drift-check
Plan-alignment verification — approved plan vs actual implementation, reporting unimplemented items, out-of-scope changes and design drift.
Skill by Sungmin-Cho
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/sungmin-cho-drift-check
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: design, testing
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/drift-check
What's inside
- Skill name
drift-check- Description
- Plan-alignment verification — approved plan vs actual implementation, reporting unimplemented items, out-of-scope changes and design drift. Read-only. Triggers on `/drift-check`, "plan drift", "drift detection", "plan alignment", "플랜 드리프트", "계획 정합성", "드리프트 검증", or `/deep-test` running it as the Required Gate.
- License
- MIT
- Instructions
- 212 lines in SKILL.md · view on GitHub
Files (1)
- SKILL.md
SKILL.md (first lines)
## Inputs (skill args)
| 인자 | 의미 |
|---|---|
| (없음) | 활성 세션의 plan.md 사용 |
| `<plan-file-path>` | 명시된 plan 파일 경로 |
빈 args / 매칭되지 않는 토큰 → 본문의 default 분기로 진입.
## Prerequisites
> **Quality Gate** — `/deep-test`가 Required Gate로 자동 실행합니다. 특정 plan 파일에 대한 독립 검증이 필요할 때 직접 사용하세요.
> Standalone: `/drift-check [plan-file]`
# Plan Alignment Check (Drift Detection)
You are performing a **Plan Alignment Check** — comparing the approved plan against actual implementation to detect drift.
## Language
Read(`${CLAUDE_PLUGIN_ROOT}/skills/shared/references/user-language.md`) and follow it.
## Critical Constraints
- **DO NOT modify any code files.** This is a verification-only operation.
- **Read plan, read code, compare, and report findings.**
- **Save results to file when in workflow mode.**
## Instructions
### 1. Determine operating mode and load plan
Read(`${CLAUDE_PLUGIN_ROOT}/skills/deep-resume/references/session-detection.md`) and apply only its **Reusable session-state resolution** section to resolve `$STATE_FILE`; retain this skill's own no-session and standalone-mode behavior.
Check if `$STATE_FILE` exists and has an active session (`current_phase` is not `idle`).
**Workflow Mode** (active deep-work session):
- Read `work_dir` from the state file
- Set `WORK_DIR` to the value of `work_dir`
- Load `$WORK_DIR/plan.md` as the plan
- Get comparison baseline: `plan_approved_at` from state file, or fall back to the commit closest to plan.md's modification time
**Standalone Mode** (no active session):
- If `$ARGUMENTS` is provided: use as plan file path
- If `$ARGUMENTS` is empty: error — "Plan 파일 경로를 지정해주세요. 예: /drift-check path/to/plan.md"
### 2. Extract plan items
Parse the plan file and extract 3 categories:
**A. 파일 목록** — "Files to Modify", "수정 대상 파일", "File Changes" 등의 섹션에서 파일 경로 추출.
**B. 구현 항목** — 체크리스트(`- [ ]`, `- [x]`), 번호 매긴 단계, 또는 "Implementation Steps" 섹션의 각 항목. 각 항목에서 다음을 추출:
- 대상 파일 (있으면)
- 구현할 내용 요약
- plan 내 위치 (섹션 번호/이름)
**C. 설계 결정** — "Design Decisions", "설계 지침", "Architecture", "Constraints" 등의 섹션에서 명시적 설계 지시사항 추출. 예:
- "Strategy 패턴으로 구현"
- "인터페이스를 통한 의존성 주입"
- "기존 XXX 클래스는 수정하지 않음"
Display extracted items:
```
Plan 항목 추출 완료:
파일: [N]개
구현 항목: [N]개
설계 결정: [N]개
```
### 3. Collect actual implementation
```bash
# Plan 승인 이후 변경된 파일 목록
git diff --name-only [baseline-commit]..HEAD
# 변경 내용 상세 (각 파일)
git diff [baseline-commit]..HEAD -- [file]
```
baseline-commit 결정:
1. `plan_approved_at` 타임스탬프가 있으면 → `git log --before="[timestamp]" -1 --format=%H`
2. 없으면 → plan.md 파일의 mtime 기준 가장 가까운 커밋
3. 둘 다 없으면 → `HEAD~10` (최근 10커밋과 비교, 경고 표시)
### 4. Compare and classify
각 Plan 항목을 다음 중 하나로 분류:
| 분류 | 조건 | 기호 |
|------|------|------|
| **구현 완료** | Plan 항목이 실제 코드에 반영됨 | PASS |
| **미구현** | Plan에 있지만 코드에 없음 | FAIL |
| **범위 초과** | Plan에 없지만 코드에 있음 (신규 파일/대규모 변경) | WARN |
| **설계 이탈** | Plan의 설계 결정과 다르게 구현됨 | FAIL |
**비교 방법**:
**파일 수준**: Plan의 파일 목록과 git diff의 파일 목록을 집합 비교.
- Plan에만 있는 파일 → 미구현
- git diff에만 있는 파일 → 범위 초과 후보 (테스트 파일, 설정 파일 등은 자동 제외)
