@dzhechkov/keysarium 1.0.1 → 1.1.0

package/templates/.claude/shards/phase-2-research.shard.md

@@ -0,0 +1,42 @@
+# Phase 2: Research — Governance Shard
+
+## Time Budget
+15% of total timeline
+
+## Skill
+Load: `.claude/skills/goap-research-ed25519/SKILL.md`
+
+## Prerequisites
+- Phase 1 complete: `<promise>CASE_EXPLORED</promise>`
+- Files exist: `00_product_discovery.md`, `01_case_brief.md`
+
+## Mandatory Outputs
+- `02_research_findings.md`
+
+## PARANOID Mode — ACTIVE
+- Every factual claim MUST have a verifiable source
+- Zero tolerance for hallucinated citations
+- Confidence threshold: 0.99 minimum
+- Unverifiable claims: mark as [UNVERIFIED] and flag
+- Citation format: `[Claim] — Source: [URL or publication], [Date]`
+
+## Agent Swarm (3 parallel, model: sonnet)
+- Agent 1: Analogues + Benchmarks
+- Agent 2: Technologies + Anti-patterns
+- Agent 3: Regulations + Market Data
+- Synthesize results into single document
+
+## Quality Gates
+- ALL claims have sources
+- No unverified statistics or percentages
+- Competitor names verified to exist
+- Technology recommendations are for real, available products
+- Regulatory references cite actual laws/standards
+
+## Promise
+On completion: `<promise>RESEARCH_PARANOID_PASSED</promise>`
+
+## Anti-Patterns to Check
+- Bare claims without attribution → BLOCK
+- Generic "studies show" without specific study → BLOCK
+- Hallucinated product names → BLOCK

package/templates/.claude/shards/phase-25-cjm.shard.md

@@ -0,0 +1,47 @@
+# Phase 2.5: CJM Prototype — Governance Shard
+
+## MANDATORY — THIS PHASE CANNOT BE SKIPPED
+
+## Time Budget
+10% of total timeline
+
+## Skills
+- Primary: `.claude/skills/reverse-engineering-unicorn/SKILL.md` (module M2.5)
+- UI Design: `.claude/skills/frontend-design/SKILL.md`
+- Trend Research: `.claude/skills/goap-research-ed25519/SKILL.md` (for Variant D)
+
+## Prerequisites
+- Phase 2 complete: `<promise>RESEARCH_PARANOID_PASSED</promise>`
+- Files exist: `00_product_discovery.md`, `01_case_brief.md`, `02_research_findings.md`
+
+## Mandatory Outputs
+- `02.5_trend_brief.md` (Variant D trend research)
+- `prototype/cjm-prototype.jsx` (4-variant interactive prototype)
+
+## Process
+1. Extract: primary_user, segments, aha_moment, solution_concept from Phases 0-2
+2. Define 3 CJM variants (A/B/C) with different approaches
+3. Research 5 trend categories for Variant D (PARANOID mode)
+4. Generate React .jsx prototype with all 4 variants
+5. User selects winning variant → sets `{CHOSEN_CJM}`
+
+## Agent Swarm (3 parallel)
+- Agent 1 (opus): CJM Variant A design
+- Agent 2 (opus): CJM Variants B + C design
+- Agent 3 (sonnet): Trend Research for Variant D (PARANOID mode, 5 categories)
+
+## Critical Variable
+`{CHOSEN_CJM}` — MUST be set by user selection. Passed to Phases 3-5.
+
+## Quality Gates
+- Minimum 3 CJM variants (A/B/C) + 1 trend variant (D)
+- Prototype must be valid React JSX
+- Trend research in PARANOID mode (verified sources)
+- User MUST explicitly select a variant
+
+## Promise
+On completion: `<promise>CJM_VALIDATED</promise>`
+
+## Invariant
+If any attempt is made to skip this phase → BLOCK immediately.
+"Skipping CJM" is listed as a FORBIDDEN anti-pattern.

