universal-dev-standards 6.8.0 → 6.9.0

package/bundled/ai/standards/reverse-engineering-standards.ai.yaml

@@ -3,8 +3,8 @@
 
 id: reverse-engineering
 meta:
-  version: "1.2.0"
-  updated: "2026-06-27"
+  version: "1.3.0"
+  updated: "2026-08-20"
   source: core/reverse-engineering-standards.md
   description: Standards for reverse engineering code into structured specifications
 
@@ -54,6 +54,49 @@ workflow:
         - "PHP-specific: flag loose-comparison / type-juggling (==, truthy/falsy of \"0\"/\"\"/null/0, string-to-number coercion) — no equivalent under strongly-typed targets"
       gate: pre-flight; unanswered three-question field marked not_implemented and blocks cutover
 
+    - stage: AUTH_SCOPE_EXTRACTION
+      description: >
+        Extract every legacy query predicate that limits results by the OPERATOR'S OWN
+        identity (tenant / org / dept / owner boundary) and verify the ported code has an
+        equivalent predicate — not merely an equivalent output shape.
+      output: Self-scoping predicate inventory, each locked by a negative cross-tenant test
+      certainty: "[Confirmed]/[Inferred]/[Unknown]"
+      position: alongside IMPLICIT_RULE_SCAN — after CODE_SCAN/TEST_ANALYSIS, before GAP_IDENTIFICATION
+      why_its_own_stage:
+        - "The security-critical line looks like a filter: the boundary lives in a WHERE clause, not in the request/response contract"
+        - "Shape-based assertions cannot see scope: '200 + non-empty list' passes identically with the scoping clause deleted"
+        - "A shared scope resolver existing elsewhere is not evidence THIS call site uses it"
+      derive:
+        - "Direct self-binding: WHERE <col> = :currentUser / :uid / $operatorId / session.user_id"
+        - "Tenant/org/owner columns (tenant_id, org_id, master_account, owner_id, dept_id, company_id) compared against a caller-derived value"
+        - "Subquery resolving the caller's own scope: WHERE dept_id = (SELECT dept_id FROM member WHERE account = :uid)"
+        - "Role-branched query construction — enumerate EVERY branch, not just the happy path"
+        - "Scope applied outside SQL: ORM global scopes, query-builder mixins, repository base classes, row-level-security policies, filter-injecting middleware"
+      derive_warning: >
+        A grep over SQL alone reports a SMALLER set than exists, because a predicate can be
+        enforced by something the SELECT never mentions — and a smaller set here reads exactly
+        like a safer one.
+      record_per_match:
+        - "Location (file:line in legacy source)"
+        - "The exact predicate text"
+        - "Self-binding: which caller-derived value it resolves to, and how that value is obtained"
+        - "Triggering branch: which role / parameter / code path reaches it"
+        - "Certainty [Confirmed] (file:line required) / [Inferred] / [Unknown]"
+      oracle_predicate_equivalence:
+        - "A literal equivalent exists in the ported code — same field, same notion of 'self'. A different field that happens to return the same rows today is a coincidence with an expiry date"
+        - "Every role branch is covered. Collapsing five role branches into one coarse 'is this an administrator?' check replaces a boundary with a permission"
+        - "THIS call site invokes the shared scope resolver, if one exists. The utility existing elsewhere is not evidence about this endpoint"
+      mandatory_negative_test: >
+        Operator A issues the request; the assertion is that operator B's records are ABSENT
+        from the response, with A and B both valid and authenticated. The assertion must be
+        about ABSENCE — a 200 / non-empty-list / matching-schema assertion is not sufficient
+        evidence and must not be counted as coverage. Self-check: delete the scoping predicate
+        in a scratch copy and re-run; if the test still passes it is not testing scope.
+      gate: >
+        Pre-flight for extraction, pre-UAT for the negative tests. A self-scoping predicate
+        whose ported equivalent is unverified, or which has no negative test, is marked
+        not_implemented and blocks cutover.
+
     - stage: GAP_IDENTIFICATION
       description: List unknowns requiring human input
       output: Gap analysis document
@@ -284,6 +327,32 @@ rules:
       is marked not_implemented and blocks cutover.
     priority: required
 
