Skill: Migrate CLAUDE.md

Purpose: Bring existing CLAUDE.md to current intake standard without losing data. When: After validate-claude-md found issues; on standard change; onboarding old project. Output: Updated CLAUDE.md with added sections and fixed placeholders. Each change confirmed.

PROCESS

Step 1 — Read and prepare plan

1. Read CLAUDE.md fully
2. Check if plan.md already exists in same directory
3. Read ~/.claude/skills/intake/SKILL.md — OUTPUT TEMPLATE section (current standard)
4. List discrepancies in three categories:

• Plan extraction — ## PHASES section exists in CLAUDE.md → must be moved to plan.md
• Structural — missing sections (QUICK CONTEXT, INFRA, SECURITY, PLAN line)
• Format — Russian prose where English key-value is expected

Show plan before any changes:

Найдено несоответствий: N

ПЛАН:
1. Извлечь ## PHASES из CLAUDE.md → создать plan.md (новый формат с YAML frontmatter)
2. Убрать ## PHASES из CLAUDE.md, добавить строку PLAN: → plan.md в QUICK CONTEXT

СТРУКТУРНЫЕ:
3. Добавить SECURITY (первой секцией, с OPERATOR_TOKEN)
4. Добавить QUICK CONTEXT блок (9 строк key-value)
5. Добавить INFRA строку в QUICK CONTEXT
...

ФОРМАТ:
6. Session Log: сжать в однострочный формат S[N] [date]: [summary]
7. ADR таблица: перенести в docs/adr.md если >3 записей
...

Применяю по одному. Начать?

Step 2 — Apply changes one by one

For each missing section:

1. Show what you're adding:

── Добавление N/M: [section name] ──────────────

ДОБАВЛЯЮ:
[new section content, filled with data from existing CLAUDE.md]

Куда: [after section X / before section Y]
Данные взяты из: [source]

Добавить? (да / нет / изменить)

2. When adding SECURITY:

• Generate unique OPERATOR_TOKEN (4 words: adjective-noun-noun-number)
• Adapt stop-ops to project stack (from Architecture / Dependencies)
• Insert as FIRST section

3. When adding missing sections with data:

• Extract from existing sections (e.g., Dependencies from Architecture)
• No empty placeholders — only concrete data or explicit TODO

4. When updating stale data (⚠):

• Show BEFORE / AFTER
• Explain where updated data comes from

Step 3 — Final validation

After all changes — run validate-claude-md:

── Финальная валидация ──────────────────────────
[validate-claude-md output]

If ⚠ or ✗ remain — offer to fix manually with hint.

Step 4 — Report

── Migrate CLAUDE.md — Итог ────────────────────

Добавлено секций:   N
Обновлено полей:    M
Пропущено:          K

CLAUDE.md приведён к стандарту intake ✓

RULES

• Never delete existing content — only add and fix
• No empty placeholders — if data missing, ask user
• Don't change project-specific sections (Detectors Map, Stage Plan, etc.) — project specifics
• OPERATOR_TOKEN — always new, unique; don't copy from other projects
• One change at a time — don't combine sections in one step

BATCH MODE

When: target format already known; 5+ files to migrate with identical rules. Difference from normal mode: no per-file confirmation — applies rules silently, summary at end.

Trigger

Use Batch Mode when user says:

• "migrate all files in [folder]"
• "apply same format to N files"
• "batch migrate, no questions"
• explicitly: "just apply, don't ask for each"

Batch Process

Step B1 — Confirm rules once:

Batch migration: [N] files
Rules:
1. [rule — e.g. Translate RU prose → EN]
2. [rule — e.g. Keep RU user-facing strings]
3. [rule — e.g. Move heavy tables → docs/]

Apply to all [N] files? (да / нет)

One confirmation for all files. Wait for "да".

Step B2 — Process silently: For each file: Read → apply rules → Write. No diff shown per file. On error — log and continue.

Step B3 — Summary:

── Batch Migrate ────────────────────────────
Done:    N files
Skipped: M (errors below)
Errors:  [file]: [reason]

Files: ✓ [file1] ✓ [file2] ...

NEW FORMAT RULES

When converting to current standard:

PHASES → plan.md extraction:

If ## PHASES section exists in CLAUDE.md:

1. Check if plan.md already exists in same directory:

• Exists → skip creation, only remove PHASES from CLAUDE.md and add PLAN line
• Not found → create plan.md from PHASES content

1. Convert PHASES content to plan.md format:

   BEFORE (in CLAUDE.md):
   ## PHASES
   > Gate between phases — do NOT proceed without explicit "go"
   Phase 1: Foundation — [ ] pending
   - [ ] task A
   - [ ] Gate: criterion

   AFTER (plan.md):
   ---
   project: "[name from CLAUDE.md PROJECT line]"
   updated: YYYY-MM-DD
   session: [session number from CLAUDE.md SESSION line]
   active_phase: [first phase without all [x]]
   ---

   ## Phase 1: Foundation — active
   - [ ] task A
   **Gate:** criterion

1. Conversion rules:

• Phase N: Name — [ ] pending → ## Phase N: Name — pending
• Phase N: Name — [x] done → ## Phase N: Name — done
• First phase not fully done → status active, others stay pending/done
• - [ ] Gate: criterion → **Gate:** criterion (remove checkbox, bold prefix)
• > Gate between phases... line → drop entirely (redundant)
• active_phase = N of first phase with status active

1. After creating plan.md — remove entire ## PHASES section from CLAUDE.md
2. Add PLAN: → plan.md line to QUICK CONTEXT (after IN_PROG line)

Language: all new content in English. Existing Russian prose → convert when compressing.

Session Log compression:

BEFORE (multi-line block):
### 2026-03-10 — Session 16
DONE: implemented 3-screen architecture...
NEXT: Stage 13

AFTER (one line):
S16 2026-03-10: 3-screen arch, pos_map pattern | next: Stage 13

Keep only last 5 sessions in CLAUDE.md. Others → offer to move to docs/session-log.md.

ADR compression: If ADR table > 3 rows → create docs/adr.md, move there, leave in CLAUDE.md:

ADR: → docs/adr.md

QUICK CONTEXT: Fill from existing CLAUDE.md sections. INFRA line: take from ENV, ENVIRONMENT, or README sections. If INFRA not found anywhere — ask user one question.