prepare-inspection-readiness
について
このスキルは、FDA、EMA、MHRAなどの特定の規制当局に対する準備状況を評価することで、組織が規制検査に備えることを支援します。警告文書や483所見の分析、模擬検査の調整、応答テンプレートを含む文書パッケージの作成を行います。予告された検査に直面したとき、定期的な自己評価を実施するとき、または過去の指摘後に新システムを導入した後の対応にご活用ください。
クイックインストール
Claude Code
推奨npx skills add pjt222/agent-almanac -a claude-code/plugin add https://github.com/pjt222/agent-almanacgit clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/prepare-inspection-readinessこのコマンドをClaude Codeにコピー&ペーストしてスキルをインストールします
ドキュメント
備檢查就緒
評並進組織受監管檢查之備度,含文件、人員備、事務、應對謀。
用時
- 監管檢查已宣或預期
- 定期檢查就緒自評
- 上檢以來已施新系或新程
- 業界趨(警告信、483)示新焦點區
- 重大審計發現或致監管注
入
- 必要:檢查機構(FDA、EMA、MHRA、或他)
- 必要:預期檢查範(數據完整、CSV、GMP、GLP、GCP)
- 必要:合規架構與系庫
- 可選:往檢報與觀察
- 可選:近業界警告信與 483 主題
- 可選:開放 CAPA 與審計發現
法
第一步:析機構特定焦點區
研檢查機構當前優先:
# Inspection Focus Area Analysis
## Document ID: IFA-[SITE]-[YYYY]-[NNN]
### FDA Current Focus Areas
| Focus Area | Regulatory Basis | Recent 483 Trends | Risk to This Site |
|-----------|-----------------|-------------------|-------------------|
| Data integrity | 21 CFR Part 11, CGMP | #1 cited observation in drug 483s since 2016 | [H/M/L] |
| Audit trail | 21 CFR 11.10(e) | Disabled/incomplete audit trails | [H/M/L] |
| Electronic signatures | 21 CFR 11.50-11.300 | Missing manifestation, shared accounts | [H/M/L] |
| Computer system validation | GAMP 5, FDA guidance | Inadequate validation of Category 4/5 systems | [H/M/L] |
| Change control | ICH Q10 | Undocumented or retrospective changes | [H/M/L] |
| Laboratory controls | 21 CFR 211.160-211.176 | OOS investigation, test repetition | [H/M/L] |
### EMA/MHRA-Specific Considerations
| Area | EU Annex 11 Reference | Focus | Risk to This Site |
|------|----------------------|-------|-------------------|
| Computerized system validation | §4, §5 | Risk-based validation, supplier assessment | [H/M/L] |
| Operational phase | §6-§13 | Security, audit trail, backup, business continuity | [H/M/L] |
| Cloud and outsourced systems | §3.4 | Service level agreements, data sovereignty | [H/M/L] |
| Data governance | MHRA DI guidance | ALCOA+, culture of integrity | [H/M/L] |
得:附險評之檢查焦點區析,限於檢機構。 敗則:若近 483/警告信數不可得,諮 FDA 警告信庫、EMA 檢報、業界刊以得最近趨。
第二步:行就緒自評
對各焦點區評廠:
# Inspection Readiness Assessment
## Document ID: IRA-[SITE]-[YYYY]-[NNN]
### Readiness Scoring
| Focus Area | Weight | Current State | Score (1-5) | Gap | Remediation Priority |
|-----------|--------|--------------|-------------|-----|---------------------|
| Data integrity controls | High | [Description of current state] | [1-5] | [Gap description] | [Critical/High/Medium/Low] |
| Validation documentation | High | [Description] | [1-5] | [Gap] | [Priority] |
| Audit trail compliance | High | [Description] | [1-5] | [Gap] | [Priority] |
| Electronic signatures | Medium | [Description] | [1-5] | [Gap] | [Priority] |
| Change control | Medium | [Description] | [1-5] | [Gap] | [Priority] |
| Training records | Medium | [Description] | [1-5] | [Gap] | [Priority] |
| SOPs (current, approved) | Medium | [Description] | [1-5] | [Gap] | [Priority] |
| CAPA management | Medium | [Description] | [1-5] | [Gap] | [Priority] |
| Vendor qualification | Low | [Description] | [1-5] | [Gap] | [Priority] |
| Periodic review | Low | [Description] | [1-5] | [Gap] | [Priority] |
Score: 1 = Critical gaps, 5 = Fully compliant
Overall readiness score: [Sum / Max] = [X]%
### Remediation Plan
| Gap ID | Description | Owner | Action | Deadline | Status |
|--------|------------|-------|--------|----------|--------|
| GAP-001 | [Gap] | [Name] | [Remediation action] | [Date] | [Open/In Progress/Closed] |
得:量化之就緒評,附分優之補救動。 敗則:若總就緒於 70% 下,考請延檢(若許)並施緊急補救。
第三步:備文件包
理文件為檢查就緒之包:
# Inspection Document Bundles
### Bundle 1: Validation Pack (per system)
- [ ] Current validation status summary (one-page per system)
- [ ] User Requirements Specification (URS)
- [ ] Risk Assessment
- [ ] Validation Plan
- [ ] IQ Protocol and Report
- [ ] OQ Protocol and Report
- [ ] PQ Protocol and Report (if applicable)
- [ ] Traceability Matrix
- [ ] Validation Summary Report
- [ ] Periodic review records
