Lazy Skills — Development skill for Claude Code

lazy-skills

A Claude Code skill that routes multi-step or multi-tool tasks to the right combination of connected skills — without loading every skill's full instructions upfront. Includes an optional prompt-keyword hint hook and a one-command installer for a known-good skill set.

Why

A large skill library only stays cheap if the model doesn't pay a token tax just for it existing. Every skill's frontmatter is already in context all the time (cheap, unavoidable). The expensive part is a skill's *body*, which only loads once it triggers. The risk with a big library isn't the skill count — it's tasks that quietly need two or three skills working together, where only loading the first one leaves you improvising the rest.

`lazy-skills` handles that narrow job: when a task looks like it needs more than one connected skill, it figures out which ones *before* starting, loads only those, and gets out of the way entirely for single-skill tasks.

The actual selling point: you don't need to type `/lazy-skills`

This is a skill, not a command you have to remember. Claude Code auto-triggers it from its description whenever a task's shape matches — multiple chained actions, or uncertainty about whether companion skills exist. Just describe what you want normally:

"Query my orders table, format the SQL, and validate it against my schema before running it in prod."

That's it. No slash command required. The `/lazy-skills` wrapper (see below) exists purely as a convenience for when you want to invoke it explicitly — not because you need it to work.

How it works

1. Decide if the task needs the graph at all. Most tasks don't — single-skill work skips everything below with zero overhead.
2. Consult the dependency graph (references/skill_map.json) — either by reading it directly or resolving it programmatically with scripts/check_deps.py .
3. Load connected skills just-in-time, not upfront. Only read a connected skill's SKILL.md right befo