> ## Documentation Index
> Fetch the complete documentation index at: https://docs.aident.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Build professional workflows

> Design repeatable agent workflows with Loadout discovery, schema checks, Vault readiness, preflight, execution, and audit.

The Loadout CLI is the primary interface for a shell-capable agent. It gives a workflow a repeatable path through discovery, account access, validation, execution, and evidence. Your own code or scheduler can decide when to run that path.

## The workflow contract

<Steps>
  <Step title="Describe the operation">
    Name the source, task, and constraints. Search **Actions** for a concrete external operation. Search **Skills** when the missing piece is the procedure or decision sequence for a broader outcome.
  </Step>

  <Step title="Resolve exact identities">
    Copy the canonical Action name from search. Read `capabilities get` for the current input and output schemas. Do not derive names from a provider label or reuse a dotted legacy alias.
  </Step>

  <Step title="Check access and cost">
    Use Vault status to verify the required Integration is ready. Use `capabilities preflight` to validate the exact input and estimate any Aident-managed cost without executing.
  </Step>

  <Step title="Execute with the intended account">
    For multiple connected accounts, select the alias confirmed for this operation. Handle credit approval or Action acknowledgement only when the response requests it and the user approves the exact action.
  </Step>

  <Step title="Keep evidence">
    Save the response or a safe summary and inspect Audit. If a response is interrupted, check Audit before retrying. A write may have succeeded even when the caller lost its response.
  </Step>
</Steps>

## Action and Skill discovery

```bash theme={null}
aident capabilities search --query "search customer email" --types '["action"]' --json
aident capabilities get --name '<exact-action-name-from-search>' --json
```

For a larger task whose method is unclear:

```bash theme={null}
aident skills search --query "triage customer requests" --json
aident skills read --name '<skill-id>' --artifactVersionId '<version-id>' --json
```

Search results for Skills are snippets. Read one exact revision, treat its instructions as guidance, and resolve any referenced Action separately. An unknown provider by itself is still an Action search problem.

## Scope a production automation

* Give each step a clear input, expected output, and observed success condition.
* Keep read steps and write steps separate. Require the authorization appropriate to the write's recipient, destination, and effect.
* Persist a request ID or external object ID from a successful write so retries can check whether it already happened.
* Store only the minimum data needed for the job. Keep provider credentials in Vault and avoid logging private Action payloads.
* Use [Audit](/loadout/vault-and-audit) to reconcile failures and measure actual calls; use your application's own monitoring for its orchestration logic.

Continue with the [CLI command reference](/loadout/cli-reference), [public endpoint schemas](/loadout/api-reference), and [worked example](/loadout/practical-workflow).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.