309
|
+
|
|
310
|
+
**回覆的第四種結果(1.4.0)。** 下一步就是*那個回覆本身*(「等廠商回信」「等對方排時間」),沒有點名任何檔案、指令、測試或編號,過去被回報為未點名。那判定的是錯的東西:對一個已經問出去的項目,起點不是「動詞加對象」,而是 **OWT-022** 要求的三件事——何時問的(`asked-at`)、在等什麼、什麼事件解除它。所以處於 `asked-awaiting` 並滿足 OWT-022 的項目,由 OWT-022 判定、不由 OWT-019 判定,檢查把它回報為第四種結果:**等待回覆**。它不是違反;它**不算「已點名」**,也**不算完成**;它被單獨列出並顯示**年齡**,超過宣告門檻的標為**已舊**(沿用 OWT-025 用的門檻,未校準)。這個免除不是藏身處:只有項目是 `asked-awaiting` **而且** OWT-022 找不出缺漏時才適用。沒有 `asked-at`(或在未來)、沒有寫在等什麼、或沒有解除事件的列,照舊由 OWT-019 判定;`not-yet-asked` 的項目也一樣(它的下一步是送出那份草稿,OWT-021 要求它點名草稿);其他狀態的項目,以及寫成小節標題而不是帶狀態欄之列的下一步,也一樣。沒有狀態欄的載體,讀法與過去完全相同。兩個檢查以同一個函式讀這個項目,所以不可能對它給出相反的結論:`uds open-work waiting` 接受的列,在 `next-action` 底下絕不是 OWT-019 違反;`waiting` 以 OWT-022 報違反的列,絕不被免除。**這無法判定的事:**請求是否真的送出、是否仍未回覆。檢查只判定欄位存在且格式正確,與 OWT-023 對一筆觀察所做的相同(OWT-014)。
|
|
311
|
+
|
|
312
|
+
---
|
|
313
|
+
|
|
314
|
+
## A waiting item nobody has asked about is not waiting
|
|
315
|
+
|
|
316
|
+
**OWT-002** asks that a waiting item say what it waits for and what event releases it. Two very different things satisfy that sentence. A message that is **drafted and never sent** waits for "somebody sends it". A message that **was sent** waits for "the other side replies". The carrier cannot tell them apart, and neither can an assistant that cannot see the chat — and the difference is the whole point: nothing will ever arrive for the first until a person acts. **OWT-020** names the two states (`not-yet-asked`, `asked-awaiting`) and requires every waiting item to be in one of them; one that is in neither is named individually, the way OWT-012 names an item left unclassified, and is never counted as done.
|
|
317
|
+
|
|
318
|
+
Each state then carries what makes it checkable. A `not-yet-asked` item names the draft or action (**OWT-021**) — the same recognition OWT-019 uses, so there is one vocabulary — because "ask legal" with no file or command behind it is a mood, not a draft. An `asked-awaiting` item carries `asked-at` (**OWT-022**), which is what starts a clock: from then on the item has an age, and an item with an age can be surfaced again by a threshold. Both read the carrier's structural status field (OWT-010), never the wording around it. The severity is `warning` and the state words are uncalibrated (OWT-016): the aim is to make the two states nameable, not to turn every existing carrier red on the day the standard is published.
|
|
319
|
+
|
|
320
|
+
---
|
|
321
|
+
|
|
322
|
+
## A fact nobody can derive needs a stamp, an honest "unknown", and a visible age
|
|
323
|
+
|
|
324
|
+
Some facts about open work have no source an assistant can read: whether a message was sent, whether the other side answered, whether something was approved. **OWT-003** allows only generation from a derivable source, and the table below rejects a hand-written state file for good reason — it goes stale, and a stamp newer than stale content is invisible. For these facts there is nothing derivable to generate from; refusing hand-written rows outright leaves them nowhere to live, and they end up in free text where nothing counts them.
|
|
325
|
+
|
|
326
|
+
1.2.0 therefore admits a **narrow exception** (**OWT-023**–**OWT-026**), on three conditions. The row says **who saw it and when** (`observed-by`, `observed-at`). Its value is `yes`, `no` or **`unknown`**, and `unknown` is a real value, counted and shown apart and never complete (**OWT-024**) — the same reading as OWT-006: what could not be seen is not zero. And the report shows **every observation's age** (**OWT-025**); one older than the declared threshold is reported as stale, counted apart, and never confirmed. The exception does not reach what version control, a spec marker or a CI result already determines (**OWT-026**), because OWT-003 is not relaxed.
|
|
327
|
+
|
|
328
|
+
What this does not claim is that the whole thing is honest. A stamp says that someone saw it and when; **it does not say it is still so**, and the report is built around that: the age is printed beside the value, so the staleness a stamp would otherwise hide is the first thing a reader sees. And what a check can decide is only what is decidable: that `observed-by` is filled in, that `observed-at` is a real day, that the value is in the domain. **It cannot decide that `observed-by` names the person who actually looked, or that the value is true** — that is a claim about the world, not a relation over artefacts (OWT-014), and this standard does not pretend otherwise. The stale threshold has no measurement behind it (OWT-016).
|
|
329
|
+
|
|
330
|
+
**Why these severities (1.2.0).** OWT-020 to OWT-026 are all `warning`: the state words, field names and the stale threshold are a first vocabulary with no measurement behind it (OWT-016), and a violation costs a reader time rather than costing the work. A stricter level would turn every carrier that predates the two states red on the day this standard is published; `warning` lets an adopter see the list first. **The commands exit 1 on any violation, `warning` included** — as `uds open-work next-action` already does for OWT-019 — so an adopter who wires one into a gate decides whether a warning blocks.
|
|
331
|
+
|
|
332
|
+
---
|
|
333
|
+
|
|
334
|
+
## A wait on another project names it, and does not locate it
|
|
335
|
+
|
|
336
|
+
Projects that depend on each other produce a wait OWT-002 was never written for: one project's item waits for something another project produces. 1.3.0 lets the release condition name it, `uds:v6.14.0` or `vibeops:docs/runbook.md`, and says what keeps that from rotting (**OWT-027**). The project is a **logical name**, never a directory. A directory belongs to one machine and one user (it leaks the user's name, and it is wrong on the next machine), and two readers of one carrier must mean the same project by one name. So the carrier holds the name, and the mapping from a name to a directory on a given machine is declared **outside** the carrier (the reference check takes `--root NAME=DIR`, repeatable, or the adopting project's settings). A machine-specific absolute path in a release condition is a violation. An item that waits for another project's object has no request that could have been "asked", so OWT-020's two states do not apply to it; it is still named and counted on its own line.
