SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsreview-methodology-foundations

review-methodology-foundations

Writing & research

View repository

Updated 10-16 00:02

About this skill

review-methodology-foundations

This Skill is a foundation, not a user-facing Skill. Its chapter order differs from the 6 user-facing Skills—it does not include user-facing Skill chapters such as trigger scenarios, main Agent role, 12-stage workflow, word count control, or mode selection. This Skill only carries 20 SSOTs and provides the specification for other Skills to reference SSOTs.

§1 What you produce and what you do not produce

§1.1 What this Skill is

review-methodology-foundations is the sharedmethodological foundationof the six academic review Skills. It does not directly respond to user requests; instead, it carries 20 Single Source of Truth (SSOT) documents for the following six user-facing review Skills to reference via references/ :

  • field-overview-review(Domain overview review: integration + scope)
  • field-evolution-review(Domain development review: narrative + meta-narrative)
  • literature-set-review(Review based on an existing literature set: integration + narrative)
  • systematic-review(Systematic review: strict PRISMA 2020)
  • meta-analysis(Meta-analysis: systematic-review + quantitative synthesis)
  • theoretical-review(Theoretical review: construct analysis and theory building)

§1.2 What this Skill is not

  • Not a user entry point—users should not directly trigger this Skill; "NOT directly triggered" in the description is a hard constraint.
  • Not a workflow executor—this Skill does not dispatch SubAgents, call MCPs, or produce review chapters; execution is undertaken by each of the six user-facing review Skills.
  • Not a restatement of textbook source text—SSOTs are methodological decisions that have been structured, operationalized, and arbitrated by the main Agent; they are not a repetition of Cooper/Cochrane/Borenstein/JBI.

§1.3 Why it is a foundation (rationale for centralized SSOT maintenance)

If the foundation Skill did not exist, the 20 methodological documents would need to be copied separately into the 6 review Skills, causing:

  • Redundant storage: the same PRISMA-S template copied 6 times, with any update requiring manual synchronization in 6 places
  • Version drift: after the 6 Skills each maintain their own versions, definitions will gradually diverge (e.g., Cohen's d interpretation thresholds, I² thresholds, AMSTAR 2 critical domain counts)
  • Inconsistent judgments: when users switch between different Skills, they encounter different answers to the same issue (e.g., selection of publication bias test tools)

The foundation architecture eliminates these risks through "define once, reference many times." Any SSOT update automatically benefits all Skills that reference it.

§2 SSOT index: when to reference which

The following decision table is organized by the 12-stage review methodology lifecycle (L1–L12) + cross-cutting categories.Each trigger situation uses the granularity of a "specific decision action" (applying the C3.1 entry granularity rule), to prevent SubAgents from misreading SSOTs at the coarse granularity of stage names.

