AgentSkillsCN

edgecase-plan

在功能落地之前，先制定一份按风险等级排序的边缘场景预案。适用于功能规划、生产环境加固、API 设计、状态机设计、事故预防、悲情路径分析、失效模式枚举、鲁棒性规划，或当用户提及边缘场景、错误处理、失效场景、风险矩阵、安全网、防御性编程、异常路径，以及生产就绪性审查时使用。

SKILL.md
--- frontmatter
name: edgecase-plan
description: Produce a risk-ranked edge-case plan before implementing features. Use for feature planning, production hardening, API design, state-machine design, incident prevention, sad path analysis, failure mode enumeration, robustness planning, or when user mentions edge cases, error handling, failure scenarios, risk matrix, safety nets, defensive programming, unhappy paths, or production readiness review.

Edge Case Planning Skill

Create an Edge Case Register + Safety Nets Plan for the specified feature or scope.

Mode: Planning only. Do NOT modify files. Use Read/Grep/Glob to gather context.

Inputs to Gather

Ask if missing:

  1. •Feature statement - what changes, who uses it, success criteria
  2. •Surfaces - UI routes, API endpoints, jobs, DB tables, external deps
  3. •Constraints - latency, scale, cost, SLOs
  4. •Risk tolerance - P0 vs acceptable degradation
  5. •Rollout - flagged? staged? backwards compatibility?

If code paths provided, scan relevant files for context.

Core Strategy

Handle "infinite" edge cases through:

  • •Categorize failures into repeatable classes (see references/taxonomy.md)
  • •Prioritize using Probability × Impact scoring
  • •Add safety nets for low-prob/high-impact unknowns (generic controls, not bespoke)

Risk Scoring

Score = Probability (1-5) × Impact (1-5)

PriorityCriteriaAction
P0(Prob ≥4 & Impact ≥4) OR data loss/security/moneyMust fix now
P1Score 9-15, user-visible correctnessShould fix
P2Score 4-8 OR low-prob/high-impactSafety net
P3Score ≤3, low impactAccept risk

Output Format

Follow this structure exactly:

1) Feature Summary

  • •What changes / Users / Success criteria / Non-goals

2) Invariants & Trust Boundaries

  • •Invariants (must always be true)
  • •Trust boundaries (user→server, server→DB, server→3rd party)
  • •Data classification (PII? payments? auth?)

3) Happy Path

5-10 step narrative of ideal flow.

4) Edge Case Register

Use template from templates/edge-case-register.md:

IDCategoryScenarioProbImpactScorePriorityMitigationTestsObservability

Rules:

  • •Prefer category-level rules over one-off conditions
  • •Every P0/P1 needs: mitigation + test + observable signal
  • •Reference references/taxonomy.md for comprehensive coverage

5) Safety Nets Checklist

Select applicable controls:

  • • Input validation + normalization (server-side)
  • • Idempotency (retries/double-click/duplicate webhooks)
  • • Timeouts + retries + backoff (external calls)
  • • Circuit breaker / fallback behavior
  • • Concurrency control (locks/uniqueness/transactions)
  • • Consistent error mapping (no leaks, stable codes)
  • • Rate limits + abuse controls
  • • Feature flags + kill switch + safe defaults
  • • Rollback / compensating actions
  • • Monitoring: logs, metrics, traces, alerts

6) Test Plan

Use template from templates/test-matrix.md:

  • •Unit / Integration / E2E / Load / Chaos (as needed)

7) Release & Monitoring Plan

  • •Rollout strategy
  • •Backwards compatibility concerns
  • •Alerts + dashboards
  • •Rollback steps

8) Open Questions

List unknowns blocking confident planning.

Stop Condition

Stop after delivering the plan. Only implement if user explicitly requests.

References

  • •Failure taxonomy: references/taxonomy.md - comprehensive edge case categories
  • •Safety nets guide: references/safety-nets.md - detailed implementation patterns
  • •Templates: templates/ - copy-ready output formats