- [ ] Change control history since last validation
### Bundle 2: Data Integrity Evidence
- [ ] Data integrity policy and programme
- [ ] ALCOA+ assessment results
- [ ] Audit trail review records (last 12 months)
- [ ] Data integrity monitoring metrics and trends
- [ ] Data integrity training records
### Bundle 3: Operational Evidence
- [ ] Current SOPs (master list with effective dates)
- [ ] Training matrix (all GxP personnel)
- [ ] Change control log (last 24 months)
- [ ] Deviation/incident log (last 24 months)
- [ ] CAPA log with closure status
- [ ] Internal audit reports and CAPA follow-up
### Bundle 4: System Configuration Evidence
- [ ] User access list (current active users with roles)
- [ ] System configuration documentation
- [ ] Backup and recovery test records
- [ ] Security patch log
- [ ] Business continuity/disaster recovery plan
得:諸包皆備、索、可於檢者請後 30 分內取。 敗則:若文缺或不全,立缺列、分優補救、記謀。檢者察雜亂。
第四步:設模擬檢查協
# Mock Inspection Protocol
## Document ID: MIP-[SITE]-[YYYY]-[NNN]
### Scope
- **Focus areas:** [Top 3-5 risk areas from readiness assessment]
- **Systems in scope:** [Systems likely to be inspected]
- **Duration:** [1-2 days]
### Participants
| Role | Name | Mock Inspection Role |
|------|------|---------------------|
| Mock inspector | [Experienced QA or external consultant] | Ask questions, request documents |
| System owner(s) | [Names] | Respond to questions, demonstrate systems |
| QA | [Name] | Observe, note findings |
| Back room coordinator | [Name] | Locate and provide documents |
### Mock Inspection Scenarios
| Scenario | Focus | Inspector Might Ask |
|----------|-------|-------------------|
| 1: Show me the audit trail | Data integrity | "Show me the audit trail for batch record BR-2025-1234" |
| 2: Walk me through a change | Change control | "Show me the change control for the last system upgrade" |
| 3: Show training records | Training | "Show me the training records for user [Name] on system [X]" |
| 4: Explain your validation | CSV | "Walk me through how you validated this system" |
| 5: Show a deviation | CAPA | "Show me your last critical deviation and its CAPA" |
| 6: User access review | Access control | "Show me how you manage user access when people leave" |
### Post-Mock Assessment
| Scenario | Outcome | Findings | Actions |
|----------|---------|----------|---------|
| [#] | [Satisfactory/Needs Work] | [Description] | [Remediation if needed] |
得:模擬檢查於實檢前揭問題。 敗則:若模擬檢揭關鍵缺,當實檢觀察之急視之。
第五步:謀檢查事務
# Inspection Logistics Plan
### Room Setup
| Room | Purpose | Equipment | Assigned To |
|------|---------|-----------|-------------|
| Front room | Inspector workspace | Table, chairs, network access, printer | Facility manager |
| Back room | Document retrieval and strategy | Copier, network access, phone | QA team |
| Demo room | System demonstrations | Workstation with system access | IT support |
### Roles During Inspection
| Role | Person | Responsibilities |
|------|--------|-----------------|
| Inspection coordinator | [Name] | Single point of contact with inspector, schedule management |
| Subject matter experts | [Names] | Answer technical questions in their domain |
| Back room lead | [Name] | Coordinate document retrieval, track requests |
| Scribe | [Name] | Document all questions, requests, and responses |
| Executive sponsor | [Name] | Available for escalation, opening/closing meetings |
### Communication Protocol
- All document requests flow through the back room lead
- No documents provided without QA review
- Questions requiring research get a "we will get back to you" response (track and follow up)
- Daily debrief with inspection team after each day
得:事務謀確專業、有序之檢查應對。 敗則:若關鍵人於檢日不可得,識並簡介替代。
第六步:立應對模板
# Inspection Response Templates
### Template 1: 483 Observation Response
[Date]
[FDA District Office Address]
Re: FDA Form 483 Observations — [Inspection Dates] — [Facility Name]
Dear [Inspector Name],
We appreciate the opportunity to address the observations identified during the inspection of [facility] on [dates].