§2.1 Index by 12 stages + cross-cutting categories

Trigger situation (specific decision action) Referenced SSOT Layer
The main Agent at L1 needs to make the 5-question review type decision / inter-Skill switching decision SSOT-REVIEW-TYPES.md 1.A
The main Agent at L2 needs to select one from the 14 PICO frameworks / dispatch the PICO Constructor SubAgent SSOT-PICO-FRAMEWORKS.md 1.B
The main Agent at L4 needs to determine the database list / convert PICO keywords into Boolean strings SSOT-SEARCH-STRATEGY.md 1.B
The main Agent at L5 needs the PRISMA-S 16-item exhaustive requirement to document the search SSOT-PRISMA-S-TEMPLATE.md 1.B
The main Agent at L6/L11 needs to draw the PRISMA 2020 flow diagram (4-stage counts) SSOT-PRISMA-FLOW-DIAGRAM.md 1.B
The main Agent at L7 needs to determine the extraction fields (six major categories R/P/I/C/O/M) / SR L7 → MA L9.1 field mapping SSOT-EXTRACTION-FIELDS.md 1.B
The main Agent at L8 needs to select the RoB tool according to the study design (10 categories of designs) SSOT-ROB-TOOL-MAPPING.md 1.B
The main Agent at L9 needs to select the synthesis method (meta-analysis / SWiM / Whittemore / Wong / Popay) SSOT-SYNTHESIS-METHODS.md 1.B
The main Agent at L9.1 needs to make the single-study effect size type decision + SE back-calculation SSOT-EFFECT-SIZE-DECISION.md 1.B
The main Agent at L10 needs GRADE 5 downgrades + 3 upgrade assessments SSOT-GRADE-FRAMEWORK.md 1.B
The main Agent at L11 needs to select the reporting standard according to the review type (PRISMA / PRISMA-ScR / RAMESES / APA JARS) SSOT-REPORTING-STANDARDS.md 1.B
The main Agent at L12 needs to perform the AMSTAR 2 16-item self-assessment (SR/MA strict mode only) SSOT-AMSTAR-2-CHECKLIST.md 1.B
The main Agent must read immediately after startup (shared by all 6 Skills) SSOT-LIFECYCLE-12-PHASES.md 1.A
SubAgent gets stuck on terminology meaning / cross-Skill terminology consistency check SSOT-METHODOLOGY-TERMINOLOGY.md 1.A
The main Agent must read immediately after startup (shared by all 6 Skills, mandatory activation, unconditional loading) SSOT-AI-FAILURE-DEFENSES.md 1.A
Textbook methods conflict on a specific decision / LLM framework boundary issues / Open Science era scrutiny SSOT-EXPERT-METHODOLOGY-INSIGHTS.md 1.A
User scenario contains psychology trigger words (APA, JARS, scales, constructs, replication crisis, etc.) / submission to APA 89 journals SSOT-PSYCHOLOGY-SUPPLEMENTS.md 1.C
User input contains Chinese / submission to Chinese journals / GB/T 7714 citation format / MCP Chinese retrieval fallback SSOT-CHINESE-CONTEXT.md 1.C
The main Agent at L3 needs to decide strict vs learning mode / learning mode simplification disclosure SSOT-MODE-PROFILES.md 1.C
The main Agent at L1/L11 needs to determine the word count formula / severe word count mismatch threshold determination SSOT-WORDCOUNT-RULES.md 1.C

§2.2 SubAgent loading rules (apply C7 frontmatter type layering)

The frontmatter of each SSOT is layered into three fields: "required dependencies / boundary partners / weak references":

Field Semantics SubAgent behavior
hard_dependencies Required dependencies (without reading them, the SSOT cannot be used independently) When loading this SSOTmustload these first
boundary_partners Boundary partners (complementary at the same layer, clarifying scope boundaries) As neededjump, not mandatory
weak_references Weak references (terminology alignment / concept supplementation only) Optionalloading (only when terminology meaning is unclear)

The beginning of each SSOT file contains > 使用判定 block (explicitly stating trigger conditions), and the end of the file contains an "Application and Regression Navigation" section (explicitly stating which Skills reference it at which stage, and where to return after reading). When loading an SSOT, the SubAgent only needs to read the decision block at the beginning to decide whether to load it, without reading the SKILL.md decision table (progressive disclosure level 3).

§2.3 Small decision table for downstream Skills

Downstream Skills provide, in their own SKILL.md "§8 When to load the foundation SSOT" section, thesubsetsmall decision table relevant to this Skill (without copying all 20 rows of this section). For example:

  • meta-analysis §8 focuses on listing EFFECT-SIZE / SYNTHESIS / PSYCHOLOGY / GRADE / AMSTAR-2
  • theoretical-review §8 does not list PRISMA-FLOW-DIAGRAM / ROB-TOOL-MAPPING (not applicable)

§3 Reference relationship diagram among SSOTs

The 20 SSOTs are distributed by layer. The reference direction is strictly bottom-up (high Layer → low Layer), with no reverse or circular references.

