---
title: "Plan a feature with your library"
description: "Use when you want a plan that draws on your books. Covers how to ask your agent explicitly to search your Trove library while it plans, and what changes in the plan."
url: https://docs.heytrove.ai/usage/plan-a-feature
updated: 2026-10-06
---

# Plan a feature with your library

> Load this page when: you need to plan work with your books

## For agents

- Run `trove config get skill.planMode`. If it is not `always`, run `trove config set skill.planMode always` and start a new session.
- In plan mode, explore the code first. Then start the librarian once, before you write the plan.
- Start one item reader per angle on the librarian's reading list. Do not add a limit of your own.
- When two books disagree, show both with citations, recommend one, and ask the user.

You get a plan that uses what your [books](https://docs.heytrove.ai/concepts/library-and-books.md) say, with page [citations](https://docs.heytrove.ai/concepts/citations.md), instead of a plan from the agent's memory.

## Before you start

- Trove is installed (see [Install](https://docs.heytrove.ai/install.md)).
- Your library has books on the topic. If it is new, see [Quickstart](https://docs.heytrove.ai/quickstart.md).
- You plan to ask your agent explicitly to search your Trove library. Plan mode (setting `skill.planMode`, default `always`) can start a check on its own, but do not rely on it. See [Plan mode and Auto mode](https://docs.heytrove.ai/concepts/plan-and-auto-mode.md).

## Steps

1. **Ask explicitly.** Say what you want to build. Then tell the agent to search your Trove library during planning, for example: "Search my Trove library for our API rules before you write the plan." Do not rely on plan mode to start this for you.
2. **Let the agent explore the code.** In plan mode, the agent reads the repository before it plans.
3. **Let the librarian choose.** The [librarian](https://docs.heytrove.ai/subagents/librarian.md) picks the books and the depth. It splits the task into angles and puts books from your project [shelf](https://docs.heytrove.ai/concepts/shelves.md) first. A narrow task gets few books. A broad task gets more.
4. **Let the readers read.** The librarian starts one [item reader](https://docs.heytrove.ai/subagents/item-reader.md) for each angle. Each reader returns an answer with page citations. You do not start readers yourself.
5. **Read the plan.** Each decision that comes from a book has a citation. If two books disagree, the plan shows both, recommends one, and asks you to choose.

## Prompt to paste

```text
Plan this feature: <what you want to build>. Search my Trove library for the rules that apply before you write the plan, and cite the pages you use.
```

## What you will see

When you ask, the agent starts the librarian after it explores the code and before it writes the plan. The librarian returns a reading list. Each angle on the list has its books, the pages to read, and one sentence on why each book fits. If a [marketplace book](https://docs.heytrove.ai/concepts/marketplace.md) fits and your library has no good book, you see an Offer box. Nothing is added until you answer yes.

Then the item readers run. The plan that follows has citations in this form:

```text
Use cursor pagination for the list endpoint (API Design Guide, p. <page>).
```

If the library has no book on a part of the task, the plan says so. It does not invent a rule.

## Why ask explicitly

Trove also runs automatic checks (see [Plan mode and Auto mode](https://docs.heytrove.ai/concepts/plan-and-auto-mode.md)). We are actively working on making the librarian start automatically and reliably. Today, do not rely on it. When the answer matters, ask your agent explicitly.

## Next

- [Plan mode and Auto mode](https://docs.heytrove.ai/concepts/plan-and-auto-mode.md)
- [Librarian](https://docs.heytrove.ai/subagents/librarian.md)
- [Research a topic](https://docs.heytrove.ai/usage/research-a-topic.md)