---
title: "Instructions"
description: "Apply your team’s coding conventions and project guidance across agent runs."
canonical_url: "https://hoplite.sh/docs/agent/instructions"
markdown_url: "https://hoplite.sh/docs/agent/instructions.md"
---

# Instructions
URL: /docs/agent/instructions
LLM index: /llms.txt
Description: Apply your team’s coding conventions and project guidance across agent runs.
Related: /docs/agent/skills, /docs/agent, /docs/workspace/project-settings

# Instructions

Use organization instructions for guidance shared across projects, and project instructions for conventions specific to one project. Instructions are included in context automatically; use [skills](/docs/agent/skills) for workflows the agent loads only when needed.

## Organization instructions

Organization owners and admins set shared instructions under **Settings → Workspace → Agents**, in **Workspace instructions**, and can also set or clear them through the API. All organization members can read them. These instructions apply to every run in the organization, including automations and Linear-delegated runs. Clearing them leaves project and repository instructions unchanged.

Hoplite includes organization instructions before project instructions in the assembled context. Keep organization-wide house rules here and project-specific details at the project layer. Instructions do not grant additional tool permissions or override platform policy.

## Project instructions

Set project instructions under **Settings → Project → Agents**. Hoplite includes them alongside the selected model and the context assembled for the thread.

## Repository instructions

Hoplite loads root `AGENTS.md` before the first model call, with `CLAUDE.md` as a fallback. Nested instruction files are read as the agent works in their directories. Commit these files with the repository so the instructions follow the checked-out revision.

Automatically injected ancestor instruction files are limited to 32 KiB each. If a file exceeds that limit, the agent sees a truncation notice and can read it explicitly for the remaining guidance. Repository instructions are reference material, not permission to bypass tool or approval boundaries.

Committed Cursor rules in `.cursor/rules/*.mdc` (including nested directories) and `.cursorrules` are also available to cloud agents with their directory scope and frontmatter preserved. Keep project rules in the repository; [Cursor onboarding](/docs/import) imports local personal rules separately.

## Scope

Organization instructions apply across the organization's projects; project instructions apply across the project's threads. A message sent in one thread applies to that thread; it does not update project instructions. Personal output preferences and memories have their own [per-user scope](/docs/agent#personal-memories).

Repository skills are discovered from the checked-out repository and read on demand. Hosted organization skills are available to every run; personal skills come from the initiating user's library and are excluded from automations and Linear-delegated runs. See [Skills](/docs/agent/skills) for precedence and import limits.

## Execution settings

Changing instructions does not change the commands Hoplite runs during workspace setup. Configure executable setup and preview commands in [project scripts](/docs/sandboxes/scripts). The agent can inspect their effective values and update project overrides directly.
