返回 Skill 列表
extension
分类: 开发与工程无需 API Key

acceptance-criteria-design

为已批准的需求设计可执行的验收标准,将目标/规格转化为二元通过/失败检查,并具有可观察的结果。在需要可测试的标准来进行实施交接、质量保证验证或发布决策时使用,以确保`REQ-*`/`NFR-*`基线的稳定性;不应用于需求发现、优先级排序或冲刺拆分。

person作者: jakexiaohubgithub

Acceptance Criteria Design

Overview

Use this skill to translate approved requirements into acceptance criteria that engineering and QA can execute without interpretation drift.

Scope Boundaries

  • Use this skill when the task matches the trigger condition described in description.
  • Do not use this skill when the primary task falls outside this skill's domain.

Inputs To Gather

  • Approved requirements (REQ-*) and relevant non-functional requirements (NFR-*).
  • Business intent, out-of-scope boundaries, and policy constraints.
  • User-visible behaviors, system side effects, and error paths.
  • Test environment capabilities and observability limits.

Deliverables

  • Acceptance criteria set with binary pass/fail expectations.
  • Scenario matrix covering happy path, boundary, negative, and failure cases.
  • Traceability map: requirement -> criterion -> verification method.
  • Open ambiguity list that blocks objective verification.

Quality Standard

  • Each criterion is independently testable and yields one clear result.
  • Each criterion states preconditions, action, expected observable outcome, and evidence source.
  • Boundary values, invalid inputs, authorization errors, and integration failure behavior are covered when applicable.
  • Criteria describe required behavior, not implementation detail.
  • Overlap and contradiction across criteria are removed.

Workflow

  1. Normalize requirement intent and explicit assumptions.
  2. Split requirements into observable behavior units.
  3. Draft pass/fail criteria for normal, boundary, negative, and failure scenarios.
  4. Define evidence sources (API response, UI state, logs, metrics, events) for each criterion.
  5. Run ambiguity and overlap review; publish unresolved blockers.

Failure Conditions

  • Stop when requirement intent is unstable or contradictory.
  • Stop when a criterion cannot be verified with available test/observability capabilities.
  • Escalate when missing instrumentation prevents objective pass/fail judgment.