AgentSkillsCN

spec

在开始复杂度较高的实现工作之前，先分析需求、提出澄清性问题，并产出清晰明确的需求规格说明书。可通过 /spec 命令快速调用此功能。

SKILL.md
--- frontmatter
name: spec
description: Analyze requirements, ask clarifying questions, and produce a clear specification. Invoke with /spec before starting non-trivial implementation work.
user-invocable: true

Spec — Requirements Specification Skill

Produce a clear, unambiguous requirements spec before implementation begins. This skill answers "what exactly should we build?" — not "how do we build it?"

When to Use

  • •Before non-trivial implementation — any task that touches multiple files or has ambiguous requirements
  • •Unfamiliar codebases — when you need to understand existing patterns before deciding what to build
  • •Ambiguous requirements — when the task description leaves room for interpretation
  • •Cross-cutting changes — features that interact with multiple existing systems

Usage

Provide the task description after the command:

code
/spec Add pagination to the users API
code
/spec Implement rate limiting for the webhook endpoint

Workflow

Phase 1: Requirements Analysis & Clarifying Questions

The agent analyzes the task description, identifies ambiguities, and asks targeted questions grouped by:

  • •Requirements clarity — expected behavior for the primary use case
  • •Edge cases — unusual inputs, failure modes
  • •Scope boundaries — what's in and out of scope
  • •Backward compatibility — potential breakage
  • •Acceptance criteria — definition of done

When a question benefits from concrete context, the agent looks up relevant code to make the question specific rather than generic.

Answer each question. The agent will proceed once ambiguities are resolved.

Phase 2: Spec Output

The agent produces a structured specification:

  • •Context — problem statement and motivation
  • •Requirements — numbered, testable requirements
  • •Constraints — technical constraints (looked up from the codebase as needed)
  • •Edge Cases & Error Handling — identified edge cases with expected behavior
  • •Acceptance Criteria — concrete, verifiable criteria
  • •Open Questions — any remaining uncertainties

Tips

  • •The more context you provide in the initial task description, the fewer questions the agent needs to ask.
  • •If the task is small and unambiguous, the agent will skip questions and produce the spec directly.
  • •Use the spec output as input for implementation planning or hand it to an implementation agent.