Skip to main content

​
How Context Works

When you send a message, the agent assembles context from multiple sources to understand your task. The more relevant context it has, the better the output.
1

Workspace analysis

The agent reads your project’s file structure, dependencies, and language configuration to understand the environment.
2

Active file context

The file currently open in your editor is automatically included. The agent sees the full contents and cursor position.
3

Skill matching

The agent checks available Skills and loads any whose description matches the current task.
4

Explicit context

Anything you attach via @ mentions, file selections, or pasted content is added to the context window.
5

Tool-gathered context

During execution, the agent can search your codebase, read additional files, run shell commands, and query MCP tools to gather more context as needed.

​
Providing Context with @ Mentions

Use @ in the chat input to attach specific context:
MentionWhat it does
@fileAttaches a specific file from your workspace
@folderAttaches all files in a folder
@JiraPulls issue details from a connected Jira ticket
@urlFetches content from a URL (docs, API references)
Be selective with @folder — attaching a large directory can dilute the relevant context. Prefer @file for targeted results.

​
Multi-Repository Context

With Multi-Repository Search, agents can search across indexed repositories beyond the current workspace. This is useful for:
  • Understanding cross-service dependencies
  • Finding shared utility implementations
  • Checking how other services handle similar patterns
Multi-repo context is fetched on demand — the agent searches indexed repos when your prompt references other services or projects.
Multi-Repository Search requires Pro Plus or higher. See Multi-Repository Search for setup instructions.

​
Skills as Context

Skills inject domain knowledge and team conventions into the agent’s context automatically. Unlike @ mentions, skills are loaded based on task matching — you don’t need to attach them manually. Examples of context that skills provide:
  • Coding standards and naming conventions
  • Error handling patterns your team uses
  • Testing requirements and framework preferences
  • API design guidelines
Skills are stored as SKILL.md files in .agents/skills/ and are version-controlled with your repo.

​
Image Context

Attach images directly in chat to provide visual context:
  • Screenshots of UI bugs or desired designs
  • Figma exports for design-to-code workflows
  • Error screenshots from browsers or terminals
  • Architecture diagrams for reference
Image attachments are available across VS Code and JetBrains IDEs.

​
Context from External Tools

Chrome Extension DiscontinuedThe Zencoder Chrome Extension has been discontinued and is no longer available. Use @url mentions to pull in context from browser-based tools, or connect them via MCP servers.

​
MCP Tools

Connected MCP servers let agents query external systems during execution — databases, APIs, Slack, GitHub, and more. The agent decides when to call these tools based on the task.

​
Tips for Better Context

The active editor file is always included. Open the most relevant file before sending your prompt.
Instead of describing a file’s contents, attach it with @file. The agent gets the exact source instead of your paraphrase.
Long, multi-topic chats accumulate stale context. Start a new chat for each distinct task so the agent’s context stays clean.
You don’t need to attach every relevant file upfront. The agent has codebase search and file read tools — it will find what it needs during execution.
If you find yourself repeating the same instructions across chats, create a Skill so the agent loads that context automatically.

Coding Agent

The primary agent that uses all context sources

Skills

Reusable instruction packages loaded by context matching

Multi-Repository Search

Cross-repository context for multi-service architectures

Integrations & MCP

Connect external tools for richer context