Skip to main content
Context is the information an agent relies on to act—system instructions, tool definitions, reference material, etc. Context engineering is the practice of building and optimizing that context to improve agent performance and capabilities. This guide walks through the core concepts of context engineering in LangSmith’s context hub, including skills, agents, versioning, and sharing.

​
Skills

A skill packages a single, reusable capability that an agent can invoke. Skills are focused on doing one thing well, and they’re designed to be used by agents. Skill repos contain: Always:
  • SKILL.md file in its root directory (the core content of the skill)
  • Description of what the skill does / when to use it
Optional:
  • References (Technical documentation, guides, etc.)
  • Assets (Schema, templates, etc.)
Examples: email formatting, code review, web search

​
Agents

An agent packages an entire agent: its high-level instructions, the skills and subagents it can use, and the tools it has access to. Agents are designed to complete tasks end to end. Agent repos contain: Always
  • AGENTS.md — the agent’s system prompt and operating instructions
Optional
  • tools.json — set of tools the agent can call
  • Subagents — smaller agents this agent can delegate to
  • Skills - set of skills the agent can use
Examples: email assistant, coding copilot, customer support agent.

​
When to use Skills vs Agents

  • Use skills for smaller tasks, reusable context, specifications
  • Use Agents for longer tasks, tool calling, decision making
A good rule of thumb: if you find yourself copying the same block of context into several agents, pull it out into a skill repo and reference it from each agent.

​
Hub backend vs store backend

Context in LangSmith can be managed by two different backends: the Context Hub and a store backend. They serve different purposes, and most agents use both. The Context Hub is built for versioning, sharing, and continuous improvement. It is the long term context database that your agent can use. Think of it as a Git-like home for your agent’s context. A store backend is built for runtime state. It holds the information an agent accumulates while running: memories, conversation history, user preferences, learned facts, and other data that evolves per session or per user.

​
Versioning

Every change to a repo in the Context Hub creates a new commit. Commits are immutable, browsable, and comparable, so you can:
  • See exactly what changed between two versions of an agent.
  • Revert to any prior commit if a change regresses behavior.
  • Tag important commits (for example, the commit you shipped on a specific date) for easy reference.
  • Promote a commit to an environment like Staging or Production so downstream agents pull a known-good version rather than the latest edit.

​
Sharing and permissions

The Context Hub is designed for teams. Every repo lives in a workspace, and its visibility and permissions are managed at the repo level:
  • Private repos are visible only inside the workspace.
  • Public repos can be discovered and pulled by anyone.
  • Per-repo permissions control who can view, commit, tag, and promote within a workspace.
This makes the Hub a natural place to collaborate on agents and skills, and improve them over time.
Edit this page on GitHub or file an issue.
Connect these docs to Claude, VSCode, and more via MCP for real-time answers.