What to ask your agent

Last reviewed: 16 Sep 2026

You do not click anything on Naetive to make it work. You ask your agent, in whichever tool you use, and the agent calls Naetive. What agents can do (under Guides) lists everything it can call; this page is what to say. None of these are magic words. Say it your way; the agent works out the rest.

Orient

"Where is this project? Give me the picture." The agent loads the digest: what the project is, its current focus, the decisions and constraints that shape it.
"What should we work on next?" A short list worked out from the real state of the jobs: now, next, blocked.
"What did we decide about payments?" A lookup in the understanding. Ask this before any non-trivial change, so the agent builds on what is known rather than re-deriving it.
"What changed in the repo since Monday?" Recent commits and pushes from the linked repository.

Take work on

"Create a job for the invite flow and write the spec." The agent opens a job and drafts the spec: what, why, edge cases, how we know it is done. It cannot start building on a thin spec, so this is a real step.
"Quick: fix the typo on the pricing page." A job whose name starts with Quick: needs no spec. For small, safe things.
"Pick up the export job where it left off." The agent reads the job's updates, todos and documents and continues. Any tool, any window.
"Make the todos from the spec." The spec's execution section becomes tracked todos the agent ticks as it goes.

While it works

"Post an update before you stop." A short work-log entry: what changed, what is next, with the files and commits attached. Work shows it.
"Leave a note for the other window that the migration is half done." A scratchpad note every window on the project can read.
"Bind to the invite job." Says which job this window is on, so Who's here shows it and updates land in the right place. Claude Code does this by itself.

Record what it learned

"Commit what you learned this session." The durable things, as entries with evidence: decisions, constraints, lessons. This is the step that makes the next agent smarter. Ask for it before you close a chat if the agent has not done it.
"Record that we chose Postgres over Mongo, and why." One entry, with the reasoning in it.
"Cite the file for that claim about the tunnel." Adds evidence to an entry that showed as not yet checked.
"I think the claim about caching is wrong. Dispute it." Marks it disputed and puts it in Needs you for your call.
"Write the architecture overview for the orientation." A narrative section of the understanding, the why the map cannot assemble by itself.

Close

"Finish the invite job and write it up." Closing note, next steps, and the job closes. If the job did substantive work, the agent commits understanding first or states that nothing durable came out of it.
"Mark the tooltips item on the roadmap shipped."

Plan

"What is on the roadmap?" Now, next, later, and what shipped recently.
"Add invite links to the roadmap for next."
"Spec the export item and make it a draft job." Turns a roadmap line into ready-to-start work.

The code, from a chat agent

ChatGPT and Claude on the web reach your files through the tunnel; Claude Code reaches them directly.

"Read the server entry point."
"Find everywhere we call the auth API."
"Run the tests."

Be specific

"Commit what you learned" is better than "save this". "Create a job for the invite flow" is better than "track this". The agent decides how; you decide what. The more specific the what, the better the how.