[MD.SAVE]

AI & AGENT MARKDOWN GUIDE

Clear context leads to better output.

Structure project documentation, agent instructions, context, and reusable prompts so people and compatible AI tools can understand the work faster.

Choose the right file

README.md — the project entrance

A README explains what the project is, why it exists, and how to run it. Lead with the outcome for the user, then keep executable commands and expected results close together.

  • A one-line overview and the problem being solved
  • Setup commands and prerequisites
  • A representative usage example and configuration
  • Project structure, contribution rules, and license
# Project name

One sentence that explains the project.

## Quick start
## Usage
## Configuration
## Contributing

AGENTS.md — working rules and boundaries

AGENTS.md can collect the commands, code rules, and definition of done that compatible coding agents should follow in a repository. Prefer real paths and runnable checks over abstract advice.

  • The repository purpose and important directories
  • Setup, test, lint, and build commands
  • Generated files and secrets that must not be changed
  • A definition of done covering accessibility and security
## Required commands

```bash
npm test
npm run lint
```

## Working rules

- Keep changes focused.
- Never expose secrets.
- Update tests with behavior.

CLAUDE.md — Claude Code project instructions

Claude Code loads CLAUDE.md as persistent project guidance. Keep it concise and reserve it for build commands, architecture, and coding rules that matter in most sessions.

  • The project purpose and important directories
  • Required development and verification commands
  • Rules Claude Code repeatedly needs
  • Minimal duplication when AGENTS.md also exists

GEMINI.md — Gemini CLI context

Gemini CLI discovers GEMINI.md files to build project and directory-specific context. Put shared guidance at the root and keep local rules close to the area where they apply.

  • Technology and repository orientation
  • Development commands and implementation principles
  • Context specific to a nested part of the project
  • Definition of done and verification steps

.github/copilot-instructions.md — Copilot repository guidance

GitHub recommends short, self-contained statements for repository-wide Copilot instructions. After download, place the file at .github/copilot-instructions.md in your repository.

  • Language, framework, and package manager
  • Code style and testing conventions
  • Sensitive data and change boundaries
  • Only guidance that applies broadly across requests

AI_CONTEXT.md — general context attached manually

AI_CONTEXT.md is not an auto-discovered standard filename for a specific development tool. Use it as a general document that you attach manually when a conversation needs product goals, current state, and settled decisions.

  • Define the product and audience in one sentence
  • Record goals and non-goals together
  • Date the current state and known issues
  • Preserve why a decision was made

TASK_PROMPT.md — a reusable task loaded manually

TASK_PROMPT.md is also not an auto-discovered standard filename. Use it to store a role, goal, input, constraints, output format, and final checks that you load directly for repeated work.

  • Write one outcome-focused goal that starts with a verb
  • Separate source material from instructions
  • State what must not be guessed
  • Provide a compact example of the desired output

Keep sensitive information out

Do not place API keys, passwords, real customer data, or private internal URLs in a template. MD.SAVE does not transmit your source text, but you should review the storage, training, and retention policy of any other service before sharing the finished file.