Layer 1.A（基础层，5 SSOT，无对外引用或仅引用 1.A 内部）
├── SSOT-REVIEW-TYPES
├── SSOT-LIFECYCLE-12-PHASES         (→ REVIEW-TYPES)
├── SSOT-METHODOLOGY-TERMINOLOGY
├── SSOT-AI-FAILURE-DEFENSES
└── SSOT-EXPERT-METHODOLOGY-INSIGHTS  (→ AI-FAILURE-DEFENSES)
                │
                ▼
Layer 1.B（功能层，11 SSOT，向 1.A 引用；同层仅"边界伙伴"互引）
├── SSOT-PICO-FRAMEWORKS              (→ REVIEW-TYPES, LIFECYCLE)
├── SSOT-SEARCH-STRATEGY              (→ REVIEW-TYPES, LIFECYCLE, PICO)
├── SSOT-PRISMA-S-TEMPLATE            (→ SEARCH-STRATEGY)
├── SSOT-PRISMA-FLOW-DIAGRAM          (→ LIFECYCLE)
├── SSOT-EXTRACTION-FIELDS            (→ LIFECYCLE)
├── SSOT-EFFECT-SIZE-DECISION         (→ TERMINOLOGY; ⇄ EXTRACTION, SYNTHESIS [边界伙伴])
├── SSOT-ROB-TOOL-MAPPING             (→ LIFECYCLE, TERMINOLOGY, REVIEW-TYPES)
├── SSOT-SYNTHESIS-METHODS            (→ TERMINOLOGY, REVIEW-TYPES, LIFECYCLE; ⇄ EFFECT-SIZE [边界伙伴])
├── SSOT-GRADE-FRAMEWORK              (→ REVIEW-TYPES, LIFECYCLE, TERMINOLOGY, ROB, SYNTHESIS)
├── SSOT-REPORTING-STANDARDS          (→ REVIEW-TYPES, LIFECYCLE)
└── SSOT-AMSTAR-2-CHECKLIST           (→ LIFECYCLE)
                │
                ▼
Layer 1.C（语境层，4 SSOT，向 1.A + 1.B 引用）
├── SSOT-PSYCHOLOGY-SUPPLEMENTS       (→ REPORTING, SYNTHESIS, EFFECT-SIZE, ROB, EXPERT, AI-FAILURE, LIFECYCLE)
├── SSOT-CHINESE-CONTEXT              (→ SEARCH, REPORTING, REVIEW-TYPES, LIFECYCLE, PSYCHOLOGY)
├── SSOT-MODE-PROFILES                (→ LIFECYCLE, REVIEW-TYPES)
└── SSOT-WORDCOUNT-RULES              (→ REVIEW-TYPES)

§3.1 Boundary partner description (apply C7 §A.2 + W5_FIX_PLAN A.P1.1 + W5.4 R4 R11)

In the current SSOT, there are 2 pairs of boundary partners with bidirectional mutual references:

Boundary partner pair 1: EFFECT-SIZE-DECISION ⇄ SYNTHESIS-METHODS (same layer, Layer 1.B)

EFFECT-SIZE-DECISION and SYNTHESIS-METHODS each declare the other as boundary_partners. Their scope division of labor:

  • EFFECT-SIZE-DECISION Coverssingle-studyeffect size calculation and interpretation
  • SYNTHESIS-METHODS Coverscross-studypooling, heterogeneity, publication bias

3 specific scenarios for SubAgent proactive jumps：

  1. After loading EFFECT-SIZE, the SubAgent finds that I² / τ² / publication bias need to be done → proactively jump SYNTHESIS-METHODS
  2. After loading SYNTHESIS-METHODS, the SubAgent finds that single-study SE / handling unit-of-analysis need to be back-calculated first → proactively jump EFFECT-SIZE
  3. The SubAgent, at L9, decides "switch to meta-analysis vs stay in SR L9 SWiM" → load simultaneouslyboth SSOTs to compare decision thresholds (k threshold + heterogeneity)

When not in the above 3 scenarios, reading either SSOT will not force a jump to the other (jump on demand, not mandatory).

Boundary partner pair 2: AI-FAILURE-DEFENSES ⇄ EXPERT-METHODOLOGY-INSIGHTS (same layer, Layer 1.A, newly added in W5.4 R4)

The two each declare the other as boundary_partners (bidirectional mutual references). Scope division of labor:

  • AI-FAILURE-DEFENSES Coversidentification and defense mechanisms of LLM failure modes(4 types of failures: citation hallucination / effect exaggeration / confidence hallucination / sampling bias amplification)
  • EXPERT-METHODOLOGY-INSIGHTS Coversmethodological feasibility boundaries + textbook conflict arbitration stance + era-reality constraints(5-dimensional judgment)

3 specific scenarios for SubAgent proactive jumps：

  1. After loading EXPERT, the SubAgent encounters LLM implementation differences that need to be disclosed → proactively jump to AI-FAILURE to see §1-§4 defense mechanisms
  2. After loading AI-FAILURE, the SubAgent encounters feasibility boundary conflicts in textbook methods → proactively jump to EXPERT to see dimension 1 (LLM framework feasibility boundaries)
  3. When writing the C4 three-part disclosure, the SubAgent needs to cite both AI-FAILURE's failure defenses + EXPERT's methodological stance → load simultaneouslyboth SSOTs

When not in the above 3 scenarios, jump on demand, not mandatory.

