ai

AI Navigation

Maps from topic to source file: short what/where docs that point an agent at the right code instead of restating it.

ai-devkit is a documentation-first toolkit shipped as a Claude Code plugin and through standard install paths (skills.sh, vendored .agents/skills/). It bundles a set of cmk:* skills that scaffold and maintain a /docs tree (requirements, design, decisions, rules, knowledge, and the rest of the canonical taxonomy), plus a cmk:codebase-docs skill that builds this very docs/ai/ tree, plus a repo-setup facet family that scaffolds local dev, layout, toolchain, agent instructions, infra, CI/CD, MCP config, skill vendoring, and upstream sync. It also bundles a tracker-neutral delivery family that runs tracked work end to end (context intake through spec, implementation, review, and shipping) and a knowledge family of standalone domain reference packs consulted whenever their subject comes up. The repo itself is the canonical source of those skills — there is no application runtime to document, only skill content, plugin manifests, and the docs scaffold.

For the user-facing pitch and install instructions see the root README.md.

Areas

• skills/ — the cmk:* skill packages that ship with the plugin.
• docs/ — the /docs taxonomy this repo defines, scaffolds, and dogfoods.
• integrations/ — how the skills are exposed to Claude Code, other agents, and CI.

Conventions

• Mirror the shape of the codebase: one folder per area, each with its own README.md index.
• Name paths and packages; point at code rather than duplicating it. No ticket IDs, no delivery status.
• Update the affected map in the same change that moves or renames the code it describes.

When to read

First, on any task — to find which source files a topic lives in.