How do you start working with a coding agent in an unfamiliar repo?
Answer out loud or in your head first, then compare.
Answer
First I establish the goal, the scope and how the result will be checked. Then I read the project's instructions, find the right module and the existing tests. The agent can help explore, but I want to know which files its conclusion is based on.
I don't paste the whole repo or ask for a big rewrite straight away. I give a small task with a concrete outcome, e.g. fixing one bug without changing the public API. The error message, the relevant file and the criterion are worth more than a long general conversation. If the agent proposes a bigger change than needed, I stop and check whether it's justified.
Why it matters: The role requires steering implementation, not just handing it to a model.
Possible follow-ups: What do you do when the agent sounds right but points to a function that doesn't exist?
How did it go? Be honest — “Not yet” puts the question on your review list.
Start with
- I-P01 · What is an agent harness? — Harness, MCP and coding agents
- I-P02 · What is MCP? — Harness, MCP and coding agents
- I-P04 · How does a coding agent differ from a chat that suggests code? — Harness, MCP and coding agents
Further reading
- Anthropic: Effective context engineering for AI agents(opens in a new tab)29.09.2025. Choosing context, limits, and references instead of blindly growing the input.