|
|
337
|
+
|
|
338
|
+
What a check reports about such a wait is four-way (**OWT-028**), for the same reason OWT-006 exists: **what could not be seen is not zero.** Released; not yet released; **not visible from here**, which means the other project is not on this machine or its version control cannot be read, and which is counted apart and is neither released nor zero; and **needs a person**, which means the project is visible but a test name, a command or a requirement identifier in it has nothing a check can read. A path or a version-control tag is machine-observable: it exists in the other project or it does not. The look-up reads this machine's files and version control and opens no connection, and "released" says the object exists there, never that it is the right object (OWT-014).
|
|
339
|
+
|
|
340
|
+
**OWT-029** says where such a record lives: in a version-controlled repository that the sessions of every project involved can read, and not only in one person's AI-tool memory. That memory is bound to one user and one machine, is invisible to another session and to CI, and gives a check no named carrier to decide a relation over (OWT-014). Which repository is the adopting project's choice. This version covers the sessions of one person on one machine; how work is shared between people, and whether a shared record needs review before it is trusted, are not decided here.
|
|
341
|
+
|
|
342
|
+
**Why `warning`, and what is uncalibrated (1.3.0).** OWT-027 to OWT-029 are `warning` for the reasons given above for OWT-020 to OWT-026. The pattern that makes a logical name, the list of machine-specific path prefixes (home and volume directories, drive letters, `~/`), and the two kinds of object a check can look up (a path and a tag) are a first judgment (OWT-016); coverage is unknown (OWT-011), because a path written in a form the list does not hold is not seen as absolute.
|
|
343
|
+
|
|
344
|
+
---
|
|
345
|
+
|
|
346
|
+
## 一個沒人問過的等待項目,不是在等待
|
|
347
|
+
|
|
348
|
+
**OWT-002** 要求等待中的項目說明在等什麼、什麼事件算解除。有兩件差別很大的事都滿足這句話。一則**草擬了、從沒送出**的訊息,等的是「有人把它送出去」;一則**已經送出**的訊息,等的是「對方回覆」。承載庫分不出這兩者,看不到聊天的助理也分不出——而差別正是重點:第一種在有人動手之前,永遠不會有任何東西到來。**OWT-020** 為這兩個狀態命名(`not-yet-asked`、`asked-awaiting`),並要求每個等待項目必處於其一;兩者皆非的項目被個別點名——與 OWT-012 點名未分類項目的方式相同——而且絕不算成已完成。
|
|
349
|
+
|
|
350
|
+
每個狀態再帶上讓它可被檢查的東西。`not-yet-asked` 的項目點名那份草稿或那個動作(**OWT-021**)——沿用 OWT-019 的辨認方式,所以詞彙只有一份——因為只有「問法務」、背後沒有檔案或指令,是一種心情,不是草稿。`asked-awaiting` 的項目帶有 `asked-at`(**OWT-022**),那是讓時鐘開始走的東西:從那一刻起項目有了年齡,有年齡的項目才能被門檻再次浮現。兩者都讀承載庫的結構狀態欄(OWT-010),絕不讀它周圍的措辭。嚴重度是 `warning`、狀態詞彙未校準(OWT-016):目的是讓這兩個狀態能被命名,不是在標準發布當天讓所有既有載體一次變紅。
|
|
351
|
+
|
|
352
|
+
---
|
|
353
|
+
|
|
354
|
+
## 沒有人能推導的事實,需要戳記、誠實的「未知」與看得見的年齡
|
|
355
|
+
|
|
356
|
+
開放工作裡有些事實,助理沒有任何來源可以讀:訊息是否已寄出、對方是否已回覆、某事是否已核准。**OWT-003** 只允許從可推導的來源生成,而下方的表格拒絕手寫狀態檔是有道理的——它會過期,而比過期內容新的戳是隱形的。這類事實沒有可推導的來源可供生成,把手寫列一律拒絕,就讓它們無處可住,最後流落到沒有任何東西會計數的自由文字裡。
|
|
357
|
+
|
|
358
|
+
因此 1.2.0 容許一個**窄例外**(**OWT-023**–**OWT-026**),附三個條件。那一列寫明**誰在何時看到**(`observed-by`、`observed-at`)。它的值是 `yes`、`no` 或 **`unknown`**,而 `unknown` 是正式的值,單獨計數、單獨顯示、永遠不算完成(**OWT-024**)——與 OWT-006 同一種讀法:看不到的不是零。而且報告顯示**每筆觀察的年齡**(**OWT-025**);超過宣告門檻的被回報為「觀察已舊」、另計、不算已確認。這個例外不及於版本控制、規格標記或 CI 結果已能決定的東西(**OWT-026**),因為 OWT-003 沒有被放寬。
|
|
359
|
+
|
|
360
|
+
這不是在宣稱整件事都誠實。戳只說有人在某時看到;**它不說現在仍然如此**,報告就是圍繞這一點建的:年齡印在值的旁邊,所以戳原本會藏起來的過期,是讀者第一眼看到的東西。而檢查能判定的只有可判定的部分:`observed-by` 有沒有填、`observed-at` 是不是一個真實的日子、值在不在值域內。**它判定不了 `observed-by` 填的是不是真的去看過的人,也判定不了那個值是否為真**——那是關於世界的宣稱,不是 artefact 之間的關係(OWT-014),本標準不假裝做得到。「觀察已舊」的門檻沒有任何量測支持(OWT-016)。
|
|
361
|
+
|
|
362
|
+
**這些嚴重度的理由(1.2.0)。** OWT-020 到 OWT-026 全是 `warning`:狀態詞、欄位名與「觀察已舊」的門檻是第一版詞彙,沒有任何量測支持(OWT-016),而違反的代價是讀的人的時間,不是工作本身。更嚴的等級,會讓所有早於這兩個狀態存在的載體在標準發布當天變紅;`warning` 讓採用者先看到清單。**但指令對任何違反都回結束碼 1,`warning` 也一樣**——`uds open-work next-action` 對 OWT-019 本來就是如此——所以把其中一支接進閘門的採用者,自己決定警告要不要擋。
|
|
363
|
+
|
|
364
|
+
---
|
|
365
|
+
|
|
366
|
+
## 對另一個專案的等待,只點名它,不定位它
|
|
367
|
+
|
|
368
|
+
彼此相依的專案會產生一種 OWT-002 從未為它而寫的等待:一個專案的項目,在等另一個專案的產出。1.3.0 讓解除條件可以點名它——`uds:v6.14.0` 或 `vibeops:docs/runbook.md`——並說明什麼東西能防止這種紀錄腐壞(**OWT-027**)。專案是**邏輯名稱**,絕不是目錄。目錄屬於一台機器、一個使用者(它會洩漏使用者名稱,換一台機器就是錯的),而同一份載體的兩個讀者,必須用同一個名稱指同一個專案。所以載體裡放的是名稱,「名稱對應到某台機器上的哪個目錄」則宣告在載體**之外**(參考檢查接受可重複的 `--root NAME=DIR`,或採用專案的設定)。解除條件裡出現機器專屬的絕對路徑即違反。一個在等另一個專案之物件的項目,沒有任何「可以被問」的請求,所以 OWT-020 的兩個狀態不適用於它;它仍被點名,並在自己那一行被計數。
|
|
369
|
+
|
|
370
|
+
檢查對這種等待的回報是四種(**OWT-028**),理由與 OWT-006 相同:**看不到的,不是零。** 已解除;尚未解除;**從這裡看不到**——對方專案不在這台機器上、或讀不到它的版本控制——它單獨計數,既不是已解除也不是零;以及**需要人判斷**——專案看得到,但裡面的測試名稱、指令或需求編號沒有任何檢查讀得到的東西。路徑與版本控制標籤是機器可觀察的:它在對方專案裡存在,或不存在。查找只讀這台機器上的檔案與版本控制、不開任何連線,而「已解除」只說該物件在那邊存在,絕不說它是對的那個物件(OWT-014)。
|
|
371
|
+
|
|
372
|
+