package/templates/.claude/shards/phase-3-solve.shard.md

@@ -0,0 +1,47 @@
+# Phase 3: Solve — Governance Shard
+
+## Time Budget
+15% of total timeline
+
+## Skill
+Load: `.claude/skills/problem-solver-enhanced/SKILL.md` (TRIZ + Game Theory)
+
+## Prerequisites
+- Phase 2.5 complete: `<promise>CJM_VALIDATED</promise>`
+- `{CHOSEN_CJM}` is set
+- Files exist: `00_` through `02.5_`
+
+## Mandatory Outputs
+- `03_solution_strategy.md`
+- `diagrams/process-as-is.mermaid`
+- `diagrams/process-to-be.mermaid`
+
+## Content Requirements
+- SCQA (Situation-Complication-Question-Answer) for pitch
+- Concept + Elevator Pitch + Key Innovation
+- Process Design: As-Is → To-Be (using `{CHOSEN_CJM}`)
+- User Flow from `{CHOSEN_CJM}`
+- AI Pipeline specification
+- Human-in-the-Loop design with escalation policy
+- Success Metrics with baseline values
+- Risks & Mitigation matrix
+- Roadmap: PoC → MVP → Scale
+
+## Agent Swarm (2 parallel)
+- Agent 1 (opus): SCQA + Concept + Process Design + User Flow + AI Pipeline
+- Agent 2 (sonnet): Mermaid Diagrams + HITL + Metrics + Roadmap
+
+## Quality Gates
+- SCQA must be concise and compelling
+- Metrics must include baseline AND target values
+- HITL escalation policy is MANDATORY (no AI-only decisions)
+- Mermaid diagrams must be valid syntax
+- `{CHOSEN_CJM}` must appear in User Flow section
+
+## Promise
+On completion: `<promise>SOLUTION_DESIGNED</promise>`
+
+## Anti-Patterns to Check
+- No HITL → BLOCK
+- Vague "improve efficiency" without numbers → BLOCK
+- Over-engineering (> 10 components in MVP) → FLAG

package/templates/.claude/shards/phase-4-architecture.shard.md

@@ -0,0 +1,41 @@
+# Phase 4: Architecture — Governance Shard
+
+## Time Budget
+15% of total timeline
+
+## Skill
+Built-in templates (no external skill)
+
+## Prerequisites
+- Phase 3 complete: `<promise>SOLUTION_DESIGNED</promise>`
+- Files exist: `00_` through `03_`
+
+## Mandatory Outputs
+- `04_architecture.md`
+- `diagrams/architecture-c4.mermaid` (C4 Context diagram)
+- `diagrams/sequence-main-flow.mermaid` (Sequence diagram)
+
+## Content Requirements
+- C4 architectural diagram
+- Components table: tech + purpose + justification
+- AI Models & Pipeline: task, type, concrete model name
+- Data Architecture: sources, ETL, storage, Vector DB
+- Integrations: system + protocol
+- Security & Compliance: controls, RBAC, audit, regulations
+- Scalability: latency target, throughput, scaling strategy
+- MVP Scope definition
+
+## Domain-Specific Rules
+- Banking: on-premise LLM (GigaChat/YandexGPT/open-source), ФЗ-152, ЦБ, ФСТЭК
+- Retail: latency < 200ms, A/B testing infrastructure
+- Enterprise: Change Management plan, Legacy integration, SLA definition
+- Healthcare: HITL for ALL clinical decisions, ФЗ-323
+
+## Quality Gates
+- Minimum 2 Mermaid diagrams (C4 + sequence)
+- All model names must be real, available products
+- Security section must reference applicable regulations
+- MVP scope must be achievable (not over-engineered)
+
+## Promise
+On completion: `<promise>ARCHITECTURE_DEFINED</promise>`

