ContextBot

ContextBot is a GitHub app that automates a core part of compound engineering: turning a team’s code review feedback into updates to the context files, like AGENTS.md and CLAUDE.md, that their AI coding agents follow. Reviewers tend to leave the same comments on agent-written code week after week, while the files that should teach the agent those conventions go stale. Each week, ContextBot reads a team’s pull requests and review comments, works out which conventions matter, and opens a pull request that updates the context files for the team to review.

At ContextBridge, I shared the engineering on ContextBot with my co-founders. As an AI-native team, we went from a strategy document to a release for early users in 28 days, including a production GitHub app and the durable workflows behind it. My work spanned the backend and the AWS infrastructure, with a focus on security. ContextBot’s agents read comments that anyone with access to a repository can write, so I built safeguards against prompt injection. The agents run without any credentials they could leak, and each run can only reach the data for the repository it’s working on.

ContextBridge retired ContextBot in 2026 to focus on its other products.

Key Skills

  • AI Agent Orchestration
  • Prompt Injection Defense
  • Serverless AWS Architecture
  • GitHub App Development
  • Rapid Product Development

For more information on ContextBot, read the launch post.

Get in Touch

Interested in working together? Let's connect.

Contact me