§3.2 Citation cycle self-check

Before writing to disk, check:

  • The hard_dependencies of a Layer 1.A SSOT should contain only 1.A internal SSOTs (or be empty)
  • The hard_dependencies of a Layer 1.B SSOT must not contain 1.B / 1.C SSOTs;boundary_partners only same-layer (only the EFFECT-SIZE ⇄ SYNTHESIS pair)
  • A Layer 1.C SSOT may reference 1.A or 1.B,must neverreference other 1.C SSOTs to form a horizontal closed loop

Actual status (as of v1.1.0): the three frontmatter fields of the above 20 SSOTs all satisfy the above constraints.

§3.3 Long citation chain SubAgent loading efficiency verification (W5.4 R6 footnote)

Background: W5.3-B Sub-dimension 3 "Long Citation Chain SubAgent Loading Efficiency" scores 9.0/10, with the remaining 1.0-point gap being "not yet validated by end-to-end fresh SubAgent simulation testing". Currently, each 4-level citation chain (e.g., MA L9.6 → references/L9.6-subgroup-metaregression.md → SSOT-SYNTHESIS-METHODS §3.6 → SSOT-METHODOLOGY-TERMINOLOGY) has its "clear jump conditions" fully in place through structural remediation (§A usage determination + §B dual blocks of when to read / where to return after reading), but actual behavior validation with a fresh SubAgent awaits end-to-end simulation supplementation after Wave 5.4.

Validation method (to be executed): Dispatch a fresh SubAgent (with no context whatsoever), run the following 3 chains end-to-end, and quantify the actual effective rate of each level's jump:

  1. Chain 1：meta-analysis L9.6 → references/L9.6-subgroup-metaregression.md → SSOT-SYNTHESIS-METHODS §3.6 → SSOT-METHODOLOGY-TERMINOLOGY
    • Expectation: Level 4 (METHODOLOGY-TERMINOLOGY) jump rate ≤ 30% (design expectation: terminology SSOT is a fallback, not a routine jump)
  2. Chain 2：systematic-review L11 → references/L11-prisma-reporting.md → SSOT-REPORTING-STANDARDS → SSOT-PSYCHOLOGY-SUPPLEMENTS
    • Expectation: In psychology scenarios, the whole-chain effective rate ≥ 50%
  3. Chain 3：literature-set-review L4 D11 → SSOT-AI-FAILURE-DEFENSES
    • Expectation: Maintain ≥ 86% effective rate (already reached 86% before remediation)

Identified constraint: MCP MCP-METHODOLOGY-TERMINOLOGY is a terminology fallback, not a routine jump target—the SubAgent should not jump when the terminology meaning is clear. This footnote explicitly declares this design expectation to prevent subsequent audits from misjudging the 30% jump rate as a defect.

Tracking: This validation is to be completed by an end-to-end simulated SubAgent after Wave 5.4, with results written back to the "Actual Status" paragraph of this section.

§4 Relationship with the Six Review Skills

§4.1 Strong-Reference SSOTs for Each of the Six Review Skills

Skill Strong-reference SSOT (required reading) Weak-reference SSOT (as needed)
field-overview-review(integration + scope) REVIEW-TYPES, LIFECYCLE, AI-FAILURE, PICO, SEARCH, EXTRACTION, SYNTHESIS, REPORTING, WORDCOUNT, MODE TERMINOLOGY, EXPERT, GRADE, PSYCHOLOGY, CHINESE
field-evolution-review(narrative + metanarrative) REVIEW-TYPES, LIFECYCLE, AI-FAILURE, SEARCH, EXTRACTION, SYNTHESIS, REPORTING, WORDCOUNT, MODE TERMINOLOGY, EXPERT, PICO, PSYCHOLOGY, CHINESE
literature-set-review(existing literature set) REVIEW-TYPES, LIFECYCLE, AI-FAILURE, EXTRACTION, SYNTHESIS, REPORTING, WORDCOUNT, MODE TERMINOLOGY, EXPERT, ROB, GRADE, PSYCHOLOGY, CHINESE
systematic-review(strict PRISMA) All 16 1.A+1.B SSOTs + WORDCOUNT, MODE PSYCHOLOGY, CHINESE
meta-analysis(systematic + quantitative synthesis) All 16 1.A+1.B + EFFECT-SIZE emphasis + PSYCHOLOGY emphasis CHINESE, MODE, WORDCOUNT
theoretical-review(construct analysis) REVIEW-TYPES, LIFECYCLE, TERMINOLOGY, AI-FAILURE, EXPERT, SEARCH, REPORTING, WORDCOUNT, MODE EXTRACTION, SYNTHESIS, PSYCHOLOGY, CHINESE

Downstream Skills should explicitly state their strong/weak reference sets in the reference decision table at the top level of their own SKILL.md, to avoid reading all 20 SSOTs in a single task.