**OWT-029** 說這種紀錄放在哪裡:放在每個相關專案的 session 都讀得到、受版本控制的 repo 裡,而不是只放在某個人的 AI 工具記憶裡。那種記憶綁一個使用者、一台機器,另一個 session 與 CI 都看不到,也沒有給檢查任何具名的載體去判定關係(OWT-014)。用哪個 repo,是採用專案的選擇。本版涵蓋的是一個人在一台機器上的各個 session;工作如何在人與人之間共享、共享的紀錄要不要先審核才可信,這裡不決定。
|
|
373
|
+
|
|
374
|
+
**為什麼是 `warning`、什麼未校準(1.3.0)。** OWT-027 到 OWT-029 是 `warning`,理由與上面 OWT-020 到 OWT-026 相同。構成邏輯名稱的樣式、機器專屬路徑前綴的清單(家目錄與磁碟區目錄、磁碟機代號、`~/`),以及檢查能查找的兩種物件(路徑與標籤),都是第一版判斷(OWT-016);涵蓋率未知(OWT-011),因為用清單裡沒有的形式寫的路徑,不會被認成絕對路徑。
|
|
279
375
|
|
|
280
376
|
---
|
|
281
377
|
|
|
282
378
|
## What this standard deliberately does not adopt
|
|
283
379
|
|
|
284
|
-
The two additions
|
|
380
|
+
The two 1.1.0 additions (OWT-017–OWT-019: goal apart from progress with a revision record, and a next action that names an object) were prompted by a prompt a user forwarded, **whose author and provenance are unknown and which ships no implementation**. Only its design shapes were borrowed; none of its claims is cited here. The rest of what it proposes was examined and **not** adopted, for reasons about mechanism rather than taste:
|
|
285
381
|
|
|
286
382
|
| Not adopted | Why (mechanism) |
|
|
287
383
|
|---|---|
|
|
288
|
-
| A hand-written state file as the source of truth | A hand-written state goes stale, and a stamp newer than stale content is invisible (the failure OWT-004 and OWT-005 exist for). Saying "trust version control when they disagree" without a mechanism that reconciles the file with version control adopts a known-stale source. OWT-003 already requires derivable fields to be generated |
|
|
384
|
+
| A hand-written state file as the source of truth for any state a reliable source already determines | A hand-written state goes stale, and a stamp newer than stale content is invisible (the failure OWT-004 and OWT-005 exist for). Saying "trust version control when they disagree" without a mechanism that reconciles the file with version control adopts a known-stale source. OWT-003 already requires derivable fields to be generated. **Since 1.2.0 there is one narrow exception (OWT-023–OWT-026)**: a hand-written row for a fact nothing can derive — whether a message was sent, whether the other side replied. It is admitted only with a stamp of who saw it and when, a value that may honestly be `unknown`, and an age shown in every report, and it never reaches a field version control, a spec marker or a CI result determines. The objection above is answered, not set aside: it was that staleness is invisible, and here the age is printed beside every value |
|
|
289
385
|
| A fixed start-of-work ritual (read the files, then check version control, then verify) | The hand-off points are already governed: OWT-007 at turn end, and [turn-completion-integrity](turn-completion-integrity.md). A start-of-work ritual is configured in each agent tool's own instructions; writing it here yields a requirement no check over an artefact can decide, which OWT-014 excludes |
|
|
290
386
|
|
|
291
|
-
|
|
387
|
+
1.1.0 的兩項新增(OWT-017–OWT-019:目標與進度分開並留下修訂紀錄、以及點名對象的下一步),起因是使用者轉貼的一份提示詞——**作者與出處不明,也沒有任何實作**。
|
|
292
388
|
只借了它的設計形狀,本文不引用它的任何宣稱。它提出的其餘部分經過檢視、**沒有**採納,
|
|
293
389
|
理由是機制層的,不是口味:
|
|
294
390
|
|
|
295
391
|
| 不採納 | 理由(機制層) |
|
|
296
392
|
|---|---|
|
|
297
|
-
|
|
|
393
|
+
| 以手寫狀態檔作為任何「已有可靠來源可決定」之狀態的真相 | 手寫狀態會過期,而「戳比過期內容新」是隱形的(OWT-004、OWT-005 存在的起因)。只說「兩者不一致時以版本控制為準」,卻沒有任何機制讓該檔與版本控制對帳,就是採納一個已知會過期的來源。OWT-003 已經要求可推導的欄位改用生成。**自 1.2.0 起有一個窄例外(OWT-023–OWT-026)**:對「沒有任何東西能推導」的事實——訊息是否已寄出、對方是否已回覆——容許一列手寫。它只有在附上誰何時看到的戳記、值可以誠實地是 `unknown`、且每份報告都顯示年齡時才被容許,並且絕不及於版本控制、規格標記或 CI 結果已能決定的欄位。上面的反對理由是被回答了,不是被擱置:那個理由是過期看不見,而這裡年齡就印在每個值旁邊 |
|
|
298
394
|
| 固定的開工儀式(讀檔→查版本控制→驗證) | 交接點已被管住:回合結束有 OWT-007,另有 [turn-completion-integrity](turn-completion-integrity.md)。開工儀式要靠各代理工具自己的指示來設定;寫進這裡只會得到一條沒有任何 artefact 上的檢查判定得了的要求,而那正是 OWT-014 排除的東西 |
|
|
299
395
|
|
|
300
396
|
---
|
|
@@ -344,8 +440,14 @@ DEX-003 扮演的角色相同。上面每一條都指名了 artefact 與它們
|
|
|
344
440
|
| Triage status read only from prose wording | Correct until an item is phrased a way the wording list did not anticipate |
|
|
345
441
|
| An item that silently vanishes from the carrier | Indistinguishable from a bug that lost it |
|
|
346
442
|
| A goal or acceptance criterion edited with no record of who agreed | "Every criterion is met" stays true, of a different set of criteria; the edited goal reads as if it always said that |
|
|
347
|
-
| A hand-written state file treated as the source of truth | Goes stale, and a stamp newer than the stale content cannot be seen |
|
|
443
|
+
| A hand-written state file treated as the source of truth for what version control or CI already knows | Goes stale, and a stamp newer than the stale content cannot be seen |
|
|
348
444
|
