Rules > Prompts — portable rule files

Logic that you'd otherwise paste into prompts lives in versioned rules/*.md files in the vault. Each agent's main config (CLAUDE.md, AGENTS.md, .cursorrules) references the rule files by path rather than embedding the content. A rule looks like: when X tool breaks, follow these recovery steps; or, when writing outreach, never use these banned phrases; or, when publishing, post to saved-messages first, never directly to channels.

The rule layer outlives any specific agent vendor. When the team switched from Claude Code to Codex and back, the prompts that were embedded in tool configs all broke — the rules in markdown files kept working unchanged. Cross-agent portability is a hedge against an ecosystem that's churning every quarter.

1. Pick one piece of logic that currently lives inside a prompt — say, "how to write internal docs" or "how to publish a post."
2. Move it to rules/{name}.md with a one-paragraph header explaining when to apply it.
3. Reference it from AGENTS.md by relative path.
4. Test in two agents (e.g., Claude Code and Codex) to confirm both pick it up.
5. Migrate the rest of your prompts over a week. Audit for rules that should be skills (multi-step) versus rules that stay as guidance text.
6. On any breakage, fix the rule file, not the agent config. Rule changes propagate to every project that references them.

• Resist embedding company-specific data into rules. Rules are logic; data goes elsewhere (graph, MCP). When the two mix, rules become brittle.