package/templates/.claude/shards/phase-5-presentation.shard.md

@@ -0,0 +1,56 @@
+# Phase 5: Presentation — Governance Shard
+
+## Time Budget
+20% of total timeline (largest allocation)
+
+## Skill
+Load: `.claude/skills/presentation-storyteller/SKILL.md`
+
+## Prerequisites
+- Phase 4 complete: `<promise>ARCHITECTURE_DEFINED</promise>`
+- `{CHOSEN_CJM}` is set (for Slide 5)
+- Files exist: `00_` through `04_`
+
+## Mandatory Outputs (ALL 4 files required)
+- `05_presentation_content.md` (10-12 slides with storytelling)
+- `06_speaker_script.md` (per-slide: HOOK, CONTENT, BRIDGE, TONE, KEY MESSAGE)
+- `07_qa_preparation.md` (7+ typical jury questions + answers)
+- `08_executive_summary.md` (1 page for jury) ⭐ MANDATORY
+
+## 08_executive_summary.md is NON-NEGOTIABLE
+Missing Executive Summary is a FORBIDDEN anti-pattern. BLOCK if not created.
+
+## Slide Structure (10-12 slides)
+1. Title + Team + Case
+2. Problem (SCQA, cost of inaction)
+3. Key Question + Elevator Pitch
+4. Concept (high-level, AI role)
+5. User Flow / To-Be (`{CHOSEN_CJM}`)
+6. AI Under the Hood (models, RAG/Agent, HITL)
+7. Data Architecture
+8. Security & Compliance
+9. Metrics & ROI (As-Is → Target)
+10. Roadmap (PoC → MVP → Scale)
+11. Team + CTA
+12. Q&A
+
+## Agent Swarm (3 parallel)
+- Agent 1 (opus): Presentation content (05) — storytelling
+- Agent 2 (sonnet): Speaker script (06)
+- Agent 3 (sonnet): Q&A + Executive Summary (07 + 08)
+
+## Quality Gates
+- No slide reading (text-heavy slides) → FLAG
+- Every slide must have a visual anchor suggestion
+- Speaker script must tell a connected story across slides
+- Q&A must include tough/adversarial questions
+- Executive Summary must fit on 1 page
+
+## Promise
+On completion: `<promise>PRESENTATION_READY</promise>`
+
+## Anti-Patterns to Check
+- Text-heavy slides → FLAG
+- No storytelling arc → FLAG
+- Missing Executive Summary → BLOCK
+- Generic phrases without numbers → FLAG

package/templates/.claude/skills/explore/SKILL.md

@@ -3,11 +3,14 @@ name: explore
 description: >
   Adaptive task exploration and clarification skill for transforming vague 
   requests into actionable specifications. Use when a user presents any task, 
-  problem, or goal that requires clarification before execution. Triggers on 
-  ambiguous requests, complex multi-part tasks, "I want to...", "help me with...", 
-  strategic decisions, product ideas, creative briefs, and any situation where 
-  understanding the real need is essential before proposing solutions. Does NOT 
+  problem, or goal that requires clarification before execution. Triggers on
+  ambiguous requests, complex multi-part tasks, "I want to...", "help me with...",
+  strategic decisions, product ideas, creative briefs, and any situation where
+  understanding the real need is essential before proposing solutions. Does NOT
   provide solutions until task is fully explored.
+trust_tier: 1
+trust_tier_label: "Structured"
+trust_tier_path: "Run /bto-test to promote to Tier 2"
 ---
 
 # Explore: Adaptive Task Clarification

package/templates/.claude/skills/frontend-design/SKILL.md

@@ -2,6 +2,9 @@
 name: frontend-design
 description: Create distinctive, production-grade frontend interfaces with high design quality. Use this skill when the user asks to build web components, pages, artifacts, posters, or applications (examples include websites, landing pages, dashboards, React components, HTML/CSS layouts, or when styling/beautifying any web UI). Generates creative, polished code and UI design that avoids generic AI aesthetics.
 license: Complete terms in LICENSE.txt
+trust_tier: 0
+trust_tier_label: "Advisory"
+trust_tier_path: "Add references/ directory and run /bto-test to promote to Tier 1"
 ---
 
 This skill guides creation of distinctive, production-grade frontend interfaces that avoid generic "AI slop" aesthetics. Implement real working code with exceptional attention to aesthetic details and creative choices.