| A next action of "continue implementation" or "handle the rest" | Names nothing to start from; indistinguishable from a forgotten item |
|
|
445
|
+
| A waiting item with no record of whether anyone has asked | A draft nobody sent and a request nobody answered look identical; waiting never releases the first |
|
|
446
|
+
| A hand-written "sent / replied / approved" with no name and no date | Cannot be told from a guess, and when it goes stale nothing shows it |
|
|
447
|
+
| An observation nobody could make, counted as "no" or as done | A fact that could not be seen is read as absent or complete; `unknown` is a value, not a zero |
|
|
448
|
+
| A wait on another project recorded with a directory of one machine | Wrong on the next machine, leaks the user's name, and two readers of one carrier no longer mean the same project |
|
|
449
|
+
| A project this machine cannot see, counted as released or as zero | What could not be looked at is read as done; "not visible from here" is its own count |
|
|
450
|
+
| Work that waits on another project kept only in one person's AI-tool memory | Another session and CI cannot read it, and no check has a named carrier to decide anything over |
|
|
349
451
|
|
|
350
452
|
| 反模式 | 為什麼會失敗 |
|
|
351
453
|
|---|---|
|
|
@@ -359,8 +461,14 @@ DEX-003 扮演的角色相同。上面每一條都指名了 artefact 與它們
|
|
|
359
461
|
| 分類狀態只靠散文措辭判讀 | 正確到某個項目用清單沒預料到的方式寫出來為止 |
|
|
360
462
|
| 項目從承載庫裡無聲消失 | 與一個弄丟它的 bug 無從分辨 |
|
|
361
463
|
| 目標或驗收條件被改了,卻沒有任何「誰同意」的紀錄 | 「每一條驗收都滿足」依然為真,只是針對另一組條件;被改過的目標讀起來像一直這麼寫 |
|
|
362
|
-
|
|
|
464
|
+
| 把手寫的狀態檔當成版本控制或 CI 早已知道之事的真相 | 會過期,而比過期內容新的戳看不見 |
|
|
363
465
|
| 下一步寫「繼續實作」或「處理剩下的」 | 沒有點名任何可以開始的東西;與被遺忘的項目無從分辨 |
|
|
466
|
+
| 等待項目沒有任何「有沒有人問過」的紀錄 | 沒人送出的草稿與沒人回的請求看起來一模一樣;等待永遠不會讓前者解除 |
|
|
467
|
+
| 手寫的「已寄出/已回覆/已核准」,沒有名字也沒有日期 | 與猜測無從分辨,過期時也沒有任何東西顯示 |
|
|
468
|
+
| 一個沒有人能做的觀察,被算成「否」或已完成 | 看不到的事實被讀成不存在或已完成;`unknown` 是一個值,不是零 |
|
|
469
|
+
| 對另一個專案的等待,用某一台機器的目錄記下來 | 換一台機器就是錯的、洩漏使用者名稱,同一份載體的兩個讀者也不再指同一個專案 |
|
|
470
|
+
| 這台機器看不到的專案,被算成已解除或零 | 沒看過的被讀成做完了;「從這裡看不到」是它自己的一個計數 |
|
|
471
|
+
| 等著另一個專案的工作,只放在某個人的 AI 工具記憶裡 | 另一個 session 與 CI 讀不到,檢查也沒有任何具名的載體可以判定 |
|
|
364
472
|
|
|
365
473
|
---
|
|
366
474
|
|
|
@@ -368,6 +476,12 @@ DEX-003 扮演的角色相同。上面每一條都指名了 artefact 與它們
|
|
|
368
476
|
|
|
369
477
|
**Nothing in UDS gates on it, and that is recorded rather than implied.** UDS states the relations a carrier of open work must satisfy; whether anything decides them is the adopting project's call, per the [writing constraint](#how-this-standard-is-written--and-why-it-is-written-that-way) above — the same boundary [deferred-item-exit](deferred-item-exit.md) draws for its own exits. Since 1.1.0 UDS does ship one **reference decision procedure** for OWT-017–OWT-019 — `uds open-work next-action | revision | separation` from the npm package (`uds open-work self-test` runs the checker's own arms; from a clone of the UDS repository `node scripts/check-open-work-tracking.mjs` runs the same code) — offered as evidence in the OWT-015 sense — it has been observed to fail against violating samples — for an adopter to run or to reimplement. It is not wired into any UDS release gate, because UDS carries no open-work carrier for it to check. For OWT-019 it reads a next-action field in three shapes, all through one vocabulary: a heading section, an inline label, and every row of a table column whose header is in that vocabulary. A table row whose cell count differs from its header is listed as undecidable (never read as empty; with no violation elsewhere the exit code is 2, not a pass), and an empty, `—`, `-` or done cell is counted and not evaluated — not a violation, because OWT-019 judges a next action that was written, and a missing one is a different failure it does not decide.
|
|
370
478
|
|
|
479
|
+
Since 1.2.0 the same module also ships `uds open-work waiting` (OWT-020–OWT-022) and `uds open-work observations` (OWT-023–OWT-026), with the same exit codes (0 no violation, 1 violation, 2 cannot decide). Today's date is injected (`--now`) and never read inside a rule, so a result does not drift with the day; `--stale-after` declares the threshold. Both read tables and list items through their structural field names, in English and in Chinese. A table with neither an `observed-by` nor an `observed-at` column is not read as an observation carrier, and a waiting item written in words the vocabulary does not hold is read as not waiting — so a clean pass covers only what was recognised, and every run says so (OWT-011). `uds open-work self-test` runs a violating and a satisfying sample for each of OWT-020–OWT-026, and mutation tests show that removing any one detection turns a test or the self-test red (OWT-015). Like the 1.1.0 checks, none of it is wired into a UDS release gate.
