AI Runbook

A comprehensive collection of AI skills and tools for enhancing agent behavior and productivity.

AI Runbook

A dotfiles repo that treats AI agent behavior as infrastructure.

[!IMPORTANT] The contents of this repository, including skills, prompts, rules, documentation, code, and generated outputs, are likely AI-generated or AI-assisted (this note included 😄).

Purpose

  • Share AI that's too good to keep to myself.
  • Update docs automatically so I don't have to.
  • Make it easier to install and reuse skills.
  • Show people how to actually use these things.
  • Treat agent behavior like infrastructure.
  • Keep it versioned, diffable, and deployable.
  • Let naming conventions and file placement do the work of configuration.
  • Treat the workspace as disposable.
  • Use the repo as the only persistence.

Usage

Dotfiles Setup

Fork this repo and select it as your dotfiles repository in Codespaces settings. Add environment variables as Codespace secrets. Every new codespace runs script/bootstrap automatically.

Run the bootstrap script to install tools, configure agents, and wire up skill directories.

bash script/bootstrap

Bootstrap also sets AI-friendly defaults like GIT_PAGER=cat and NPM_CONFIG_LOGLEVEL=warn to suppress pagers and noisy output so agents can parse tool results cleanly.

CORES

An identity-engineering framework built around a meta mnemonic.

It keeps the persona model coherent, consistent, explainable, and easy to extend.

Fill the 5 slots (C/O/R/E/S):

  • Context: Who this is
  • Obstacle: Why this matters
  • Routine: How this works
  • Effect: What this did
  • Scope: Which this retains

The result is a reusable structure for defining and evolving personality modules.

Curiously, this aligns with the “personality spheres” concept from a certain puzzle game.

Environment Variables

Set these to install and configure AI agents automatically.

OPENROUTER_API_KEY: Installs Cline globally and runs auth with the openrouter/auto model.

ANTHROPIC_API_KEY: Installs Claude Code globally. Pay-as-you-go via the Anthropic Console.

CLAUDE_CODE_OAUTH_REFRESH_TOKEN + CLAUDE_CODE_OAUTH_SCOPES Experimental: Installs Claude Code globally for subscription users. See Claude Code Authentication for GitHub Codespaces for setup.

In GitHub Codespaces, GITHUB_REPOSITORY is detected automatically to set the skills manifest path.

Skills

Naming

Generally skills follow the format: {subject…}-{predicate…}.

Install

npx openskills install pbierkortte/ai-runbook

Update

npx openskills update

Invoke

Slash command:

/conventional-commit

Or just ask naturally:

do a conventional commit

List

/skills

Tools

Each tool file contains a specific prompt with instructions.

Adaptive Indexing

Runs on markdown and one bash script. Let docs tell agents what exists so you never hand-write instructions again.

AGENTS.md Files

An AGENTS.md at the repo root activates agent features in supported IDEs. It doubles as a table of contents agents read on first contact. It writes itself.

Surfacing Rules

Rules live in RULES.md files scoped to the directory they apply to. Write them as headings and they surface into the index automatically. RULES.md follows its own rules and serves as the example.

One rule per line keeps each statement atomic so agents can follow it.

Hydrate and Dehydrate Scripts

Scans markdown headings and regenerates every AGENTS.md it finds. Add an empty AGENTS.md to a directory and the docs there get noticed.

Before hydrating, it runs script/stage-protocols, which automatically collects protocols from both the local protocols/ directory (staged as {NAME}_PROTOCOL.md) and any relative ../dotfiles/protocols/ directory (staged as DOT_{NAME}_PROTOCOL.md). Staged protocols are kept out of project commits via .git/info/exclude.

Bootstrap creates ~/.dotfilesrc with an AGENTS.md guard so the hooks only fire in agent-enabled repos. It also sets BASH_ENV so non-interactive shells pick it up. Both stage-protocols and hydrate-agents operate on the current working directory, not the dotfiles installation path.

AGENTS.md files are stored empty in the repo and hydrated at runtime. script/dehydrate-agents resets them back to empty before committing so diffs stay clean and content stays authoritative at runtime.

Run hydration manually with:

bash script/hydrate-agents

Run dehydration manually with:

bash script/dehydrate-agents

Contents

Cores

One who wears many hats.

<!-- CORES START -->

Black

Blue

Evan

Green

Grey

Maya

Orange

Otis

Red

White

Yellow

<!-- CORES END -->

Skills

<!-- SKILLS START --> <!-- SKILLS END -->

Tools

<!-- TOOLS START --> <!-- TOOLS END -->

Resources

Ecosystem

Links