How to write a strong agent prompt

A good agent prompt gives Rox enough structure for a useful first draft.

Include these four things:

1. The trigger

Say exactly when the agent should run.

Examples:

• every Friday at 4 PM

• before a timed event

• when a transcript is completed

• when a deal changes stage

• when a webhook is called

2. The context

Tell Rox what information it should use.

Examples:

• account data

• contact data

• meetings

• transcripts

• emails

• deal history

• generated values

• webhook payload data

3. The output

Be specific about what you want at the end.

Examples:

• Slack message

• email draft

• HTML report

• prep brief

• slides

• downloadable file

4. Guardrails

Tell Rox what matters and what to avoid.

Examples:

• do not invent dates or owners

• only run for meetings with external attendees

• only continue if a meaningful signal exists

• keep the output concise and factual

• notify the owner when the agent is complete

A strong agent prompt is not just descriptive. It is directional. It tells Rox the job, relevant inputs, and the desired result.