How we stopped re-explaining decisions to Claude Code, Codex and Cursor

On Saturday a Codex session completed search-visibility work by writing a guide, submitting the sitemap to Google and Bing, and logging the steps taken and remaining tasks in the project's central knowledge base. On Monday, I launched a separate session of Claude Code (a different tool) and asked it to continue from where Codex left off.

Claude Code had access to the guide that had already been created, the fact that the pull request had been merged after an earlier note indicated it hadn't been merged, and the information that Bing had not yet indexed the guide. Rather than writing a new guide, Claude Code extended the existing one. Previously, when we had to delegate tasks from one session to the next, I had to spend time re-explaining the previous session's decisions, often missing something in the process.

Each tool maintained its own memory of previous sessions, but none of them had knowledge of what the other tools had decided. The key to solving this issue was to establish stable conventions in a rules file like AGENTS.md, and to keep track of all decisions, corrections, evidence, and open questions in a single project brain accessible to all the tools you use.

This approach ensures that information is automatically shared between tools and that the shared project brain serves as a single source of truth.