|
|
480
|
+
|
|
481
|
+
Since 1.3.0 `uds open-work waiting` also resolves a release condition that names another project (OWT-027–OWT-028): `--root NAME=DIR`, repeatable and alongside the single `--root DIR` it always took, or `open_work.projects` in `uds.project.yaml`, says where each project lives, and a project it cannot see is counted apart. `uds open-work next-action` and `separation` take `--next-action-word WORD` (repeatable), or `open_work.next_action_words`, for the words an adopting project's carriers use, and `uds open-work next-action` and `waiting` take `--command-word PROGRAM` (repeatable; `glab` and `dotnet` are built in, `go`, `make` and `sh` deliberately are not) or `open_work.command_words`, for the programs it reads as a command. `next-action` also says which directory a path is looked up under and where that came from, and splits `named-unresolved` into `path-missing` (looked up, not there) and `not-resolvable` (a command, a test name or a requirement identifier, which is never looked up). The checker itself reads no settings file; the `uds` command passes them in, so `node scripts/check-open-work-tracking.mjs` takes the flags only.
|
|
482
|
+
|
|
483
|
+
Since 1.4.0 `uds open-work next-action` reads each row's state through the same function `waiting` runs, with the same date, and reports a row that is `asked-awaiting` and complete under OWT-022 as `waiting-on-reply`: counted apart (`waiting-on-reply=N` is appended to the counts line only when N is not 0, so a carrier with no such row prints what it always printed), listed in its own section with its age, and not a violation. It takes `--now` (the date, never read inside a rule) and `--stale-after` (default 7 days, uncalibrated). **The exit code of `next-action` changes for a carrier that has such rows: a row that used to make it exit 1 as unnamed no longer does.** A carrier with no status column, or whose rows are in no such state, prints exactly what it printed in 1.3.0. `uds open-work self-test` carries a satisfying sample and violating samples (no `asked-at`, `asked-at` in the future, nothing waited for, no release, `not-yet-asked`, another state, a carrier with no status column, a heading section) for this outcome, and an arm that reads `waiting` and `next-action` over the same rows and requires them to agree.
|
|
484
|
+
|
|
371
485
|
**UDS 不對本標準設任何閘門,而這件事是被記錄的,不是被暗示的。** UDS 陳述一個承載開放工作的地方
|
|
372
486
|
必須滿足的關係;有沒有東西去判定它,依上面的[寫法約束](#how-this-standard-is-written--and-why-it-is-written-that-way),
|
|
373
487
|
是採用專案的決定——與 [deferred-item-exit](deferred-item-exit.md) 對自己出口劃的界線相同。
|
|
@@ -375,6 +489,12 @@ DEX-003 扮演的角色相同。上面每一條都指名了 artefact 與它們
|
|
|
375
489
|
作為 OWT-015 意義上的證據——它已被觀察到對違反的樣本回報失敗——供採用者直接執行或自行重做。
|
|
376
490
|
它沒有接進任何 UDS 發版閘門,因為 UDS 本身沒有承載開放工作的地方可供它檢查。
|
|
377
491
|
|
|
492
|
+
自 1.2.0 起,同一份程式還附上 `uds open-work waiting`(OWT-020–OWT-022)與 `uds open-work observations`(OWT-023–OWT-026),結束碼相同(0 無違反、1 違反、2 判定不了)。今天的日期是被注入的(`--now`),絕不在規則裡讀時鐘,所以結果不會隨日子漂移;`--stale-after` 宣告門檻。兩者都以結構欄位名讀表格與清單項目,中英文皆可。沒有 `observed-by` 也沒有 `observed-at` 欄的表格不被讀成觀察載體;用詞彙裡沒有的字寫成的等待項目,被讀成「不是等待」——所以乾淨的通過只涵蓋被辨認出來的部分,而每次執行都這麼說(OWT-011)。`uds open-work self-test` 對 OWT-020–OWT-026 每一條各跑一個違反的樣本與一個符合的樣本,突變測試顯示拿掉任何一個偵測都會讓某個測試或自測變紅(OWT-015)。與 1.1.0 的檢查相同,它都沒有接進任何 UDS 發版閘門。
|
|
493
|
+
|
|
494
|
+
自 1.3.0 起,`uds open-work waiting` 也會解析點名另一個專案的解除條件(OWT-027–OWT-028):`--root NAME=DIR`(可重複,並與它一直接受的單一 `--root DIR` 並存),或 `uds.project.yaml` 的 `open_work.projects`,說明每個專案在哪裡;看不到的專案被單獨計數。`uds open-work next-action` 與 `separation` 接受 `--next-action-word WORD`(可重複),或 `open_work.next_action_words`,放採用專案的載體所用的詞;`uds open-work next-action` 與 `waiting` 接受 `--command-word 程式名`(可重複;`glab` 與 `dotnet` 內建,`go`、`make`、`sh` 刻意不內建),或 `open_work.command_words`,放它讀成指令的程式。`next-action` 另外會說路徑是在哪個目錄下查找的、那個目錄從哪來,並把 `named-unresolved` 分成 `path-missing`(查過、不在)與 `not-resolvable`(指令、測試名稱、需求編號,從不查找)。檢查本體不讀任何設定檔,是 `uds` 指令把設定傳進去,所以 `node scripts/check-open-work-tracking.mjs` 只接受旗標。
|
|
495
|
+
|
|
496
|
+
自 1.4.0 起,`uds open-work next-action` 以 `waiting` 所用的同一個函式、同一個日期讀每一列的狀態,把處於 `asked-awaiting` 且滿足 OWT-022 的列回報為 `waiting-on-reply`:單獨計數(只有在數量不是 0 時才在計數行尾附上 `waiting-on-reply=N`,所以沒有這種列的載體印出的與過去一模一樣)、在自己的一節裡列出並顯示年齡、不算違反。它接受 `--now`(日期,絕不在規則裡讀時鐘)與 `--stale-after`(預設 7 天,未校準)。**對含有這種列的載體,`next-action` 的結束碼會改變:過去因「未點名」而回 1 的列,現在不再讓它回 1。**沒有狀態欄的載體,或列不在這種狀態的載體,印出的與 1.3.0 完全相同。`uds open-work self-test` 對這個結果附上一個符合的樣本與多個違反的樣本(沒有 `asked-at`、`asked-at` 在未來、沒寫在等什麼、沒有解除事件、`not-yet-asked`、其他狀態、沒有狀態欄的載體、小節標題),以及一個同時讀 `waiting` 與 `next-action`、要求兩者一致的自測臂。
|
|
497
|
+
|
|
378
498
|
What this standard does do is make that call visible: OWT-014 guarantees every requirement here **can** be decided, OWT-015 fixes what it takes for a decision to count, and OWT-005/OWT-011 fix what a partial decision is allowed to print.
