Skip to content

AGENTS.md

Latest commit

 

History

History
58 lines (37 loc) · 2.75 KB

AGENTS.md

File metadata and controls

58 lines (37 loc) · 2.75 KB

Agents and skills workflow repository

This repository contains custom AI coding agents and skills that extend LLM capabilities with specialized knowledge and workflows.

Project structure

  • agents/ - Custom AI coding agents (*.agent.md files)
  • skills/ - Reusable skills for specialized tasks (each in its own directory with SKILL.md)

General guidelines

  • Language: Always use English for file contents and any generated content (code, documentation, commit messages, etc.)
  • Markdown titles: Use sentence case for all markdown headings
  • Nested instructions: Some subdirectories contain their own AGENTS.md files with more specific instructions for that context. These nested files take precedence when working within those directories.

Workflow rules

Pre-task checklist

STOP! Before modifying ANY file in this repository, complete this checklist:

  • Read nested AGENTS.md: Check if the directory you're working in (or its parents) has an AGENTS.md file and read it
  • Identify relevant skills: Search the skills catalog for skills related to your task and read their documentation
  • Understand conventions: Ensure you know all applicable naming, formatting, and structural conventions

Task scope definition: A "task" is ANY user request that modifies files, including:

  • Adding/removing/modifying code or content
  • Creating new files or directories
  • Updating documentation
  • Refactoring existing code

Even "small edits" or "quick additions" are tasks that require the pre-task checklist.

Post-task checklist

After completing ANY task, review and update if needed:

  • AGENTS.md files: Did you introduce new patterns, conventions, or important context? Update relevant AGENTS.md file(s)
  • README.md files: Did user-facing behavior or documentation change? Update relevant README.md file(s)
  • Only update when meaningful: Skip updates for trivial edits that don't change conventions or user experience

Task verification

After completing ANY task, run verification commands to ensure quality:

  • Run markdown linting: Execute pnpm run lint:markdown to verify all markdown files comply with style rules
  • Fix violations: If linting fails, fix violations before marking task as complete
  • Task completion criteria: A task is considered successfully completed ONLY when all verification commands pass with exit code 0

Available verification commands:

# Markdown linting (required for all tasks that modify .md files)
pnpm run lint:markdown

Note: These verification commands are also enforced by git pre-commit hooks (if Husky is configured), but running them manually after completing work provides faster feedback and prevents surprises during commit.