Writing Effective Agent Prompts

The Agent node is one of the most powerful tools in your Trigify workflow. It uses AI to evaluate every post against your criteria, filtering out noise and surfacing what matters. This guide shows you how to write prompts that actually work.

What the Agent Node Does

The Agent node takes each post that flows through your workflow and runs it through an AI evaluation. Based on your prompt, the Agent decides what to do with the post: pass it through, filter it out, score it, or even draft a response.

How to Structure a Good Prompt

The best Agent prompts are specific, structured, and clear about expected output. Tell the AI exactly what to evaluate and exactly how to respond.

Template

Evaluate the following post:[Your specific criteria here]Output format: [exact format you want]

Example Prompts

ICP Filtering

Evaluate if this person matches our ICP:- B2B SaaS company- 50-500 employees- VP or Director level in Sales or MarketingBased on the post content and author info, output "yes" or "no".

Sentiment Filtering

Is this post expressing frustration or dissatisfaction with [competitor name]?Look for complaints, negative experiences, or requests for alternatives.Output "yes" or "no".

Relevance Scoring

Rate how relevant this post is to [your topic] on a scale of 1-10.10 = directly about the topic, 1 = completely unrelated.Output only the number.

Comment Drafting

Draft a helpful, non-promotional reply to this post.Be genuine and add value. Do not mention our product.Keep it under 100 words.

Tips for Better Prompts

  • Keep outputs structured - Yes/no, numbers, or short labels. This makes it easy to use IF conditions after the Agent.
  • Be specific about criteria - "Relevant posts" is too vague. "Posts from marketing leaders discussing ABM strategy" is specific.
  • Test with sample posts - Run a few posts through manually before publishing to make sure the Agent behaves as expected.
  • One job per Agent - Do not try to filter, score, AND draft a comment in one prompt. Chain multiple Agent nodes instead.

Common Mistakes

  • Prompts too vague - "Find good leads" gives inconsistent results. Be explicit about what "good" means.
  • No output format specified - If you do not tell the Agent how to respond, you will get paragraphs instead of actionable outputs.
  • Trying to do too much - Complex multi-step prompts lead to unreliable results. Break them into multiple Agent nodes.