Why Scheduled Prompts Are Different
In a normal conversation, you can clarify instructions, answer the agent's questions, and steer the output in real time. Scheduled runs don't have that luxury — the agent receives your prompt and must complete the task without any human interaction.
This means your prompt needs to be completely self-contained. If the agent would normally ask you a follow-up question, it will either guess (which may produce wrong results) or fail to complete the task.
The Self-Contained Rule
Before saving a schedule, ask yourself: Could someone who knows nothing about this task complete it using only this prompt? If the answer is no, add the missing context.
Include:
What to do — the specific task or deliverable
Where to find data — which datastores, time ranges, or sources to use
How to format the output — structure, length, tone, and any templates
What to skip — explicitly exclude anything that would require clarification
Good vs. Bad Examples
Bad: vague, needs clarification
Write a report about our clients.
The agent doesn't know which clients, what kind of report, what time period, or how to format it. It would need to ask several follow-up questions.
Good: specific and self-contained
Generate a weekly pipeline report for the RM team. Pull data from the last 7 days. Include: total leads, conversion rate by stage, top 5 deals by value, and any deals that have been stalled for more than 14 days. Format as a summary with bullet points, no longer than 500 words.
Bad: assumes prior conversation context
Update the spreadsheet we discussed last week.
Each scheduled run starts a fresh conversation — the agent has no memory of previous chats.
Good: provides all context
Check the SEO Tracker datastore for any pages with a ranking drop of more than 5 positions in the last 7 days. List each page with its current rank, previous rank, and the target keyword. If no pages dropped, respond with "No significant ranking changes this week."
Combine Agent and Project Instructions with the Prompt
Remember that the agent already has its own instructions (the system prompt you configured when creating the agent). If the schedule is attached to a project or client, the project's instructions and context also apply to every run. Your schedule prompt works together with both.
Use the agent instructions for general behavior — tone, formatting preferences, which tools or datastores to use by default.
Use the project instructions for client or project context — who the client is, their goals, and any project-specific rules.
Use the schedule prompt for the specific task — what to do on this particular run.
This way, you can reuse the same agent across multiple schedules and projects with different prompts, while keeping the prompts short and focused.
Test Before You Go Live
Always use the Run Test feature to validate your prompt before relying on the schedule. A test run executes immediately with the exact same setup as a real run, so you can verify the output matches your expectations. See Testing and Monitoring Your Schedules for details.