Coding agents: how they work and where they can leak your data

Coding agents can read repos, run commands and change files. That makes them useful, and risky.
In this free live session, Nikalina Ogorodova, an AI/ML engineer at Hyperskill who has trained over 300 engineers, covers:

1. Inside the agent loop: one task traced through tool choice, context retrieval, and a codebase too large for the context window.
2. Where the risks come from: secrets in env or config files can end up in context, and hidden text in a file or web page can steer the agent.
3. Safer use and its limits: sandboxes reduce risk but do not guarantee isolation. You will practice limiting access, using a throwaway VM or container, and reviewing commands.

Participants get an API key to try the examples. The session runs on Zoom through Maven. To get the join link and the recording, register here: https://maven.com/p/20c602/coding-agents-architecture-and-security-intro