|
|
379
499
|
|
|
380
500
|
本標準做的事,是讓那個決定顯形:OWT-014 保證這裡每一條**能**被判定,OWT-015 固定
|
|
@@ -396,6 +516,12 @@ This standard's shape comes from one adopting project's observations made and ac
|
|
|
396
516
|
|
|
397
517
|
**1.1.0's additions (OWT-017–OWT-019)** come from two gaps found in one adopting project on 2026-09-29 (DEC-122): the standard said nothing about separating a work item's goal from its progress, and an acceptance criterion in one of that project's specifications was revised mid-work with no record of who agreed. The design shapes were borrowed from a prompt a user forwarded, author unknown (see [What this standard deliberately does not adopt](#what-this-standard-deliberately-does-not-adopt)). The reference procedure is hours old, has one author, and has run against constructed samples, not against a real backlog of revisions. Under OWT-016, everything it uses that resembles a threshold is **uncalibrated, an initial judgment**: the heading vocabulary that marks a section as intent, progress, next action, or revision record; the list of command names it recognises; the file-extension list; and the requirement-identifier pattern. None of them was measured against real usage, and a project should pass its own.
|
|
398
518
|
|
|
519
|
+
**1.2.0's additions (OWT-020–OWT-026)** come from one adopting project's test report of 6.14.0-beta.5 on 2026-10-07 (XSPEC-459): a message drafted and never sent could not be told from one sent and unanswered, and a fact the assistant could not observe had no sanctioned place to be written. The reference procedure is hours old, has one author, and has run against constructed samples, not against a real backlog. Under OWT-016, everything it uses that resembles a threshold is **uncalibrated, an initial judgment**: the state words and field names in English and Chinese, the list of subjects treated as derivable, the 7-day default for "stale", and the reading of dates (a year is required; the time of day is ignored). None of it was measured against real usage, and a project's own words and threshold should replace the defaults.
|
|
520
|
+
|
|
521
|
+
**1.3.0's additions (OWT-027–OWT-029)** come from the same adopting project on 2026-10-07 (XSPEC-460 and XSPEC-461): items in one project wait for what another project produces, and the only record of how was an ad-hoc convention; and a user whose carrier kept one table per status got exit 2 from `next-action` without being told why. Under OWT-016, everything the new look-up uses that resembles a threshold is **uncalibrated, an initial judgment**: the pattern for a logical name, the machine-specific path prefixes, the two object kinds it can look up (a path and a tag), and the next-action words, both the built-in list and its refusal to include `待辦`, `後續`, `Next`, `TODO` and `Action`. None of it was measured against real usage.
|
|
522
|
+
|
|
523
|
+
**1.4.0's addition (no new requirement identifier)** comes from the same adopting project's test of 6.14.0-beta.5 on 2026-10-07 (XSPEC-464): 14 of its 17 next-action fields were "wait for the other side's reply", "wait to schedule" or "wait to revoke"; `waiting` accepted those rows and `next-action` rejected the same rows; and the project declined to rewrite its carrier only to pass the check. Under OWT-016, what the new outcome uses that resembles a threshold is **uncalibrated, an initial judgment**: the 7-day stale threshold is the default OWT-025 already carries, reused without measurement, and the state words are those OWT-020 reads. The outcome is only as good as the carrier's records: a carrier that states a release condition but no asked date is still judged by OWT-019 until it records the date.
|
|
524
|
+
|
|
399
525
|
- **OWT-001 的「不超過兩個欄位」**與**OWT-012 的「過了宣告的門檻」**(在原始觀察中以兩週為例)
|
|
400
526
|
依 OWT-016 是**初始判斷,不是量測結果**——兩個欄位跟三個欄位、兩週跟四週的未分類門檻,
|
|
401
527
|
目前都沒有對照比較過。
|
|
@@ -411,6 +537,12 @@ This standard's shape comes from one adopting project's observations made and ac
|
|
|
411
537
|
或修訂紀錄的標題詞彙;它認得的指令名清單;副檔名清單;需求編號的樣式。
|
|
412
538
|
沒有任何一項對照過真實使用量測,採用專案應傳入自己的。
|
|
413
539
|
|
|
540
|
+
**1.2.0 的新增(OWT-020–OWT-026)**來自一個採用專案在 2026-10-07 對 6.14.0-beta.5 的測試回報(XSPEC-459):一則草擬了從沒送出的訊息,與一則已送出、沒人回的訊息分不出來;助理無法觀察的事實,沒有一個被認可的地方可以寫。那支參考判定程序只有幾小時大、只有一位作者,跑過的是人造樣本,不是真實的待辦。依 OWT-016,它用到的一切類似閾值的東西都是**未校準、初始判斷**:中英文的狀態詞與欄位名、被當成「可推導」的主題清單、「觀察已舊」預設的 7 天、以及日期的讀法(必須有年份;時間不看)。沒有任何一項對照過真實使用量測,採用專案應以自己的詞彙與門檻取代預設。
|
|
541
|
+
|
|
542
|
+
**1.3.0 的新增(OWT-027–OWT-029)**來自同一個採用專案在 2026-10-07 的工作(XSPEC-460 與 XSPEC-461):一個專案的項目在等另一個專案的產出,而當時唯一的紀錄方式是臨時約定;另有一位使用者的載體一個狀態一張表,`next-action` 回了 2,卻沒有說為什麼。依 OWT-016,新的查找所用到的一切類似閾值的東西都是**未校準、初始判斷**:構成邏輯名稱的樣式、機器專屬的路徑前綴、它能查找的兩種物件(路徑與標籤),以及「下一步」詞——內建清單,與它拒絕收進 `待辦`、`後續`、`Next`、`TODO`、`Action` 的決定。沒有任何一項對照過真實使用量測。
|
|
543
|
+
|
|
544
|
+
**1.4.0 的新增(沒有新的要求編號)**來自同一個採用專案對 6.14.0-beta.5 的測試(2026-10-07,XSPEC-464):它的 17 個下一步欄位中有 14 個是「等對方回信」「等排時間」或「等撤銷」;`waiting` 接受這些列、`next-action` 卻拒絕同一批列;而該專案拒絕只為了讓檢查通過而改寫自己的載體。依 OWT-016,新結果用到的一切類似閾值的東西都是**未校準、初始判斷**:「已舊」的 7 天是 OWT-025 本來就帶的預設,未經量測就沿用,而狀態詞就是 OWT-020 讀的那些。這個結果只和載體的紀錄一樣好:只寫了解除條件、沒寫問的日期的載體,在補上日期之前仍由 OWT-019 判定。
|
|
545
|
+
|
|
414
546
|
---
|
|
415
547
|
|
|
416
548
|
## Relationship to other standards
|
|
@@ -420,6 +552,10 @@ This standard's shape comes from one adopting project's observations made and ac
|
|
|
420
552
|
- [class-level-fix](class-level-fix.md) — the general form of the wording-list limit OWT-011 discloses, and the source of the non-vacuous-evidence procedure OWT-015 requires.
|
|
421
553
|
- [verification-evidence](verification-evidence.md) — the source of the exit-code and evidence-validity reasoning OWT-015 depends on; also where a partial-coverage exception (OWT-006, OWT-011) is registered rather than merely disclosed once.