**Observation [N]:** [Quote the exact observation text]
**Response:**
- **Root Cause:** [Brief root cause description]
- **Immediate Corrective Action:** [What was done immediately]
- Completed: [Date]
- **Long-term Corrective Action:** [Systemic fix]
- Target completion: [Date]
- **Preventive Action:** [How recurrence will be prevented]
- Target completion: [Date]
- **Effectiveness Verification:** [How effectiveness will be measured]
- Target verification date: [Date]
### Template 2: Immediate Correction During Inspection
When an inspector identifies an issue that can be corrected immediately:
1. Acknowledge the observation
2. Implement the correction (if feasible)
3. Document the correction with before/after evidence
4. Inform the inspector that the correction has been made
5. Include in the formal response as "corrected during inspection"
得:應對模板使對檢查觀察之速、結構化回。 敗則:若應對模板通用而不對特定觀察,依特定證與時程定制各應對。
驗
- 機構特定焦點區已析附險評
- 就緒自評已完附量化分
- 諸缺有補救謀附主與期
- 諸範系之文件包已備並索
- 模擬檢查已行附記發現與隨訪
- 檢查事務謀定室、角、通信協
- 常見觀察型之應對模板已備
- 諸關鍵補救項於檢日前已閉
陷
- 末刻備:檢查就緒為續程,非臨時抱佛腳。倉促之組產雜亂、不全之應。
- 匿問題:檢者經驗豐能察隱。透明附明補救謀勝匿。
- 過自願訊:答所問。供未求之訊或開新查線。
- 未訓人員:未練應檢者問之專家表現劣。模擬檢為要練。
- 忽後室:後室(文件取回與策協)與前室同要。劣文件取造雜亂之印。
參
design-compliance-architecture— 檢者欲見之基文conduct-gxp-audit— 內審當仿檢方法monitor-data-integrity— 數據完整為 FDA 檢之首焦investigate-capa-root-cause— CAPA 必檢前徹查qualify-vendor— 廠商資質檢時常請
GitHub リポジトリ
関連スキル
llamaguard
その他LlamaGuardは、暴力やヘイトスピーチなど6つの安全性カテゴリーにおいて、LLMの入力と出力をモデレートするMetaの70-80億パラメータモデルです。94〜95%の精度を提供し、vLLM、Hugging Face、Amazon SageMakerを使用してデプロイ可能です。このスキルを使用して、AIアプリケーションにコンテンツフィルタリングと安全策を簡単に統合できます。
cost-optimization
その他このClaudeスキルは、リソースの適正サイジング、タグ付け戦略、支出分析を通じて、開発者がクラウドコストを最適化することを支援します。AWS、Azure、GCPにわたるクラウド支出の削減とコストガバナンスの実施のためのフレームワークを提供します。インフラコストの分析、リソースの適正サイジング、または予算制約への対応が必要な際にご利用ください。
quantizing-models-bitsandbytes
その他このスキルは、bitsandbytesを使用してLLMを8ビットまたは4ビット精度に量子化し、精度の低下を最小限に抑えつつ50〜75%のメモリ削減を実現します。限られたGPUメモリでより大規模なモデルを実行したり、推論を高速化するのに理想的で、INT8、NF4、FP4などのフォーマットをサポートしています。HuggingFace Transformersと統合され、QLoRAトレーニングや8ビットオプティマイザーを可能にします。
dispatching-parallel-agents
その他このClaudeスキルは、複数のエージェントを配備し、3つ以上の独立した問題を並行して調査・修正します。共有状態や依存関係がなく解決可能な、無関係な障害が発生するシナリオ向けに設計されています。中核となる機能は並列問題解決であり、効率を最大化するために独立した問題領域ごとに1つのエージェントを割り当てます。