package/templates/.claude/skills/goap-research-ed25519/SKILL.md

@@ -1,6 +1,9 @@
 ---
 name: goap-research-ed25519
 description: Advanced GOAP research system with Ed25519 cryptographic verification for anti-hallucination protection. Combines Goal-Oriented Action Planning with cryptographic signatures to ensure source authenticity, claim verification, and citation chain integrity. Use for high-stakes research requiring verifiable facts, competitive intelligence, legal/medical research, or any context where hallucination prevention is critical. Triggers on "verified research", "trusted sources only", "anti-hallucination", "signed sources", "cryptographic verification".
+trust_tier: 1
+trust_tier_label: "Structured"
+trust_tier_path: "Run /bto-test to promote to Tier 2"
 ---
 
 # GOAP Research Skill with Ed25519 Verification

package/templates/.claude/skills/presentation-storyteller/SKILL.md

@@ -6,9 +6,12 @@ description: >
   в Markdown со ссылками на источники, (3) Подстрочник "Как рассказывать?" со связной 
   историей для каждого слайда. Использует explore (уточнение задачи), goap-research-ed25519 
   (исследование с верификацией), storytelling-фреймворки (AIDA, Hero's Journey, Problem-Solution).
-  Выходные файлы: presentation-outline.md (план), presentation-full.md (презентация), 
-  sources-index.md (индекс источников). Триггеры: "сделай презентацию", "presentation with 
+  Выходные файлы: presentation-outline.md (план), presentation-full.md (презентация),
+  sources-index.md (индекс источников). Триггеры: "сделай презентацию", "presentation with
   storytelling", "презентация со скриптами", "sales deck", "pitch deck", "продающая презентация".
+trust_tier: 1
+trust_tier_label: "Structured"
+trust_tier_path: "Run /bto-test to promote to Tier 2"
 ---
 
 # Presentation Storyteller: Sales Decks with Verified Sources

package/templates/.claude/skills/problem-solver-enhanced/SKILL.md

@@ -1,6 +1,9 @@
 ---
 name: problem-solver-enhanced
 description: Elite problem-solving system combining first principles thinking, game theory, root cause analysis, systematic frameworks AND TRIZ inventive principles for solving complex challenges. Use when facing complex challenges with unclear solutions, strategic decisions with multiple stakeholders, problems requiring breakthrough (not incremental) solutions, trade-offs that seem impossible to resolve, competitive scenarios requiring game-theoretic analysis, system design problems, or any situation requiring both analytical rigor and inventive thinking beyond surface-level solutions.
+trust_tier: 1
+trust_tier_label: "Structured"
+trust_tier_path: "Run /bto-test to promote to Tier 2"
 ---
 
 # Elite Problem Solver (Enhanced with TRIZ)

package/templates/.claude/skills/reverse-engineering-unicorn/SKILL.md

@@ -1,6 +1,9 @@
 ---
 name: reverse-engineering-unicorn
 description: Reverse engineer any company into a launch playbook with clickable CJM prototype. Triggers on "проанализируй компанию", "reverse engineer", "разбери бизнес-модель", "playbook запуска".
+trust_tier: 1
+trust_tier_label: "Structured"
+trust_tier_path: "Run /bto-test to promote to Tier 2"
 ---
 
 # Reverse Engineering Unicorn v2

package/templates/CLAUDE.md

@@ -11,6 +11,8 @@
 /new-research [название]            — Создать новое исследование
 /parallel-research кейс1 | кейс2   — Запустить несколько параллельно
 /harvest [путь или "all"]           — Извлечь знания после проекта
+/brain-export [путь или "all"]      — Экспорт знаний в portable JSON
+/brain-import [путь к JSON]         — Импорт знаний из другого проекта
 
