6-build-instructions

majiayu000's avatarfrom majiayu000

Final step in building instruction context for codebase

⭐0stars🔀0forks📁View on GitHub🕐Updated Jan 5, 2026
[Prompt Templates and Packaging]

When & Why to Use This Skill

This Claude skill automates the creation of a comprehensive markdown instruction file that serves as a project-specific 'meta-instruction' for AI coding assistants. By synthesizing technical stacks, architectural domains, and style guides, it ensures that AI-generated code remains strictly consistent with existing codebase conventions, preventing the introduction of non-compliant patterns or 'invented' best practices.

Use Cases

  • Standardizing AI Code Generation: Generate a master instruction file to ensure any AI assistant (like Claude or Copilot) produces code that perfectly matches the project's unique architectural patterns and naming conventions.
  • Onboarding AI Agents to Complex Codebases: Automatically bootstrap the necessary context and rules for an AI agent when starting work on a new or legacy project, reducing the manual effort of explaining project structure.
  • Enforcing Architectural Constraints: Define and package clear integration rules (e.g., 'all data fetching must use the custom useQuery wrapper') that the AI must follow to maintain system integrity.
  • Automated Feature Scaffolding: Provide AI assistants with a precise 'scaffold guide' so they know exactly which files to create, where to place them, and how to link them when implementing new features.
name6-build-instructions
descriptionFinal step in building instruction context for codebase

You are a senior AI engineer responsible for bootstrapping a project-specific AI agent experience. The goal is to generate a markdown instruction file at:

{final_output_file}

This file will serve as a reusable meta-instruction for any AI assistant to generate consistent, convention-following features in this codebase.

You must synthesize the following source materials:

  • ./{output-folder}/1-techstack.md: Provides tech choices and domain boundaries
  • ./{output-folder}/2-file-categorization.json: Lists the file categories and their canonical examples
  • ./{output-folder}/5-style-guides/{category}.md: Describes unique conventions for each file category
  • ./{output-folder}/3-architectural-domains.json: Defines how domains like ui, routing, data-layer, etc. are implemented, along with constraints and required patterns

Your Output: {final_output_file}

This file must include:

1. Overview Section

Explain the purpose of this file:

  • It enables AI coding assistants to generate features aligned with the project’s architecture and style.
  • It is based only on actual, observed patterns from the codebase — not invented practices.

2. File Category Reference

For each category in 2-file-categorization.json:

  • Explain what it is
  • List 1–2 representative file examples
  • Summarize key conventions based on its corresponding 5-style-guides/{category}.md

3. Feature Scaffold Guide

Define how to plan and implement a new feature. Include:

  • How to determine which categories of files to create
  • Where to place those files
  • How to follow naming and structure conventions
  • Example: what files to create for a new component, a hook, or an API integration

This section should refer to actual conventions in the project (e.g., if Storybook is used, include *.stories.tsx; if styles are colocated .module.css, mention that).

4. Integration Rules

From 3-architectural-domains.json, summarize constraints like:

  • "All canvas logic must use useCanvas"
  • "Components must use shared tokens from the design-system"
  • "API requests go through apiClient.ts pattern"

This prevents LLMs from generating non-compliant or inconsistent files.

5. Example Prompt Usage

Show how a user could prompt Copilot with a request like:

"Create a searchable dropdown that lets users filter by category"

And have it respond with:

  • src/components/SearchableDropdown.tsx
  • src/components/SearchableDropdown.module.css
  • src/hooks/useSearchableDropdown.ts
  • src/components/__tests__/SearchableDropdown.test.tsx
  • etc…

Only use categories and file types present in this project.

⚠️ Requirements

  • Do not include invented best practices
  • Do not list categories or conventions that aren’t supported by the codebase
  • Do not omit any categories or domains defined in the analysis
  • Maximum file lines: {max_lines}

If the file is more than {max_lines} lines, identify the most important information and distill it down to ${max_lines} lines. It's important the file stays within the required limit.

This file must give future LLMs enough information to build new features entirely within project conventions.

To clarify further, if {final_output_file} already exists, overwrite it.