SSOT-AI-FAILURE-DEFENSES mandatory activation baseline(applying W5_FIX_PLAN A.P2.2): All 6 user-facing Skills must explicitly list SSOT-AI-FAILURE-DEFENSES in the "§3.1 Main Agent Required Reading (load immediately after startup)" subsection. For specific integration strength (strong / medium / basic), see the 6-Skill integration matrix in SSOT-AI-FAILURE-DEFENSES §6.2.

§4.2 Review Type Decision Tree (High-Level)

Users entering a review task usually do not know which Skill to use. The decision tree skeleton is as follows (for detailed decisions see references/SSOT-REVIEW-TYPES.md §2 Judgment 5 Questions):

用户场景
  │
  ├─ 是否有现成文献集？
  │  ├─ 是 → literature-set-review
  │  └─ 否 → 继续
  │
  ├─ 综述目的是什么？
  │  ├─ 系统梳理领域全貌（无严格 PRISMA 要求）→ field-overview-review
  │  ├─ 追溯领域发展史/演化路径 → field-evolution-review
  │  ├─ 严格 PRISMA 2020 系统综述（医学/公共政策）→ systematic-review
  │  ├─ 量化合并效应量（含 systematic 全套）→ meta-analysis
  │  └─ 构念分析/理论建构/理论比较（无穷尽检索）→ theoretical-review
  │
  └─ 用户场景模糊 → 加载 SSOT-REVIEW-TYPES 全文 + 5 问决策

Downstream Skills need not copy this decision tree; they should directly reference SSOT-REVIEW-TYPES.md。

§4.3 Capability Delegation Between Skills

When a Skill discovers during execution that the user's actual needs exceed the scope of this Skill, it shouldproactively suggest switchingrather than reluctantly taking it on:

  • field-overview-review discovers the user actually wants to do strict PRISMA → suggest switching to systematic-review
  • literature-set-review discovers the literature set is sufficiently homogeneous and requires quantitative pooling → suggest switching to meta-analysis
  • systematic-review discovers the user only wants a scope overview → suggest switching to field-overview-review
  • theoretical-review Found that the user actually wants an empirical review → suggest switching to the corresponding empirical review Skill

The downstream Skill should explicitly list its delegation rules in the "§13 Relationship with Other Skills" section of SKILL.md.

§5 Citation Conventions (How Other Skills Cite the SSOT)

§5.1 Citation Format

参见 review-methodology-foundations/references/SSOT-PICO-FRAMEWORKS.md §2.3（v1.0.0）

Or use footnotes/end-of-paragraph markers in citation-dense places:

14 个 PICO 框架 [SSOT-PICO §1] 中，PEO 适用于质性综述 [SSOT-PICO §2.7]。

Cross-Skill Citation Syntaxshould include the version number(apply W5_FIX_PLAN A.P2.1) — this makes it easy to identify Skills that need review when the SSOT major version is upgraded.

§5.2 Copying and Pasting SSOT Content Is Not Allowed

The SKILL.md / references/ files of downstream Skillsmust notrepeatedly define standards, formulas, thresholds, or field tables already covered by the SSOT. For example:

  • Do not rewrite the 27 PRISMA 2020 items separately in each of the 6 Skills
  • Do not define Cohen's d interpretation thresholds separately in each of the 6 Skills
  • Do not rewrite the 16 AMSTAR 2 items separately in each of the 6 Skills

Downstream Skills should only state "which section of the SSOT is cited at which stage," rather than moving the content over.

