Coding agents need the context of the whole team

Context lives with the individual

A coding agent works with what is in its session: the files it can read, the instructions in files like CLAUDE.md or AGENTS.md, and the conversation with the person who started it. What was decided in that conversation usually stays in the session. Whoever starts the next agent the following day begins with the same files and explains yesterday’s decisions again.

In a team, this adds up. Each person explains the same background to their own agent, and each agent works with a slightly different picture of the task.

How teams share context today

Most teams use what already lives in the repository and in their tools.

• Instruction files in the repository

  CLAUDE.md for Claude Code and AGENTS.md for Codex, Cursor, and other agents hold conventions and background that every agent reads at the start. This works well for anything that belongs to the codebase.

• MCP servers

  The Model Context Protocol connects agents to tools and data sources such as a ticket system, a wiki, or a database. An agent can then look things up on its own, as far as the server’s permissions allow.

• Tickets and pull requests

  Many teams write decisions into an issue or a pull request so that the next person and the next agent can find them.

These approaches share context that someone has written down. The context from the conversation itself, such as which approach the team rejected and why a customer wants something a certain way, rarely makes it into a file.

The agents’ work also stays in the individual sessions. Who asked which agent what, and what came out of it, is visible only to the person at that computer.

Context belongs where the team works

In kiimu, agents work in the same channels as the team. A task starts as a message in the channel, the agent posts its steps and results there, and a person approves the steps that need a decision. The conversation that led to the task sits right next to the work.

Skills and files live in the shared workspace. A skill that one person writes, for example how the team reviews a pull request or sets up a quote, is available to every person and every agent on the team.

The agents stay the ones your team already works with. kiimu comes with its own agents and also runs Claude Code, Codex, Cursor, and OpenCode. For inference, you use your own subscriptions and API keys, or AI credits from kiimu from kiimu Pro on.

A first step

If your team already works with coding agents, pick one recurring task that several people hand to an agent. Write the background for it once, as a skill or in a shared file, and let the whole team work from that version. You will see quickly which context was missing before.