|
|
422
554
|
- OWT-018 attaches to the same hand-back point as OWT-007: an edit to intent with no approver is one more thing listed there, and, like everything listed there, never blocks.
|
|
555
|
+
- OWT-020–OWT-022 refine OWT-002 (a waiting item states what it waits for and its release) by saying which kind of wait it is; OWT-023–OWT-026 sit beside OWT-003 and OWT-006 — the first says what may be generated and what may not, the second that what could not be seen is never counted as zero.
|
|
556
|
+
- OWT-027–OWT-029 extend OWT-002 across a project boundary (a release condition may be another project's object) and OWT-006 with it: a project this machine cannot see is counted apart, never as released and never as zero.
|
|
557
|
+
|
|
558
|
+
- OWT-019 and OWT-022 meet at one point (1.4.0): an item in `asked-awaiting` that satisfies OWT-022 is judged by OWT-022, and its next step may be the reply it waits for; every other next action is judged by OWT-019 as before.
|
|
423
559
|
|
|
424
560
|
- [deferred-item-exit](deferred-item-exit.md) — 同一個形狀的上游一半:DEX 要求延後項目離開文件、
|
|
425
561
|
抵達可追蹤的出口,並刻意不規定出口的載體。本標準接手**出口存在之後**的事,
|
|
@@ -434,3 +570,7 @@ This standard's shape comes from one adopting project's observations made and ac
|
|
|
434
570
|
而不是揭露一次就放著。
|
|
435
571
|
- OWT-018 掛在與 OWT-007 相同的交回點:沒有核可者的意圖修改,是在那裡多列出來的一項,
|
|
436
572
|
而且與列在那裡的一切相同,永不阻斷。
|
|
573
|
+
- OWT-020–OWT-022 細化 OWT-002(等待項目說明在等什麼與解除條件),說明那是哪一種等待;OWT-023–OWT-026 與 OWT-003、OWT-006 並列——前者說什麼可以生成、什麼不可以,後者說看不到的絕不被算成零。
|
|
574
|
+
- OWT-027–OWT-029 把 OWT-002 延伸過專案的邊界(解除條件可以是另一個專案的物件),並連同 OWT-006:這台機器看不到的專案被單獨計數,絕不算成已解除、也絕不算成零。
|
|
575
|
+
|
|
576
|
+
- OWT-019 與 OWT-022 在一點上交會(1.4.0):處於 `asked-awaiting` 並滿足 OWT-022 的項目由 OWT-022 判定,它的下一步可以就是它等的那個回覆;其他每一個下一步照舊由 OWT-019 判定。
|
|
@@ -1,5 +1,8 @@
|
|
|
1
1
|
# Pipeline Security Gates(CI Pipeline 安全檢查點)
|
|
2
2
|
|
|
3
|
+
**Version**: 1.1.0
|
|
4
|
+
**Last Updated**: 2026-10-07
|
|
5
|
+
|
|
3
6
|
## 概述
|
|
4
7
|
|
|
5
8
|
本標準定義嵌入 CI pipeline 各階段的安全檢查點,涵蓋 SAST、DAST、SCA(含 SBOM)、機密掃描,以及發布前的供應鏈簽章驗證,並明確規定各類發現的阻斷(Block)、警告(Warn)、記錄(Log)行為。
|
|
@@ -9,6 +12,7 @@
|
|
|
9
12
|
## 核心原則
|
|
10
13
|
|
|
11
14
|
- **機密掃描在 pre-commit**:所有包含機密的提交都必須被阻擋,不得例外
|
|
15
|
+
- **機密掃描器自己的行為也要寫明**:輸出不得印出被匹配到的機密值;掃描器跑不起來視同閘門失敗(見 1.1 節)
|
|
12
16
|
- **SAST 在建置後**:Critical 和 High 等級發現阻斷 pipeline
|
|
13
17
|
- **SCA + SBOM 在封包階段**:追蹤相依套件風險並產生物料清單
|
|
14
18
|
- **Attestation 驗證在發布前**:封包後、部署前 MUST 驗證 artifact 的 checksum、簽章與 SLSA provenance(消費 supply-chain-attestation 標準的產出,不得「產生 SBOM 卻不驗」)
|
|
@@ -31,6 +35,65 @@
|
|
|
31
35
|
|
|
32
36
|
**重要**:此閘門必須在程式碼進入版本控制前執行,一旦機密進入 git 歷史便需要完整的機密輪換程序。
|
|
33
37
|
|
|
38
|
+
### 1.1 掃描器自己的三個行為(1.1.0 新增)
|
|
39
|
+
|
|
40
|
+
上表講「掃什麼、什麼時候阻斷」;本節講**掃描器本身**必須怎麼表現。以下三條都以本標準既有的兩條原則為前提:
|
|
41
|
+
「安全閘門失敗 = pipeline 失敗」與「禁止靜默跳過安全閘門」。
|
|
42
|
+
|
|
43
|
+
| 編號 | 強度 | 要求 |
|
|
44
|
+
|------|------|------|
|
|
45
|
+
| **PSG-1** | 必須(MUST) | 掃描的輸出——終端機、CI 紀錄、報告檔——**不得包含被匹配到的機密值本身**,只能指出檔案、行號與規則名稱 |
|
|
46
|
+
| **PSG-2** | 必須(MUST) | 掃描器**無法執行**視同閘門失敗,**必須阻斷**提交/pipeline,不得放行 |
|
|
47
|
+
| **PSG-3** | 建議(SHOULD) | 團隊**自己撰寫**的偵測規則,每一條都有一個該被抓到(紅)與一個不該被誤報(綠)的測試樣本,樣本在測試中執行 |
|
|
48
|
+
|
|
49
|
+
#### PSG-1 輸出不得印出被匹配到的機密值
|
|
50
|
+
|
|
51
|
+
理由:掃描紀錄會被貼進聊天室與工單。閘門把機密值印在輸出裡,等於把已被擋下的機密再洩漏一次。
|
|
52
|
+
|
|
53
|
+
工具註記(只寫已查證的):
|
|
54
|
+
|
|
55
|
+
| 工具 | 註記 | 依據 |
|
|
56
|
+
|------|------|------|
|
|
57
|
+
| gitleaks | 官方 README 的範例輸出(`gitleaks git -v`,未加 `--redact`)含 `Secret:` 一行,印出被匹配到的值全文;`--redact` 旗標的說明為「redact secrets from logs and stdout」。README 未說明 `--redact` 是否也涵蓋 `--report-path` 產生的報告檔,**採用者須自行確認** | gitleaks 官方 README,2026-10-07 讀取 |
|
|
58
|
+
| detect-secrets | 預設輸出行為**本標準未查證,採用者須自行確認** | — |
|
|
59
|
+
| trufflehog | 預設輸出行為**本標準未查證,採用者須自行確認** | — |
|
|
60
