Skip to main content
AI & Agents

AI Workflows

AI tools, Claude Code configuration, sr commits, AGENTS.md standard, skills-as-docs philosophy, and llms.txt. Use when setting up AI tooling, configuring projects for AI, or working with agent skills.

AI Workflows

Specifications

ConventionSpec
SKILL.mdAgent Skills Specification
AGENTS.mdagents.md Standard
llms.txtllms.txt Specification

Skills vs Docs

  • Skills (skills/<name>/SKILL.md) = executable agent instructions — how to do things (conventions, workflows, tool usage)
  • Docs (docs/) = project documentation — what was decided and why (rfcs/, guides/, plans/, runbooks/, architecture/)
  • Both exist in every project — neither replaces the other
  • Every repo should also have: README.md (humans) + AGENTS.md (AI) + llms.txt (LLM discovery)
  • Root-level standard files (README, AGENTS.md, CONTRIBUTING, LICENSE, CODEOWNERS, llms.txt) stay at root — everything else goes in docs/

AI Tools

ToolUsage
Claude Code (opus)Primary AI coding assistant — architecture, refactoring, multi-file edits
OpenAI CodexAlternative AI assistant for code generation
Google Gemini CLIGoogle’s AI for research and code tasks
GitHub CopilotInline code completion in Neovim

Claude Code Configuration

Claude Code is configured per-project via CLAUDE.md files at the repository root. These files define:

  • Project context — tech stack, directory structure, key patterns
  • Coding conventions — naming, formatting, commit style
  • Behavioral rules — what to avoid, how to handle edge cases
  • Memory — persistent notes stored in .claude/ directories

Best Practices

  • Keep CLAUDE.md concise — it’s loaded into every conversation
  • Use separate memory files for detailed topic notes
  • Reference project-specific tools and commands
  • Define verification steps (build, test, lint)

AGENTS.md Standard

Every repo MUST have AGENTS.md at root.

Template Structure

  1. Identity — project name, one-line description
  2. Architecture — high-level component overview
  3. Commands — build, test, lint, run
  4. Code Style — language-specific conventions
  5. Extension Guide — how to add features

Do NOT include static directory trees or file tables — structure goes stale. Instead, instruct agents to use tree and ripgrep/ag to discover project layout on the fly.

Skills Standard

Every repo with a CLI or library should have skills/<name>/SKILL.md:

  • Frontmatter: name, description, argument-hint (for CLI tools)
  • Content: step-by-step guidance for using or extending the project
  • Skills live in skills/<name>/SKILL.md within each repo

AI-Powered Git Workflows (sr)

sr provides AI-powered git commands with multi-backend support (--backend {claude|copilot|gemini}):

CommandPurpose
sr commitGenerate atomic conventional commits from staged/unstaged changes
sr rebaseAI-powered interactive rebase (reword, squash, reorder)
sr reviewCode review of staged/branch changes with severity feedback
sr prGenerate PR title + body from branch commits
sr branchSuggest conventional branch name
sr explainExplain recent commits
sr askFreeform Q&A about the repo
# Stage changes, then generate commit
git add -p
sr commit

# Or commit all changes with context
sr commit -M "refactored auth flow"

# AI-powered rebase of last 5 commits
sr rebase --last 5

Global flags: --backend, --model, --budget (Claude only, default $0.50), --debug.