+  - id: auth-scope-extraction
+    trigger: reverse engineering or migrating a system whose data layer is multi-tenant, role-scoped, or owner-scoped
+    instruction: >
+      Run AUTH_SCOPE_EXTRACTION alongside IMPLICIT_RULE_SCAN. Mechanically derive every legacy
+      query predicate that bounds results by the operator's own identity — direct self-bindings,
+      tenant/org/owner columns compared to a caller-derived value, subqueries resolving the
+      caller's scope, every branch of role-branched query construction, and scope enforced
+      outside SQL (ORM global scopes, repository base classes, RLS policies, filter-injecting
+      middleware). Record each with file:line, the exact predicate, its self-binding, the
+      triggering branch, and a certainty tag; route [Unknown] to gap analysis, never assume
+      absent. For each, verify the ported code has a LITERAL equivalent (same field, same notion
+      of self) on EVERY branch, and that this specific call site invokes the shared scope
+      resolver if one exists — the resolver existing elsewhere is not evidence about this
+      endpoint.
+    priority: required
+  - id: auth-scope-negative-test
+    trigger: locking a confirmed self-scoping predicate as a verifiable oracle
+    instruction: >
+      Require a negative test: operator A's request must NOT return operator B's records, with
+      A and B both valid and authenticated. The assertion must be about the ABSENCE of the other
+      operator's records. A "200 with data" or matching-schema assertion is not sufficient
+      evidence and must not be counted as coverage — it passes identically with the scoping
+      clause deleted. Any predicate with an unverified ported equivalent or no negative test is
+      marked not_implemented and blocks cutover.
+    priority: required
+
   - id: human-review-required
     trigger: completing spec generation
     instruction: Always mark specification as requiring human review; [Unknown] sections MUST be filled by humans
@@ -320,6 +389,7 @@ quick_reference:
       - [Code Scan, Technical inventory, "[Confirmed]"]
       - [Test Analysis, Draft AC, "[Confirmed]/[Inferred]"]
       - [Implicit Rule Scan, Non-HTTP implicit-rule inventory, "[Confirmed]/[Inferred]/[Unknown]"]
+      - [Auth Scope Extraction, Self-scoping predicate inventory + negative tests, "[Confirmed]/[Inferred]/[Unknown]"]
       - [Gap ID, Gap analysis, "[Unknown] items"]
       - [Spec Gen, Draft SPEC, Mixed]
       - [Human Review, Validated spec, "[Confirmed]"]
@@ -341,6 +411,7 @@ quick_reference:
       - ["/reverse-tdd", Analyze test coverage against BDD]
 
 related_standards:
+  - test-completeness-dimensions.md   # Dimension 4 names the cross-tenant test; AUTH_SCOPE_EXTRACTION decides which ones you need
   - anti-hallucination.md
   - spec-driven-development.md
   - behavior-driven-development.md

package/bundled/ai/standards/security-standards.ai.yaml

@@ -153,7 +153,7 @@ enforcement:
   trigger: PreToolUse
   matcher:
     tool: Bash
-  script_ref: "scripts/hooks/check-dangerous-cmd.js"
-  hook_script: "scripts/hooks/check-dangerous-cmd.js"
+  script_ref: "scripts/hooks/check-dangerous-cmd.mjs"
+  hook_script: "scripts/hooks/check-dangerous-cmd.mjs"
   severity: error
   timeout_ms: 500

package/bundled/ai/standards/spec-driven-development.ai.yaml

@@ -3,8 +3,8 @@
 
 id: spec-driven-development
 meta:
-  version: "1.4.0"
-  updated: "2026-08-12"
+  version: "1.5.0"
+  updated: "2026-08-24"
   source: methodologies/guides/sdd-guide.md
   description: Spec-Driven Development workflow where documentation precedes implementation
 
