AgentSkillsCN

first-principles

在实施前，先将问题拆解至最根本的层面。适用于处理复杂或模糊的问题。

SKILL.md
--- frontmatter
name: first-principles
description: "Break down a problem to its fundamentals before implementing. Use for complex or ambiguous problems."
argument-hint: "<problem description>"
user-invocable: true
context: normal
model: opus
allowed-tools:
  - Read
  - Glob
  - Grep
  - Bash
  - mcp__sequential-thinking__sequentialthinking

First Principles Analysis

Structured problem decomposition before jumping into implementation. Forces clarity of thought.

Process

Walk through these steps with the user, pausing for input at each stage:

1. Define the Problem

  • •What exactly are we trying to solve?
  • •What does success look like?
  • •What triggered this need?

2. Identify Constraints

  • •What are the hard technical constraints? (language, framework, infra, budget)
  • •What are the soft constraints? (timeline, team skills, existing patterns)
  • •What must NOT change?

3. Decompose

  • •Break the problem into independent subproblems
  • •Identify which subproblems are already solved (existing code, libraries)
  • •Identify which subproblems are novel and need design

4. Explore Approaches

For each novel subproblem, identify 2-3 approaches:

  • •What's the simplest approach?
  • •What's the most robust approach?
  • •What would an expert do?

5. Evaluate Trade-offs

For each approach:

  • •Complexity (implementation + maintenance)
  • •Risk (what could go wrong)
  • •Reversibility (how hard to change later)

6. Decide

  • •Select an approach with clear reasoning
  • •Document what we're explicitly NOT doing and why
  • •Define the first concrete step

Rules

  • •Use sequential-thinking MCP for structured reasoning when the problem has many dimensions
  • •Read relevant code before proposing approaches — ground recommendations in reality
  • •Challenge assumptions — ask "why" at least once for each stated constraint
  • •Prefer the simplest approach that meets requirements
  • •If the user's framing of the problem seems wrong, say so
  • •Output a clear action plan at the end, not just analysis