Best practice collection (retrieval → screening → adaptation)
When to use this skill
- Need to quickly find actionable industry best practices or success cases for domains such as architecture/process/security/operations.
- Need to determine whether a recommendation fits the current environment and provide an actionable implementation path with reference sources.
Required inputs
- Goals and decision scenarios: what problem to solve, what the success criteria are (SLO/compliance/cost/efficiency).
- Technical and business environment: language/framework/cloud provider/data sensitivity level/team size.
- Constraints and what cannot be done: budget, launch timeline, organizational policy, regional/compliance restrictions.
- Required output format: brief/comparison table/implementation checklist, whether bilingual Chinese-English is needed.
- Known sources or preferences: official white papers, standards, competitor/similar cases, internal baselines.
Workflow
- Rewrite the question and scope: state in one sentence the "best practice + applicable scenario" to look for, and clearly define exclusions.
- Design the search strategy: list 3–5 groups of keywords, including purpose terms (best practice/guideline/pattern/anti-pattern), objects (microservices, data governance, FinOps, etc.), and restrictions (cloud provider/region/version); combine operators and filters:
site:、filetype:pdf、intitle:"best practice", time range, and add language/region limits when necessary.
- Source priority：
- Official/authoritative: standards (ISO/RFC/NIST), cloud provider Well-Architected, official product guides and release notes.
- Real-world cases: conference talks, technical blogs, white papers, benchmark reports (note test conditions).
- Community signals: foundations such as CNCF/OWASP, mature project README/ADR, highly upvoted answers in discussion forums (record time and version).
- Quickly screen credibility: check publication date and version; clarify applicable scale/preconditions; identify marketing content and potential bias; exclude cases with incomparable data.
- Distill commonalities and disagreements: identify recommendations shared by 3 or more sources; flag approaches with disagreements and their underlying assumptions/risks; add common anti-patterns and how to avoid them.
- Adapt to the local environment: map recommendations to current constraints (team size, compliance, budget, SLO); list parameters that need adjustment or additional controls (e.g., access control, change process, rollback strategy).
- Output structure：
- TL;DR: 3–5 recommended actions + applicable conditions + expected outcomes.
- Consensus and differences: use a table to list approach/preconditions/benefits/risks/sources.
- Implementation path: provide an iterative plan, owners, and validation metrics by priority; flag assumptions to be validated.
- Source list: include links/publication dates/author organizations; clarify which are strong evidence and which are merely experience sharing.
Output format example
- Table:
做法 | 适用前提 | 主要收益 | 风险/限制 | 典型来源。
- Text example:
- TL;DR：
建议采用 X（来源：AWS Well-Architected 2023），因 <理由>；不建议 Y（来源：...），风险 <...>。
- Disagreement:
做法 A vs 做法 B，差异在 <前提>；若 <条件> 选 A，否则选 B。
Final checklist
- Each recommendation indicates its source and date; outdated versions or regional differences are noted.
- Applicable preconditions, exclusion conditions, and counterexamples/anti-patterns are clearly stated.
- The output includes actionable next steps (action items + owner + validation metrics), and points out evidence that still needs to be supplemented.