@@ -254,3 +254,15 @@ quick_reference:
       - [OpenSpec, "/openspec proposal, /openspec approve"]
       - [Spec Kit, "/spec create, /spec close"]
       - [Manual, "specs/SPEC-XXX.md"]
+
+# ── Deferred items produced here ─────────────────────────────────────────────
+# Pointer, not a summary: the exit invariant is stated once, in deferred-item-exit.
+deferred_items:
+  produced_here:
+    - "out-of-scope / not-in-this-version entries"
+    - "open questions"
+    - "assumptions awaiting confirmation"
+  governed_by: deferred-item-exit
+  note: >
+    規則只在 core/deferred-item-exit.md 一處陳述；此處放指標不放摘要。
+    摘要會腐壞，指標不會。

package/bundled/ai/standards/tech-debt-standards.ai.yaml

@@ -3,10 +3,10 @@
 
 id: tech-debt-standards
 meta:
-  version: "1.0.0"
-  updated: "2026-03-31"
+  version: "1.1.0"
+  updated: "2026-08-20"
   source: core/tech-debt-standards.md
-  description: Tech debt taxonomy, registry, budget, impact matrix, and metrics
+  description: Tech debt taxonomy, registry, budget, impact matrix, metrics, and overdue handling
 
 taxonomy:
   design: "Architecture/design decisions — impacts maintainability, scalability"
@@ -34,6 +34,50 @@ registry_fields:
   - "Target resolution date"
   - "Status (Open→Scheduled→In Progress→Resolved→Verified)"
 
+registry_storage_options:
+  # Each option is legal ONLY if it also satisfies its overdue-check condition.
+  options:
+    - option: "docs/tech-debt-registry.md"
+      best_for: "Small teams, simple tracking"
+      overdue_check_condition: "A repository check parses the table and fails on rows past Target Resolution Date — OR an Unattended Declaration is present"
+    - option: "Issue tracker (GitHub Issues, Jira)"
+      best_for: "Larger teams, workflow integration"
+      overdue_check_condition: "A scheduled due-date query that actually RUNS and raises; a saved filter nobody opens is not a check — OR an Unattended Declaration"
+    - option: "Dedicated spreadsheet"
+      best_for: "Non-technical stakeholders"
+      overdue_check_condition: "Only if exported on a stated cadence to a location a check reads — OR an Unattended Declaration"
+  rationale: >
+    A registry no program can read satisfies every other requirement (11 fields, Owner,
+    Target Resolution Date). Nothing is wrong with it until the date passes and nothing
+    happens. The spreadsheet option is deliberately kept; what is forbidden is any storage
+    option being a place where dates expire unobserved.
+
+overdue_handling:
+  principle: "A Target Resolution Date that passes with no consequence is indistinguishable from no date at all"
+  legal_dispositions:
+    resolve:
+      meaning: "Do the work"
+      trace: "Registry entry closed + `Tech-Debt: TD-NNN resolved` commit footer"
+    withdraw:
+      meaning: "Decide not to do it, and delete the record"
+      trace: "Entry marked withdrawn with a reason and a date"
+    extend:
+      meaning: "Set a new Target Resolution Date"
+      trace: "New date AND a written reason recorded together; earlier dates stay visible (append, never overwrite)"
+  no_fourth_disposition: "'Still open, date passed, nobody looked' is not a permitted state"
+  registry_states:
+    checked: "A named, runnable check reads the registry and fails on items past Target Resolution Date"
+    unattended: "No such check exists; an Unattended Declaration naming owner + review cadence + last-review date is present"
+    neither: "Non-compliant"
+  unattended_declaration_template: >
+    Unattended — no automated check reads this registry for overdue items.
+    Reviewed manually by <owner> on a <cadence> cadence. Last reviewed: <date>.
+  known_cost: >
+    This standard ships no checker. Adopters wanting the Checked state must write it
+    themselves and most will not — the same failure mode the section describes. The minimum
+    bar is therefore not "build a checker" but "never let a registry be silently unattended":
+    declaring Unattended is fully compliant; being unattended without declaring it is not.
+
 budget:
   new_project: "10% of dev time"
   mature_project: "15% of dev time"
@@ -54,3 +98,43 @@ metrics:
   - "High priority ratio (P0+P1 < 20%)"
 
