XSource
If you built a second brain but haven't told Claude/Codex how to use it, it's useless
by Corey Ganim (@coreyganim) · 2026-10-13
✓ Opened via the X API on 2026-10-16
Read the original ↗Part 1 · What the source did
Standing custom instructions that force agents to read your company and personal brains before every task, flag conflicts, and propose exact edits only after you approve.
- Wrote custom instructions that apply on every task for Claude/Codex.
- Agents must read a company brain and a personal brain before acting, flag conflicts with quotes, and ask which version wins.
- Brain edits are proposed as exact diffs and applied only after approval (for GitHub, as a PR).
“If you built a second brain but haven't told Claude/Codex how to use it, it's useless.”— Corey Ganim (@coreyganim), X
Summary written by Agent Cookbook. See the original for the full context.
Part 2 · Our copy-paste version
Adapted by us from what the source described. Fill in the [BRACKETS].
The prompt
These apply to every task in every project. A project's own instructions file (AGENTS.md, CLAUDE.md, a project README) can add rules, but the conflict rule and the response rules below always apply.
🛡
Approval ruleNever edit either brain until you approve the exact change. Sending, buying, posting, and deleting always need a human.
Set it up
- Fill the table with your real brain names and locations.
- Paste into Claude Code custom instructions / CLAUDE.md and Codex AGENTS.md (or project instructions).
- Run one real task and confirm it reads the brains and stops on a deliberate conflict.
Common failures
⚠
What usually goes wrong
- Agent invents brain content when the connector fails; keep the "tell me in one line" rule.
- Edits land without approval if you soften step 3; keep the exact-edit-then-wait rule.
Variations
- One brain only if you are solo.
- Add a third "clients" brain with the same conflict rule.