Skip to main content
Once your agent is set up, ask for the outcome you want in ordinary language. Name the source app, the time range, and whether the agent may write to another app.

​
Requests that work well

“Check our connected Linear project for issues updated this week. Summarize blockers and link each source. Do not change the issues.”
“Find customer messages about the launch in Gmail. Draft replies for the three that need answers, and show me each recipient and draft before sending.”
“Post the approved weekly update in the project Slack channel, then show me the message link and the Loadout audit result.”

​
What the agent should do

  1. Search Loadout for the needed Action. If the task needs a method or sequence of decisions, it may search a public Skill first.
  2. Read the exact Action schema and check its Integration connection in Vault.
  3. Ask you to complete any missing app connection in Aident. Keep provider passwords and API keys out of chat.
  4. Show the planned recipient, destination, cost, or other material effect before a consequential write when your authorization is needed.
  5. Run the Action and report the result. Use Audit when a receipt or failure diagnosis matters.

​
Check the result

Open Audit to see recent Action calls. An Action can be available in the catalog while your account is disconnected, and an attempted call can fail because the provider rejected the input. Treat a successful response and its observed destination as the evidence that the requested work happened.

Connect apps

Prepare a missing account connection.

Troubleshoot

Diagnose setup and Action failures.