 commit_marking: "Tech-Debt: TD-NNN (introduced|resolved: description)"
+
+rules:
+  - id: td-exp-001
+    trigger: an item is past its Target Resolution Date
+    instruction: >
+      Treat it as a finding, not a neutral background condition. Apply exactly one of the
+      three dispositions (resolve / withdraw / extend-with-reason). A registry holding an
+      overdue item with no disposition applied is non-compliant.
+    priority: required
+  - id: td-exp-002
+    trigger: designing or reviewing any automation over the tech debt registry
+    instruction: >
+      Never implement expiry as automatic extension or automatic closure. Automatic extension
+      makes the date unfalsifiable so it measures nothing; automatic closure deletes the record
+      without anyone deciding to. Both stop the clock and neither leaves a trace.
+    priority: required
+  - id: td-exp-003
+    trigger: extending a Target Resolution Date
+    instruction: >
+      Record the reason next to the new date. Changing only the date is not an extension, it is
+      an erasure with a timestamp on it.
+    priority: required
+  - id: td-exp-004
+    trigger: an item has been extended three or more times with no progress between extensions
+    instruction: Re-triage it as a Withdraw candidate rather than issuing a fourth date.
+    priority: recommended
+  - id: td-exp-005
+    trigger: creating or auditing a tech debt registry, in any storage format
+    instruction: >
+      Put the registry in exactly one of two states and make which one visible inside the
+      registry: Checked (a named runnable check fails on overdue items) or Unattended (an
+      Unattended Declaration naming owner, review cadence, and last-review date). Neither state
+      declared means non-compliant.
+    priority: required
+  - id: td-exp-006
+    trigger: reading an Unattended Declaration
+    instruction: >
+      Apply the declaration's own cadence to its "Last reviewed" date. A declaration whose last
+      review is older than its stated cadence is itself an overdue item.
+    priority: required