**항목 수준**: 각 구현 항목에 대해 관련 파일의 diff를 읽고, 해당 항목이 구현되었는지 판단.
- 클래스/함수/인터페이스 생성 항목 → 해당 심볼이 코드에 존재하는지 확인
- 로직 변경 항목 → diff에서 관련 변경이 있는지 확인
- 판단이 어려운 항목 → "확인 필요"로 표시 (false negative 방지)
**설계 수준**: 각 설계 결정에 대해 실제 코드가 해당 결정을 따르는지 확인.
- "Strategy 패턴" 지시 → 인터페이스 + 구현체 분리가 되어있는지
- "기존 XXX 수정 금지" → 해당 파일이 diff에 없는지
- "인터페이스 의존" → new ConcreteClass() 대신 인터페이스 참조인지
### 5. Generate report
```markdown
## Plan Alignment Report
**Plan**: [plan file path]From Sungmin-Cho/deep-work, MIT.
More from Sungmin-Cho/deep-work
- 0.630score
deep-assumptions
Sungmin-Cho
Assumption health report — per-assumption Wilson Score confidence, verdict (justified / loosen / drop), model-aware split, decay history.
SkillAFreshness: A★ 10Claude Code +9 - 0.595score
deep-status
Sungmin-Cho
Current deep-work session status — dashboard, badge, tree, or routing to the receipts / history / report / assumptions sub-pages.
SkillBFreshness: B★ 10Claude Code +9 - 0.594score
deep-resume
Sungmin-Cho
Use when resuming an interrupted or parked deep-work session from its current phase.
SkillBFreshness: B★ 10Claude Code +9 - 0.594score
deep-finish
Sungmin-Cho
Use when finishing an authorized deep-work session with keep, PR, merge or discard and a durable receipt.
SkillBFreshness: B★ 10Claude Code +9 - 0.590score
deep-insight
Sungmin-Cho
Code-metrics, complexity and dependency report for a file/directory/glob — analysis only, never modifies code.
SkillBFreshness: B★ 10Claude Code +9 - 0.590score
deep-debug
Sungmin-Cho
Debug sub-mode for Phase 3 — Investigate → Analyze → Hypothesize → Implement, with the iron rule that no fix ships before root-cause investigation.
SkillBFreshness: B★ 10Claude Code +9 - 0.590score
deep-mutation-test
Sungmin-Cho
Mutation testing on changed files to verify AI-generated test quality, with an auto-fix loop that returns to Implement for survived mutants.
SkillBFreshness: B★ 10Claude Code +9 - 0.590score
deep-receipt
Sungmin-Cho
View, dashboard, or export deep-work slice receipts (`receipts/SLICE-*.json`).
SkillBFreshness: B★ 10Claude Code +9 - 0.590score
deep-history
Sungmin-Cho
Cross-session deep-work history — completed sessions, TDD compliance rates, model usage, aggregate trend indicators.
SkillBFreshness: B★ 10Claude Code +9 - 0.590score
deep-work-workflow
Sungmin-Cho
High-level overview of the deep-work workflow (Brainstorm → Research → Spec → Plan → Implement → Test → Integrate).
SkillBFreshness: B★ 10Claude Code +9 - 0.590score
deep-sensor-scan
Sungmin-Cho
Run computational sensors (linter, type checker, coverage) inside or outside a deep-work session.
SkillBFreshness: B★ 10Claude Code +9 - 0.590score
solid-review
Sungmin-Cho
SOLID design-principles review (SRP / OCP / LSP / ISP / DIP) on a file/directory/glob.
SkillBFreshness: B★ 10Claude Code +9
Signals
- Stars
- 10
- Forks
- 2
- Open issues
- 1
- Last commit (this folder)
- Aug 13, 2026 (2 months ago)
- Last release
- Sep 20, 2026 (1 month ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/Sungmin-Cho/deep-work
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from Sungmin-Cho
- 0.630score
deep-assumptions
Sungmin-Cho
Assumption health report — per-assumption Wilson Score confidence, verdict (justified / loosen / drop), model-aware split, decay history.
SkillAFreshness: A★ 10Claude Code +9 - 0.595score
deep-status
Sungmin-Cho
Current deep-work session status — dashboard, badge, tree, or routing to the receipts / history / report / assumptions sub-pages.
SkillBFreshness: B★ 10Claude Code +9 - 0.594score
deep-finish
Sungmin-Cho
Use when finishing an authorized deep-work session with keep, PR, merge or discard and a durable receipt.
SkillBFreshness: B★ 10Claude Code +9 - 0.594score
deep-resume
Sungmin-Cho
Use when resuming an interrupted or parked deep-work session from its current phase.
SkillBFreshness: B★ 10Claude Code +9 - 0.590score
deep-mutation-test
Sungmin-Cho
Mutation testing on changed files to verify AI-generated test quality, with an auto-fix loop that returns to Implement for survived mutants.
SkillBFreshness: B★ 10Claude Code +9 - 0.590score
deep-debug
Sungmin-Cho
Debug sub-mode for Phase 3 — Investigate → Analyze → Hypothesize → Implement, with the iron rule that no fix ships before root-cause investigation.
SkillBFreshness: B★ 10Claude Code +9
Cite this page
drift-check skill. Agent Mart. https://agentmart.dev/skills/sungmin-cho-drift-check (last updated 2026-10-24)