Agent prompts

Agent prompts are instructions for CLEA to do work. They are different from mention prompts because they do not only measure visibility. They inspect, summarize, recommend, or prepare an output.

The best agent prompts run after there is enough workspace context to use.

An agent can review prompt results and look for changes, weak answers, competitor movement, or missing sources.

This is useful after daily or weekly visibility runs.

An agent can use workspace files as context. This is where workflow files become powerful.

A workflow file can describe how CLEA should review data, what evidence to use, and what output to create.

Agent prompts can produce suggestion cards, team summaries, reports, content briefs, or drafts.

The output should match the review process. If a human needs to approve the work, the prompt should say that.

Scheduled agents make CLEA feel alive. They can run after prompt data arrives and turn fresh evidence into next steps.

For example, a weekly agent can review source changes and send a suggestion card when a competitor starts appearing from a new source.

One-time agents are better for a narrow question. Use them when you want a specific review, a quick report, or one card based on current evidence.

Do not schedule a prompt until the one-time version has produced a useful result.

Files reduce repetition. They let you keep a routine in the workspace and reference it from chat or the calendar.

Goal: review this week's prompt results.
Inputs: latest chatlogs, source analytics, competitor movement.
Steps: find material changes, ignore noise, propose one next action.
Output: one suggestion card with evidence and a clear CTA.

Agent output should not skip human judgment. Review the evidence, the claim, and the next action before acting.

This matters most for website edits, public claims, and publishing flows.