package/bundled/ai/standards/turn-completion-integrity.ai.yaml

@@ -0,0 +1,131 @@
+# Turn Completion Integrity - AI Optimized
+# Source: core/turn-completion-integrity.md
+
+id: turn-completion-integrity
+meta:
+  version: "1.3.0"
+  updated: "2026-09-08"
+  source: core/turn-completion-integrity.md
+  description: An agent must not end a turn having stated a next action it did not take; enforced at turn end, not by instruction
+  related:
+    - agent-behavior-discipline
+    - ai-response-navigation
+    - anti-hallucination
+
+rules:
+  r1_commitment_must_be_kept:
+    summary: A stated next action is not optional
+    rule: If the final message states a first-person commitment to a next action, the turn must not end until that action is taken or the blocker is named
+    do:
+      - Take the action before ending the turn
+      - Or name, per item, the person or input it waits on
+    do_not:
+      - End on "the main parts are done"
+      - End on a sentence describing work that was not performed
+
+  r2_blocker_statement_is_itemized:
+    summary: A blocker statement names who or what each item waits on
+    rule: Ending without acting is permitted only when every remaining item is listed with its blocker
+    do:
+      - "State: item X waits on the user's key; item Y waits on nothing and is being done now"
+    do_not:
+      - Summarize progress in place of listing blockers
+
+  r3_enforce_at_the_decision_point:
+    summary: An instruction is not enforcement
+    rule: R1 must be evaluated at turn end by something that is not the agent
+    rationale: The instruction form of this rule was violated inside the same session that read it, and again seven days after being published as an optional standard with no checker
+
+  r4_never_block_on_standing_work:
+    summary: Fire only on a commitment made in this message
+    rule: The check must not consult repository state such as open TODOs or backlog items
+    rationale: A gate that is true on every turn gets disabled, and then protects nothing
+
+  r5_fail_open:
+    summary: Any error lets the turn end
+    rule: Unreadable transcript, missing field, malformed JSON, or unexpected schema must all exit without blocking
+    rationale: A hook that can trap a session is worse than none — the human's only recovery is to disable it permanently
+
+  r6_bound_the_blocking:
+    summary: Cooldown plus a rolling-window cap
+    rule: Enforce a cooldown between blocks and a maximum number of blocks per rolling time window
+    do_not:
+      - Count the cap per session with no reset — that is an off switch on a delay, and it disarms silently in long sessions
+
+  r7_each_language_ships_its_own_corpus:
+    summary: Language coverage is a correctness property
+    rule: Each supported language ships as a locale pack with its own must-block and must-pass corpus, and that corpus runs in CI
+    rationale: A detector carrying one language's patterns, shipped to an adopter working in another, is installed, runs, exits 0, and can never fire — indistinguishable from good behavior
+
+  r9_exempt_a_human_directed_stop:
+    summary: The human can end the turn, and the agent's words cannot say so
+    rule: The check must exempt a turn the human asked to end, decided from the human's own most recent message
+    rationale: Measured on the first real firing after shipping — a turn ending by instruction and a turn ending on an abandoned commitment produce the same words from the agent, because in both the agent names work it is not doing now
+
+  r10_keep_the_stop_pattern_narrow:
+    summary: A false exemption silences the check for the rest of the session
+    rule: The stop-request pattern must be narrow, and its corpus must include work instructions that merely contain a stopping word
+    rationale: During implementation the pattern matched "stop using the hardcoded list and walk the registry instead" — an instruction to do work, read as an instruction to stop
+
+  r11_recognise_your_own_block_message:
+    summary: The check's own output re-enters the input it reads
+    rule: The block message must carry a fixed marker, and any human turn containing it must be skipped when looking for the human's last message
+    rationale: Without this R9 works exactly once — the block message becomes "the human's most recent message" on the next run, hiding the real instruction to stop, silently and only in the situation the exemption was built for
+
+  r12_recognise_the_ending_you_demand:
+    summary: A check that cannot see the ending R2 defines punishes correct behaviour
+    rule: The check must recognise an itemized blocker list and must not block it
+    requires:
+      - two or more list items — one item with an attribution is a sentence, not an itemization
+      - a blocker attribution anywhere in the message, not per item (real writing puts it in the preamble)
+      - no vague-completion phrase ("the rest are done" is the summary R2 rejects)
+      - the attribution search must exclude the check's own headings, or a heading containing "you" satisfies it
+    rationale: Blocking the one ending the standard asks for gets the check uninstalled faster than missing a violation
+
+  r8_declare_unsupported_languages:
+    summary: Say when the check is inactive
+    rule: An adopter whose language has no locale pack must be told the check cannot fire for them
+
+detector_shape:
+  match: first-person future marker plus an action verb within one sentence
+  exclusions:
+    - reporting verbs (say, note, mention) — describing is not doing
+    - negation — a refusal is a decision, not a commitment
+    - quoted or tabular text — examples of the pattern are not instances of it
+  scope: sentence, not paragraph
+  rationale: Each exclusion was added because its absence produced a false block in testing
+
+prohibited_behaviors:
+  - id: stated-but-undone
+    description: Do NOT end a turn on a sentence describing a next action that was not performed
+    correct_action: Perform it, or list every remaining item with its blocker
+
+  - id: instruction-as-enforcement
+    description: Do NOT treat a rule in the agent's instructions as enforcement of that rule
+    correct_action: Evaluate it at the decision point with an external check
+
+  - id: monolingual-detector
+    description: Do NOT ship a prose detector without a corpus for every language it claims to cover
+    correct_action: Ship a locale pack per language, each with a corpus that runs in CI
+
+enforcement:
+  hook_type: Stop
+  trigger: Stop
+  matcher: ""
+  script_ref: "scripts/hooks/check-turn-completion.mjs"
+  hook_script: "scripts/hooks/check-turn-completion.mjs"
+  severity: warning
+  timeout_ms: 2000
+
+checklist:
+  - The check runs at turn end, not as an instruction to the agent
+  - Every failure path exits without blocking
+  - A cooldown and a rolling-window cap are both present
+  - Each shipped language has a corpus, and the corpus runs in CI
+  - Adopters in unsupported languages are told the check is inactive
+  - The check does not consult repository state
+  - A turn the human asked to end is exempt, decided from the human's message
+  - The stop-request corpus includes work instructions containing a stopping word
+  - The check recognises its own block message and does not read it as the human's
+  - The check recognises the itemized blocker ending R2 defines, and does not block it
+  - The attribution search excludes the check's own headings and scaffolding

package/bundled/core/acceptance-criteria-traceability.md

@@ -2,8 +2,8 @@
 
 > **Language**: English | [繁體中文](../locales/zh-TW/core/acceptance-criteria-traceability.md)
 