|
+
|
|
61
|
+
判定程序(示意,三行):
|
|
62
|
+
|
|
63
|
+
1. 建一個含**假**密鑰的檔案(示意記法:`FAKE-SECRET-DO-NOT-USE-0000`;實際樣本須是**你自己的閘門會匹配的格式**——例如自寫一條 canary 規則,或工具文件列出的測試樣式——且必須是假值,不得是任何真實服務的有效憑證),加入 staging 後執行閘門,**先確認閘門確實阻斷**。
|
|
64
|
+
2. 沒阻斷代表樣本沒被匹配,第 3 步不算數——此時輸出裡當然找不到它,結果是「綠得虛假」。
|
|
65
|
+
3. 在終端機、CI 紀錄、報告檔的**全部輸出**搜尋該假密鑰字串;找到任何一處即不合規。
|
|
66
|
+
|
|
67
|
+
#### PSG-2 掃描器跑不起來視為閘門失敗
|
|
68
|
+
|
|
69
|
+
「跑不起來」包含:執行檔不存在、設定檔無法讀取或無效、逾時、非預期的結束。這些情況必須阻斷,**不得**以 `|| true`、`2>/dev/null` 等方式吞掉,也不得讓閘門輸出「掃描通過」。
|
|
70
|
+
|
|
71
|
+
UDS 自己的先例:`uds init` 為非 Node 專案寫的原生 `.git/hooks/pre-commit`,曾把每個檢查接上 `2>/dev/null || true` 後印出「Pre-commit checks passed」——檢查壞掉也照樣放行,且藏起自己的錯誤。6.14.0-beta.4 修正(見 CHANGELOG 該版〈Fixed〉)。
|
|
72
|
+
|
|
73
|
+
工具註記:gitleaks 官方 README 的〈Exit Codes〉一節寫 `1 - leaks or error encountered`——找到洩漏與發生錯誤是**同一個結束碼**,所以**區分原因要靠輸出,不靠結束碼**;對阻斷與否,兩者的結論相同。README 另說明 `--exit-code` 旗標可改寫「找到洩漏時」的結束碼(2026-10-07 讀取):把它設成 0 會讓閘門在找到洩漏時放行,不得這樣設定。detect-secrets、trufflehog 的結束碼行為**本標準未查證,採用者須自行確認**。
|
|
74
|
+
|
|
75
|
+
判定程序(示意,三行):
|
|
76
|
+
|
|
77
|
+
1. 把掃描器執行檔改名或移走,或把它的設定檔改成無效內容。
|
|
78
|
+
2. 對一個**不含任何機密**的提交執行閘門(此時若被阻斷,原因只可能是掃描器跑不起來)。
|
|
79
|
+
3. 提交被阻斷(結束碼非 0),且輸出說明是掃描器無法執行;若提交被放行,或輸出看起來像「掃描通過」,即不合規。
|
|
80
|
+
|
|
81
|
+
#### PSG-3 自己寫的偵測規則,每條要有紅綠樣本
|
|
82
|
+
|
|
83
|
+
適用於團隊**自己撰寫**的規則,包含 `secret-management-standards` 的 REQ-003 所列的最低四種:AWS access key、PEM 私鑰標頭、一般 API token、含密碼的連線字串。只用工具內建規則、不自己寫規則的團隊不適用。
|
|
84
|
+
|
|
85
|
+
道理與 `open-work-tracking` 的 OWT-015 相同:沒見過失敗的檢查不算證據。一條從沒被紅樣本匹配過的規則,無法證明它抓得到東西;一條沒有綠樣本的規則,無法證明它不會把正常程式碼全擋下來。
|
|
86
|
+
|
|
87
|
+
判定程序(示意,三行):
|
|
88
|
+
|
|
89
|
+
1. 列出團隊自寫的全部偵測規則。
|
|
90
|
+
2. 逐條找對應的一個紅樣本(必被匹配)與一個綠樣本(不得匹配);樣本須是假值。
|
|
91
|
+
3. 確認樣本在測試中真的執行;缺任一個即列為缺口(本條為建議,缺口不阻斷)。
|
|
92
|
+
|
|
93
|
+
#### 範圍的誠實註明
|
|
94
|
+
|
|
95
|
+
UDS **不出貨掃描器**。上面三段是採用者在**自己的閘門**上自行驗證的程序。UDS 自己的 repo 目前也沒有機密掃描(CI 與 hook 都沒有用到上述三種工具)。
|
|
96
|
+
|
|
34
97
|
---
|
|
35
98
|
|
|
36
99
|
### 2. Post-Build — SAST(靜態應用安全測試)
|
|
@@ -2,8 +2,8 @@
|
|
|
2
2
|
|
|
3
3
|
> **Language**: English | [繁體中文](../locales/zh-TW/core/project-context-memory.md)
|
|
4
4
|
|
|
5
|
-
**Version**: 1.2.
|
|
6
|
-
**Last Updated**: 2026-
|
|
5
|
+
**Version**: 1.2.1
|
|
6
|
+
**Last Updated**: 2026-10-07
|
|
7
7
|
**Applicability**: All software projects using AI assistants
|
|
8
8
|
**Scope**: uds-specific
|
|
9
9
|
|
|
@@ -232,10 +232,18 @@ We use **Split-File Storage** (one Markdown file per topic) rather than a single
|
|
|
232
232
|
|
|
233
233
|
---
|
|
234
234
|
|
|
235
|
+
## Boundary: What This Standard Does Not Hold
|
|
236
|
+
|
|
237
|
+
- **Personal, cross-project, identifiable facts** (who owns what, how to reach a particular environment) are not project context: entries here live in the repository and are shared with everyone who clones it. Keep them in the user's own AI tool memory, not in `.project-context/`.
|
|
238
|
+
- **Coordination between projects** (which piece of work is waiting on which other) is open work, not context: see [Open Work Tracking](open-work-tracking.md) (OWT-027–OWT-029).
|
|
239
|
+
|
|
240
|
+
---
|
|
241
|
+
|
|
235
242
|
## Version History
|
|
236
243
|
|
|
237
244
|
| Version | Date | Changes |
|
|
238
245
|
|---------|------|---------|
|
|
246
|
+
| 1.2.1 | 2026-10-07 | Added: a Boundary section — personal, cross-project, identifiable facts and coordination between projects do not belong in the repository's project context (XSPEC-460 R5; patch, no behavioral change) |
|
|
239
247
|
| 1.2.0 | 2026-06-18 | Added: rationale + configurable note for the 7-day staleness threshold (XSPEC-292 T8) |
|
|
240
248
|
| 1.1.0 | 2026-03-16 | Add `workflow-state` type for cross-session state tracking |
|
|
241
249
|
| 1.0.0 | 2026-02-09 | Initial standard definition |
|