What is Big Plan?
Big Plan turns your agent’s standard text plan into a rich, inert HTML document specially designed for human review.
Run it in your favorite agent harness as a skill like this:
Give me a /big-plan for implementing the following feature...Give me a $big-plan for implementing the following feature...Why build it?
Section titled “Why build it?”Agent plans are kind of a big deal.
The more confident you are in your agent’s plan, the better your agent will perform at the execution step and the faster your final review will go.
But an agent’s standard wall-of-text plan is painful to read! If plans are so important for the human, they should present in a first-class way. These plans really are a big deal, so we call them “big plans”.
What makes a good agent plan
Section titled “What makes a good agent plan”A great agent plan should be:
-
Easy to scan. You don’t always want to read the full plan. Sometimes it’s better to scan the sections and read the details where they interest you.
-
Easy to read. There are many ways to present an idea. Sometimes a paragraph of text is best, but often a visual arrangement conveys the same information with far less cognitive load.
-
Easy to update. It’s painful to reference a specific paragraph of prose. You should be able to instantly highlight a section and give the agent feedback.
-
Token-efficient. Plans get written, revised, and re-read by agents, and every one of those interactions costs tokens. Authoring a plan, updating it, and pulling in feedback should all be as token-efficient as possible.
-
Versioned. Sometimes the plan updates and you’re not sure what changed, or you accidentally overwrite something. Versioning should allow you to easily see older plans.
-
Local and private. Your plan can contain sensitive information; it shouldn’t be shared with anyone else without a good reason.
Next step
Section titled “Next step”The fastest way to get a feel for Big Plan is to see one. Check out the demo, then render your own plan in under a minute.