-**Version**: 1.1.0
-**Last Updated**: 2026-06-19
+**Version**: 1.2.0
+**Last Updated**: 2026-08-24
 **Applicability**: All software projects using specification-driven or test-driven workflows
 **Scope**: universal
 
@@ -231,6 +231,8 @@ Exceptions to coverage requirements MUST be documented:
 | Infrastructure limitation | Test environment constraint | Workaround plan |
 | Deferred to next iteration | Agreed with stakeholders | Ticket reference |
 
+> **Deferred items** this standard produces — `Gaps` (Uncovered AC / Partial AC), the exceptions above, and the report's `Action Items` — are governed by [deferred-item-exit](deferred-item-exit.md). | 本標準產出的**延後項目**（`Gaps` 的 Uncovered AC／Partial AC、上表的例外、報告的 `Action Items`）適用 [deferred-item-exit](deferred-item-exit.md)。
+
 ---
 
 ## AC Coverage Report Format
@@ -382,3 +384,4 @@ Generated spec MUST include:
 |---------|------|---------|
 | 1.0.0 | 2026-03-18 | Initial version — traceability matrix, coverage calculation, spec generation rules |
 | 1.1.0 | 2026-05-12 | Add `not_implemented` 4th status; update CI gate formula; add decision tree (XSPEC-199) |
+| 1.2.0 | 2026-08-24 | Pointer to `deferred-item-exit` for coverage gaps, threshold exceptions and report action items (XSPEC-391 R5) |

package/bundled/core/adr-standards.md

@@ -2,8 +2,8 @@
 
 > **Language**: English | [繁體中文](../locales/zh-TW/core/adr-standards.md)
 
