When Cranes comes in
Type @cranes to make sure your AI uses your thinking; it also brings it in on its own when it applies. @cranes on its own checks your work against your thinking, and Cranes stays out of general questions.
Type @cranes to make sure your AI uses your thinking. A connected AI also checks with Cranes before work where something you’ve worked out could matter, and leaves it out when it doesn’t.
When your thinking applies
Before a decision, a piece of writing, a plan, a review, or a judgment call, your AI asks Cranes how your thinking bears on it, and answers with that in mind. It keeps doing that on follow-ups in the same conversation, even short ones like “no, shorter”.
When nothing you’ve worked out applies, Cranes sends nothing and your AI carries on as usual. See How your AI uses Cranes.
@cranes on its own
Type @cranes with nothing else, or with a word like “check” or “thoughts?”, partway through some work, and your AI checks what you’re doing against your thinking:
- what applies to this work at this point,
- your skills that fit it, in full,
- one thing where the work already departs from your thinking or a skill.
If nothing applies, Cranes sends nothing back. A check doesn’t keep anything; if you corrected your AI earlier in the conversation, it may offer to keep those words, and keeps them only if you say yes.
@cranes followed by words still works as before, and naming a group or workspace after it, like @cranes design group, still says whose thinking to use. See Use Cranes from your AI.
General questions
Cranes stays out of questions that don’t depend on your thinking: facts about the world, general help, small talk. Your AI doesn’t call it for those, and nothing from those conversations is offered for keeping. @cranes on its own still runs the check, and gets nothing back when nothing applies.
Your skills
Skills are the instructions you give your AIs, kept in Cranes as skills: a CLAUDE.md, AGENTS.md, a SKILL.md, a project’s instructions, or prompts you wrote. Cranes brings in the ones that fit the work at hand. When your AI describes the work (what kind it is, how far along, and what comes next), the skills closest to it are always considered, so a skill reaches work its words don’t match: “ship it” can still bring your skill on commit messages.
They come in whole, and your AI follows them where they fit unless you say otherwise now. When the work already departs from one of your skills, Cranes points that out. Skills that don’t fit the work are left out.
To bring skills in, see Your first setup. In Claude Code, the Cranes plugin also copies them where Claude Code loads skills. See Claude Code.
Apps decide when to call
Cranes asks every connected app to check with it before substantial work, but each app decides when to call a tool. If yours doesn’t call often enough, use @cranes, or put a line in its instructions:
- In Cranes, choose Connect your AI in the account menu. It opens Take it with you.
- Under Have your AI use Cranes on its own, choose Whose thinking it should use, and copy the line.
- Paste it into your AI’s instructions: a Claude or ChatGPT project, custom instructions, or a file like
AGENTS.mdorCLAUDE.mdin a repository.
The line names the thinking you chose, so a whole team can share it in a repository and each teammate’s AI uses the team’s thinking there, within what each person let their own AI use.
In Claude Code, the Cranes plugin checks on every message for you. See Claude Code.