The ASV Labs AI Resource Center offers 163 free, reviewed agent prompts you can give your agent as skill sets. They are prompts, not an installable skills file. Filter them by job, platform, connection and action mode, save a shortlist with no account, then adapt one to a single job and its limits.
What is in the free collection?
The prompt collection is a set of reviewed agent prompts for real jobs. Each record covers a purpose, a category, an action mode, and the connections it needs. Categories include Marketing, Ops, Personal, Productivity, Sales and Success.
You can narrow the list three ways:
- Action mode: draft only, explicit human confirmation, or read only.
- Platform: specific platforms, or platform neutral.
- Connection: the services a prompt depends on, such as Gmail, Slack, Notion, Google Calendar or GitHub.
Saving and reading progress stay in your browser. No account is needed, and you can clear saved items any time.
How do you pick the right one?
Start with one clear job. A narrow first job makes mistakes easier to notice and improvements easier to measure. "Help with my business" is too broad. "Summarize this week's customer questions" is a job both you and the agent can understand.
Then filter by the connection you already use. A listed connection is a requirement, not proof that your agent has access. Before you rely on a prompt, confirm your agent can actually reach that service. The Tool Notes page explains the difference between a built-in connector, an MCP server, browser control, a CLI tool and a data source, plus a safe first test for each.
How do you make a prompt your own?
Treat a prompt as a starting point. The first agent brief walks through four decisions you add on top:
- The job: one result you can name and review.
- Good work: a short checklist another person could use to pass or reject the result.
- Trusted facts: where the facts live, and what to do when sources disagree or are missing.
- Permissions: what the agent may read, what it may prepare, and what it must ask before changing.
For a deeper walk through the parts of a prompt, read anatomy of a reliable agent prompt.
Which action mode should you start with?
Start with read only or draft only. A draft-only agent prepares the email, post or summary and you send it. When a prompt changes something outside your desk, such as sending, deleting or paying, choose explicit human confirmation. The guide on human approval for consequential actions covers where that line belongs.
What should you check before connecting a prompt to your accounts?
Give the agent only the access the job needs. Connecting tools without over-sharing shows how. If the prompt reads web pages, email or files, plan for prompt injection: text inside those sources can try to redirect the agent. The prompt injection and secret handling checklist is a quick pass before you go live, and the Security and Privacy practice lets you try the calls in interactive scenarios.
FAQ
Are the agent prompts free? Yes. All 163 are open to browse, and saving a shortlist needs no account.
Are these skills files I can install? No. Each one is a reviewed prompt. You give it to your agent as instructions, the way you would use a skill set, and adapt it to your own job.
Do I need an account to save prompts? No. Your collection and reading progress stay in this browser.
Will a prompt work with my tools? Check its listed connections. A listed connection is a requirement, not proof your agent has access, so confirm the connection first.
Where should a beginner start? Build your first agent brief: choose one job, describe good work, name trusted facts, and set permissions.