§5.3 SSOT Version Number Alignment

Each SSOT declares in its frontmatter version: X.Y.Z. When downstream Skills cite it, theymay optionallyrecord the version they are aligned to at the time of citation, making it easy to identify Skills that need review when the SSOT major version is upgraded:

本节遵循 SSOT-EFFECT-SIZE-DECISION.md v1.0.0 §3 决策树。

§6 SSOT Maintenance Rules

§6.1 When to Update the SSOT

Legitimate triggers for updating the SSOT:

  • New edition of a textbook— review the relevant SSOT after a new edition of Cooper / Cochrane / Borenstein / JBI / Booth / APA Manual is released
  • New failure mode— a new LLM review failure mode is discovered during use (update SSOT-AI-FAILURE-DEFENSES)
  • User feedback— multiple instances of user feedback point to unclear or missing SSOT content, and it falls within the SSOT scope
  • Reporting standard updates— extensions or new editions of PRISMA, APA JARS, AMSTAR, etc. are released
  • Statistical method updates— important new statistical methods become widely accepted in the field (e.g., RoBMA, advances in selection models)

Not triggers for SSOT updates:

  • One-off problems in a single task (should be handled in that Skill's own references/)
  • Skill workflow adjustments unrelated to the SSOT
  • Personal preference

§6.2 Version Number Conventions

  • MAJOR(X+1.0.0) — major structural changes to the SSOT (deleting an entire chapter, merging multiple SSOTs, changing Layer affiliation)
  • MINOR(X.Y+1.0) — adding chapters, adding decision items, supplementing new tools or frameworks
  • PATCH(X.Y.Z+1) — errata, clarification of wording, citation completion, revisions that do not affect judgment results

§6.3 Notifying Downstream After Updates

After the SSOT is updated, notify downstream Skills according to the "update dimension":

  • MAJOR/MINOR update: maintainers should review all Skills that cite this SSOT to confirm the citation locations are still accurate (section numbers have not drifted); the version numbers recorded by downstream Skills at citation time should be updated accordingly
  • PATCH update: downstream review is usually unnecessary, but the change description should be recorded in the "Version and Maintenance" section at the end of the SSOT

Every SSOT must have a "§N Version and Maintenance" subsection at the end, recording the change log in reverse chronological order.

§7 Self-check list (pre-write check for this SKILL.md itself)

  • frontmatter name Consistent with folder name（review-methodology-foundations，kebab-case）
  • description < 1024 characters(actually < 1024)
  • description contains explicit "NOT directly triggered"(second sentence)
  • frontmatter contains is_foundation: true and consumed_by list(6 overview Skills)
  • All 20 SSOTs are under references/ under(verified before creation)
  • SSOT index decision table covers 12 stages + cross-cutting categories(§2.1, 20 rows total; trigger situations use the granularity of specific decision actions)
  • Reference relationship graph among SSOTs has no cycles(§3 + §3.2 self-check rules)
  • Boundary partner 3-scenario jumps explicitly stated(§3.1, apply W5_FIX_PLAN A.P1.1)
  • SSOT-AI-FAILURE-DEFENSES 6 Skill mandatory activation baseline explicitly stated(§4.1, apply W5_FIX_PLAN A.P2.2)
  • Cross-Skill reference syntax includes version number(§5.1, apply W5_FIX_PLAN A.P2.1)
  • SKILL.md main body < 500 lines(actually about 280 lines)
  • Does not contain README.md / XML angle brackets / uppercase underscore name

§8 Version and maintenance

Version Date Changes
1.0.0 2026-04-26 Initial version. 20 SSOTs (5 + 11 + 4) completed according to Layer 1.A/1.B/1.C.
1.1.0 2026-04-28 Wave 5.2.2-A revision: (1) apply the C5 section sequence template (simplified version for foundation Skills, §1-§8); (2) refine the index decision table entry granularity to "specific decision actions" (apply C3.1); (3) explicitly state boundary partner 3-scenario jumps (apply W5_FIX_PLAN A.P1.1); (4) make explicit the SSOT-AI-FAILURE-DEFENSES 6 Skill mandatory activation baseline (apply W5_FIX_PLAN A.P2.2); (5) include version numbers in cross-Skill reference syntax (apply W5_FIX_PLAN A.P2.1); (6) synchronize the §3 reference relationship graph with the three frontmatter field type layers (apply C7).

Next expected update triggers: (1) after Wave 2 is completed, supplement the precision of the §4.1 strong/weak reference matrix according to actual downstream Skill references; (2) any textbook publishes a new edition; (3) accumulated new LLM failure mode cases need to be incorporated into SSOT-AI-FAILURE-DEFENSES.