A repeatable workflow is a written task pattern. It defines the source account, inputs, permitted operations and completion evidence so that another authorised team member can understand what should happen.
Define the outcome#
Use a concrete result such as a weekly draft summary of approved records. Include the date range, timezone, required fields and destination. Avoid instructions such as do everything with my tools, which leave the scope unclear.
Separate the stages#
- Read the approved source using a narrow query.
- Check the account context and completeness of the returned data.
- Prepare the proposed output.
- Review any business decision or live change required by the task.
- Execute only authorised actions with available tools.
- Verify the resulting source-system state and record the evidence.
Example workflow prompt#
Prepare the weekly report for the agreed workspace and date range.
Use only the approved read tools. Identify incomplete or missing
data. Return a draft with source record references. Do not send
the report or modify source records.Keep automation expectations clear#
A tool connection does not create a background schedule by itself. A prompt used in a conversation does not continue running after the session unless a separately configured automation system supports it.
If you need scheduled work, agree the scheduler, account, trigger, retry rules, monitoring and authorised outputs with DigLab. Do not assume the AI application remains connected indefinitely.
Maintain the workflow#
Review it when an account changes, a service updates its API or a team member leaves. Run the agreed read-only check after a connection change. Keep a record of what was verified and when, rather than assuming an old test proves today's account state.