Save the prompt.
Invoke it by name.

Prompts are neurons — same brain, same revision history, same search, same scoping rules. The only thing that's prompt-specific is the slug column you use to invoke them.

With the Cortex MCP server connected, ask the agent to call `resolve_prompts` with the slugs you want and the trailing input. The natural-language pattern is `use cortex software-engineer+security-reviewer review this codebase` — well-trained agents call the tool automatically; you can also be explicit.

Yes. A prompt linked to a team via the team scope is visible to every member. A prompt linked to a project is visible to project members. Resolution priority is project > team > personal > built-in, so a project-scoped `software-engineer` automatically wins inside that project context.

The resolver picks the most specific scope first. A project-scoped prompt beats a team-scoped one, which beats your personal version, which beats a built-in. Inside personal scope, slugs are unique per user — the API rejects collisions with a 409.

You don't edit the built-in directly — built-ins live in the codebase so every account gets the same baseline. To customize, instantiate the template into your brain, then edit your copy. Yours shadows the built-in for any context that can see it.

Combining is plain text concatenation before the model sees the message — there's no extra round-trip and no extra spend on Cortex's side. The model bills as usual based on the final composed input length.