1. Buy and download
Pick the agent for the job you want gone. $19 one-time. Three files arrive instantly: skill.md (the full operating workflow: evidence rules, output contract, test cases, stop points), setup-prompt.txt (guided setup for Google, Microsoft or Zoho) and usage-prompt.txt (the short day-to-day request that runs the job). They are yours to keep forever.
2. Set up on your own tools
Attach the three files to the assistant you already use and run the setup prompt. It asks for the exact form and spreadsheet to use, requests only the access the job needs, then tests the whole workflow on fictional data and shows you a setup report.
3. Approve, then it runs
Review the setup report and approve. From then on the agent does the job and stops at every human gate: it does not send, spend, publish, schedule or decide without the exact approval that action needs. Setup-and-forget agents run a standing cycle on new or due items; give-task agents run when you hand them an exact task.
Where outputs land
Every run saves a Google Doc, Sheet or Form (Microsoft or Zoho if you prefer) in a well-named folder in your own Drive, with evidence and unknowns shown per row. You keep the files, the outputs and the final say.
Set it up on the assistant you already use
Every agent is three plain-text files, so it runs wherever you already work. What changes per assistant is where the files live and whether the setup persists between sessions - here is the honest version for each one.
Instinct
Hand the three files (skill.md, setup-prompt.txt, usage-prompt.txt) to Instinct in chat. It keeps the context across sessions, so the setup survives: run the setup prompt once, approve the fictional-data test, and the agent becomes a standing workflow you trigger with one line. This is the closest fit to how the agents are designed to run.
Full Instinct guide →ChatGPT
Create a Project for the job, upload skill.md as a Project file and paste the setup prompt into the Project instructions. Run the setup once, approve the sample-data test, then start each run inside that Project so the rules stay loaded.
Full ChatGPT guide →Claude
Create a Project for the job and add skill.md as Project knowledge with the setup prompt as instructions. Run the setup conversation, check the sample-data output, then start each run in that Project so the rules stay loaded. Artifacts Claude produces can be saved out to your Drive.
Full Claude guide →Gemini
Create a Gem whose instructions are the skill.md contents plus the setup prompt, then run the setup. Gems keep their instructions between sessions, so the agent's rules persist. For Sheets work, Gemini's Workspace integration makes it the natural home for these agents.
Full Gemini guide →Grok
Attach the three files to a chat or build a Grok Skill/Bot with skill.md as the standing expertise, then run the setup prompt and approve the fictional-data test. Skills keep the rules persistent between asks.
Full Grok guide →DeepSeek
Upload skill.md at the start of a chat, paste the setup prompt, and approve the sample-data test. Keep the three files in a folder you can reach: each new chat starts fresh, so you re-attach skill.md and re-paste the usage prompt per run.
Full DeepSeek guide →Perplexity
Create a Space for the job, upload the three files, and paste the setup prompt into the Space instructions. Run the setup, approve the sample-data test, and start each run in that Space. Save the agent's output where your work lives (Sheets, Docs).
Full Perplexity guide →Microsoft Copilot
With Microsoft 365 Copilot, create a Notebook, add skill.md as a reference and the setup prompt as the custom instructions, then run the setup and approve the sample-data test. With consumer Copilot, upload skill.md at the start of each session instead.
Full Microsoft Copilot guide →Hermes
Give Hermes the three files in its workspace: skill.md becomes the standing skill, then run the setup prompt and approve the fictional-data test. Because Hermes is open-source and local-first, your prompts and outputs stay under your control.
Full Hermes guide →OpenClaw
Drop the three files into your OpenClaw workspace and point it at skill.md, then run the setup prompt over your usual chat channel. OpenClaw keeps its workspace persistent, so the agent's rules survive between asks, and its approval style matches the agents' stop-at-human-gates design.
Full OpenClaw guide →Town
Share the three files with Town and ask it to keep skill.md as the standing rules for the job, then run the setup prompt and approve the fictional-data test. Town's email and calendar focus pairs well with agents whose output is a draft, a schedule or a follow-up queue.
Full Town guide →Muse
Give Muse the three files and ask it to keep skill.md as the standing rules for the job, then run the setup prompt and approve the fictional-data test before pointing it at real work. As with any action-taking agent, keep its approvals on.
Full Muse guide →Questions people ask
What is AI Agent Skills?
AI Agent Skills (aiagentskills.biz) is an independent store of ready-made AI agent workflows. Each agent is a $19 one-time download of three plain-text files - skill.md, setup-prompt.txt and usage-prompt.txt - that you hand to the AI assistant you already use, such as Instinct, ChatGPT, Claude, Gemini, Grok, DeepSeek, Perplexity, Microsoft Copilot, Hermes, OpenClaw, Town or Muse. The agent proves the workflow on sample data first, then runs it on your own Google Sheets, Docs and Forms, stopping at every human decision before it sends, spends, publishes or schedules.
Is AI Agent Skills a subscription?
No. Every agent is $19 one-time. You keep the files forever.
Which AI assistants are supported?
Any assistant that reads plain-text files, including Instinct, ChatGPT, Claude, Gemini, Grok, DeepSeek, Perplexity, Microsoft Copilot, Hermes, OpenClaw, Town and Muse.
Last updated 13 September 2026 · Reviewed by the AI Agent Skills team.