# clue

> A fast and simple coding agent for Claude, in your terminal. clue calls Claude's API directly.

Each page below is plain Markdown. All of them together are at https://clue.shivam.dev/llms-full.txt.

## Getting started

- [Quick start](https://clue.shivam.dev/docs.md): clue is a coding agent for the terminal. It runs on Claude's API.

## Using clue

- [Prompts](https://clue.shivam.dev/docs/prompts.md): Mention files, add images and steer clue while it works.
- [Sessions](https://clue.shivam.dev/docs/sessions.md): Each session is saved under ~/.clue/sessions/ and belongs to the folder it started in.
- [Compaction](https://clue.shivam.dev/docs/compaction.md): Compaction replaces the older part of a conversation with a summary that Claude writes.
- [Scripts](https://clue.shivam.dev/docs/scripts.md): clue ask sends one request without the interactive shell. It takes the prompt as arguments or on stdin.

## Configure

- [Models](https://clue.shivam.dev/docs/models.md): clue starts with claude-opus-5-5 at medium effort.
- [Permissions](https://clue.shivam.dev/docs/permissions.md): The ask, auto and full-access modes, and the rules that decide which actions clue runs.
- [Project instructions](https://clue.shivam.dev/docs/project-instructions.md): Instruction files tell clue how your project works. Good ones are short: the commands to build and test, the conventions to follow, and what to avoid.
- [Settings](https://clue.shivam.dev/docs/settings.md): Settings live in ~/.clue/settings.json. Every key is optional.

## Extend

- [Skills](https://clue.shivam.dev/docs/skills.md): A skill is a folder of instructions that clue reads when a task needs them.
- [MCP servers](https://clue.shivam.dev/docs/mcp-servers.md): MCP servers give clue more tools. Servers start in the background, so they never delay the prompt.

## Reference

- [Slash commands](https://clue.shivam.dev/docs/slash-commands.md): Type / to open the list of commands. tab completes the one you're typing.
- [Keys](https://clue.shivam.dev/docs/keys.md): Every key clue responds to, from sending and steering a message to editing the prompt.
- [Tools](https://clue.shivam.dev/docs/tools.md): The model chooses these tools itself. Permissions decide which calls run.

## Experiments

- [Dream](https://clue.shivam.dev/docs/dream.md): clue dream reads your past sessions in a repository and proposes memories from your corrections. Approved memories load into every new session there.

## Optional

- [Changelog](https://clue.shivam.dev/changelog): What changed in each version of clue.