| name | instruction-generation |
| description | Onboard an AI agent to an unknown codebase by generating a comprehensive instructions file. Use when asked to analyze a codebase, generate copilot instructions, create an onboarding document, or teach an AI about a project. Runs a 6-step prompt chain covering tech stack, file categorization, architecture, domain analysis, style guides, and instruction building. |
Skill: Instruction Generation
This skill guides you through a multi-step analysis chain to generate a comprehensive instructions file (e.g., copilot-instructions.md) by analyzing the structure, patterns, and intent of a codebase. The resulting file helps AI tools operate more effectively within the project by providing clear architectural context, domain understanding, and stylistic guidelines.
Overview
This prompt chain walks through structured steps to extract meaningful insights from a codebase:
- Identifying the technology stack and major frameworks
- Mapping out file purposes and categorizing project structure
- Inferring architecture and design patterns
- Understanding domain concepts and key features
- Generating stylistic and structural guidance for future code contributions
The final output serves as a high-level onboarding and guidance document that aligns AI-generated code with the project's existing conventions and design.
Parameters
Before starting, define these parameters:
- {output-folder} — Path where intermediate analysis files are saved (e.g.,
.results/)
- {final_output_file} — Final combined output file (e.g.,
/.github/copilot-instructions.md)
Recommended Defaults
| Tool | output-folder | final_output_file |
|---|
| Copilot | .results/ | /.github/copilot-instructions.md |
| Windsurf | .windsurf/ | /.windsurf/instructions.md |
| Claude | .results/ | CLAUDE.md |
Execution
Run the following steps in order. Each step reads the output of previous steps.
For each step:
- Read the corresponding sub-prompt file from this skill's folder
- Launch a subagent with the sub-prompt contents, substituting
{output-folder} and {final_output_file} with the resolved parameter values
- Wait for the subagent to complete before moving to the next step (unless steps can run in parallel — see below)
The sub-prompt files are located alongside this SKILL.md file.
Execution Order
Steps 3 and 5 have no dependency on each other and should be run in parallel to save time:
Step 1 (Tech Stack)
↓
Step 2 (Categorize Files)
↓
Step 3 (Architecture) ←──── run in parallel ────→ Step 5 (Style Guides)
↓
Step 4 (Domain Deep Dive)
↓
Step 6 (Build Instructions)
Step 1: Determine Tech Stack
Read ./1-determine-techstack.md and launch a subagent with its contents.
The subagent should analyze the codebase and write its findings to ./{output-folder}/1-techstack.md.
Step 2: Categorize Files
Depends on: Step 1
Read ./2-categorize-files.md and launch a subagent with its contents.
The subagent should read ./{output-folder}/1-techstack.md first, then categorize every file in the codebase and write the result to ./{output-folder}/2-file-categorization.json.
Step 3: Identify Architecture ⇄ Step 5: Style Guide Generation (parallel)
After Step 2 completes, launch Steps 3 and 5 as parallel subagents — they have no dependency on each other.
Step 3: Identify Architecture
Depends on: Steps 1, 2
Read ./3-identify-architecture.md and launch a subagent with its contents.
The subagent should read ./{output-folder}/1-techstack.md and ./{output-folder}/2-file-categorization.json, then identify architectural domains and write the result to ./{output-folder}/3-architectural-domains.json.
Step 5: Style Guide Generation
Depends on: Step 2
Read ./5-styleguide-generation.md and launch a subagent with its contents.
The subagent should read ./{output-folder}/2-file-categorization.json, then for each category write a style guide to ./{output-folder}/5-style-guides/{category}.md.
Step 4: Domain Deep Dive
Depends on: Step 3
Wait for Step 3 to complete. Read ./4-domain-deep-dive.md and launch a subagent with its contents.
The subagent should read ./{output-folder}/3-architectural-domains.json and ./{output-folder}/1-techstack.md, then for each domain write findings to ./{output-folder}/4-domains/{domain}.md.
Step 6: Build Instructions
Depends on: Steps 3, 4, 5 (all must be complete)
Read ./6-build-instructions.md and launch a subagent with its contents.
The subagent should synthesize all previous outputs (including ./{output-folder}/4-domains/{domain}.md files) and generate the final instruction file at {final_output_file}.
Alternative: All-in-One Usage
You can run this entire chain by providing the agent with these parameters and instructing it to execute all 6 steps in sequence:
{output-folder} = .results
{final_output_file} = /.github/copilot-instructions.md
Execute the instruction-generation skill steps 1 through 6 in order.
For each step, read the sub-prompt file and launch a subagent to perform the work.
Stop only when all steps are complete and {final_output_file} is generated.