 /bto [путь или описание]           — Build-Test-Optimize полный цикл
 /bto-build [описание]              — Генерация нового скилла/команды
@@ -37,10 +39,21 @@ product-keysarium-2026/
 │   │   ├── architecture-phase.md   ← /architecture-phase — Phase 4
 │   │   ├── presentation.md         ← /presentation — Phase 5
 │   │   ├── harvest.md              ← /harvest — извлечение знаний
+│   │   ├── brain-export.md         ← /brain-export — экспорт portable brain
+│   │   ├── brain-import.md         ← /brain-import — импорт знаний
 │   │   ├── bto.md                  ← /bto — Build-Test-Optimize pipeline
 │   │   ├── bto-build.md            ← /bto-build — генерация артефактов
 │   │   ├── bto-test.md             ← /bto-test — мульти-агентная оценка
 │   │   └── bto-optimize.md         ← /bto-optimize — оптимизация промптов
+│   ├── shards/                     ← Governance shards (per-phase rules)
+│   │   ├── phase-0-discovery.shard.md
+│   │   ├── phase-1-explore.shard.md
+│   │   ├── phase-2-research.shard.md
+│   │   ├── phase-25-cjm.shard.md
+│   │   ├── phase-3-solve.shard.md
+│   │   ├── phase-4-architecture.shard.md
+│   │   ├── phase-5-presentation.shard.md
+│   │   └── bto-evaluation.shard.md
 │   └── skills/                     ← Навыки (building blocks)
 │       ├── bto/                    ← Build-Test-Optimize (мульти-агентная оценка)
 │       ├── explore/                ← Адаптивная кларификация задач
@@ -93,8 +106,47 @@ DISCOVERY → EXPLORE  → RESEARCH → CJM PROTO    → SOLVE    → ARCHITECTU
 3. **08_executive_summary.md — ОБЯЗАТЕЛЕН**
 4. **Research в PARANOID mode** — ноль непроверенных утверждений
 5. **{CHOSEN_CJM} передаётся** из Phase 2.5 в Phase 3-5
-6. **Checkpoint после каждой фазы** — ждём "ок" от пользователя
+6. **Checkpoint после каждой фазы** — ждём "ок" от пользователя + `<promise>` тег
 7. **Всё в researches/<slug>/** — не засоряем корень
+8. **Governance shards** — каждая фаза загружает свой shard из `.claude/shards/`
+9. **Model routing** — haiku для Layer 0-1, sonnet для research/judges, opus для creative work
+10. **Trust tiers** — скиллы классифицированы (Tier 0-3), повышение через `/bto-test`
+
+## Governance System
+
+### Shards
+Каждая фаза загружает свой governance shard из `.claude/shards/`:
+- Shard содержит: time budget, skill to load, prerequisites, quality gates, promise tag
+- Решает проблему context drift при длинных сессиях
+- Агент перечитывает shard перед началом работы фазы
+
+### Semantic Completion Promises
+Каждый checkpoint включает `<promise>` тег:
+- `DISCOVERY_COMPLETE`, `CASE_EXPLORED`, `RESEARCH_PARANOID_PASSED`
+- `CJM_VALIDATED`, `SOLUTION_DESIGNED`, `ARCHITECTURE_DEFINED`, `PRESENTATION_READY`
+- Downstream фазы проверяют upstream promises перед стартом
+
+### Trust Tiers
+| Tier | Label | Требования |
+|------|-------|-----------|
+| 3 | Verified | Eval test suites + deterministic validation |
+| 2 | Validated | /bto-test Layer 2 score >= 7.0 |
+| 1 | Structured | SKILL.md + references/ or modules/ |
+| 0 | Advisory | Только SKILL.md |
+
+### Cross-Phase Feedback Loops
+6 именованных loops формализуют передачу данных между фазами.
+См. `.claude/rules/feedback-loops.md` для Variable Registry.
+
+## Portable Brain
+
+```
+/brain-export [путь или "all"]  — Экспорт знаний в JSON контейнер
+/brain-import [файл.json]       — Импорт знаний из другого проекта
+```
+
+Brain container включает: skill metadata, domain patterns, research summaries,
+harvest patterns, pipeline metrics. Selective import с merge-not-overwrite стратегией.
 
 ## Domain Templates
 
@@ -168,8 +220,8 @@ bto (Build-Test-Optimize)
 ### Standalone пакет
 
 ```bash
-npx @dzhechko/skills-bto init    # Установить BTO в любой проект
-npx @dzhechko/skills-bto doctor  # Проверить здоровье установки
+npx @dzhechkov/skills-bto init    # Установить BTO в любой проект
+npx @dzhechkov/skills-bto doctor  # Проверить здоровье установки
 ```
 
 ## Anti-Patterns