X Timeline Summarizer

Define the agent's purpose in the instructions. Start with one status

Start with one sentence that explains what the agent does. "You are an agent that does [action]. You will scan my x timeline and return the following, figure it out. (Figure it out helps you at first to see what the agent can figure out on its own and then you can refine the instructions later based on what is figures out).

Clearly define what shoud be included

You must explicitly list what the agent should look for. "You should look for X, Y, and Z." Give a definition of what you're looking for if it's helpful.

Give the agent guardrails

Give it explicit exclusions if there are things you do not want to include in your output.

Give it geographic, demographic, or any specific restraints

This allows you to get more specific.

Define how posts are received.

If you only want original posts, you can instruct the agent to exclude replies and retweets. Or you can explicitly ask for all activity.

Handle time correctly

Always normalize time in UTC. When a relative timeframe is requested (e.g., “last 60 minutes”): 1. Convert the user’s current local time to UTC 2. Subtract the requested duration 3. Generate a start_time in: YYYY-MM-DDTHH:mm:ssZ 4. Use that timestamp in the API call This prevents missing or duplicated posts and makes the agent reliable across all time zones

Specify output requirements (what the agent must return)

Be very explicit about formatting and content. Each returned result must include: The relevant post content A direct link to the X post Only posts that match the defined criteria Do NOT return: Commentary summaries without links Posts that are informational but not alerts/forecasts/damage