-**Version**: 1.0.0
-**Last Updated**: 2026-03-26
+**Version**: 1.1.0
+**Last Updated**: 2026-08-24
 **Applicability**: All software projects making architectural decisions
 **Scope**: universal
 **Industry Standards**: ISO/IEC/IEEE 42010 (Architecture Description), TOGAF ADR
@@ -81,6 +81,26 @@ Chosen option: **[Option N]**, because [justification].
 - [Related ADRs, SPECs, PRs, or external references]
 ```
 
+> **Deferred items** recorded in this template — accepted risks and negative consequences under `Consequences`, and options considered but not taken up — are governed by [deferred-item-exit](deferred-item-exit.md). | 此範本中記下的**延後項目**（`Consequences` 下的既受風險與負面後果、被考慮而未採用的選項）適用 [deferred-item-exit](deferred-item-exit.md)。
+
+---
+
+## Acceptance Criteria Live in the SPEC
+
+An ADR records a **decision**. It does not carry acceptance criteria of its own. Where a decision has acceptance criteria, they are maintained in **exactly one place — the SPEC** — and the ADR reaches them by link, from `Technical Story` or `Links`.
+
+A copied AC list has two owners and only one of them is ever updated. The measured consequence: implementation completed, the SPEC's checkboxes ticked, and the ADR's copy left entirely unticked — so a reader of the ADR concludes nothing was built, while AC coverage tooling, which reads the SPEC, cannot see the ADR's copy at all.
+
+一份 ADR 記錄的是**決策**，它不帶自己的一份驗收標準。當一個決策有驗收標準時，
+它們**只在一處維護——SPEC**，而 ADR 以連結（`Technical Story` 或 `Links`）指向它。
+
+一份被複製的 AC 清單有兩個擁有者，而只有其中一個會被更新。實測後果是：
+實作完成、SPEC 的勾選框打勾，ADR 那一份**整份留在未勾選狀態**——
+讀 ADR 的人於是認為什麼都沒做，而讀 SPEC 的 AC 覆蓋率工具**根本看不到 ADR 裡那一份**。
+
+> This is a boundary rule about **where an AC is maintained**, not about how ACs are written or labelled. For the AC format itself see [spec-driven-development](spec-driven-development.md); for AC-to-verification traceability see [acceptance-criteria-traceability](acceptance-criteria-traceability.md).
+> 本條是關於 **AC 在哪裡維護**的邊界規則，不涉及 AC 的寫法或標註慣例。
+
 ---
 
 ## ADR Numbering
@@ -179,6 +199,8 @@ Before accepting an ADR, verify:
 - [ ] **Status** is set correctly
 - [ ] **Links** to related artifacts are included
 - [ ] File is stored in `docs/adr/` with correct naming
+- [ ] **No acceptance criteria are copied into the ADR** — they are linked to the SPEC that maintains them
+- [ ] Every **deferred item** carries the identifier of its exit ([deferred-item-exit](deferred-item-exit.md))
 
 ---
 
@@ -192,6 +214,8 @@ Before accepting an ADR, verify:
 | No consequences | Incomplete analysis | Always list good and bad outcomes |
 | Vague context | Useless for future readers | Include specific constraints and drivers |
 | Editing accepted ADRs | Lost history | Supersede instead of editing |
+| Copying the SPEC's acceptance criteria into the ADR | Two owners, one updated; the ADR's copy stays unticked and coverage tooling never sees it | Keep one copy in the SPEC; link to it |
+| An accepted risk with no exit | The ADR is the only carrier, and it is now approved | Give it an exit and name the exit here |
 
 ---
 

package/bundled/core/code-review-checklist.md

@@ -2,8 +2,8 @@
 
 > **Language**: English | [繁體中文](../locales/zh-TW/core/code-review-checklist.md)
 
-**Version**: 1.4.0
-**Last Updated**: 2026-06-18
+**Version**: 1.5.0
+**Last Updated**: 2026-08-24
 **Applicability**: All software projects with code review processes
 **Scope**: universal
 **Industry Standards**: SWEBOK v4.0 Chapter 10
@@ -281,6 +281,8 @@ Is there a specific reason for this approach?
 
 > This comment prefix approach aligns with the [Conventional Comments](https://conventionalcomments.org/) specification, which standardizes review feedback across teams and tools.
 
+> **Deferred items** a review produces — non-blocking comments (`⚠️ IMPORTANT`, `💡 SUGGESTION`, `[SUGGESTION]`, `[NIT]`) accepted without a change in this change — are governed by [deferred-item-exit](deferred-item-exit.md). | Review 產出的**延後項目**（本次未改而被接受的非阻斷留言：`⚠️ IMPORTANT`、`💡 SUGGESTION`、`[SUGGESTION]`、`[NIT]`）適用 [deferred-item-exit](deferred-item-exit.md)。
+
 ### Alternative: Text Labels
 
 For teams preferring plain text labels without emojis:
@@ -664,6 +666,7 @@ Comment Prefixes:
 
 | Version | Date | Changes |
 |---------|------|---------|
+| 1.5.0 | 2026-08-24 | Added: pointer to `deferred-item-exit` for non-blocking comments accepted without a change (XSPEC-391 R5) |
 | 1.4.0 | 2026-06-18 | Added: source + configurability note for PR-size/response-time thresholds (SmartBear/Cisco study, Google practices) + bulk-change exception (XSPEC-292 T8) |
 | 1.3.0 | 2026-01-12 | Added: Comprehensive Refactoring PRs section with pre-review checklist, review focus areas, large refactoring guidelines, red flags, and best practices |
 | 1.2.0 | 2026-01-05 | Added: SWEBOK v4.0 Chapter 10 (Software Quality) to References |

package/bundled/core/context-aware-loading.md

@@ -131,7 +131,7 @@ For Claude Code users, a `UserPromptSubmit` hook can automatically inject releva
 **Requirements:**
 - Hook execution must complete in < 500ms
 - Hook failures must not block the user's prompt
-- See `scripts/hooks/inject-standards.js` for reference implementation
+- See `scripts/hooks/inject-standards.mjs` for reference implementation
 
 ---