Get clear step-by-step instructions from an agent
Know which action is yours, where to do it and how to check the result
Get clear step-by-step instructions from an agent when a technical explanation still leaves you unsure what to do. The os-step-by-step skill tells the agent to attempt authorised work itself, find a suitable route and hand you only the part that needs your action.
Each step should name one action, its target and a way to recognise success. A decision is presented as a choice rather than disguised as an instruction. The skill supplies no new permissions or tools, and a check depends on what the host can actually inspect.
What it does
- Request instructions without unexplained jargon
- Separate your actions from the agent’s work
- Identify the right app or environment
- Give each step a visible success condition
- Distinguish decisions from execution
- Correct the failed step without repeating everything
Install and use
- Prepare the selected skill folder and licence in a compatible agent, with authorised task context. Any verification tool must already be available and permitted. Full Open Steps hooks and report history have separate setup; they are not required for this instruction file.
- Describe where you are stuck. For example, ask how to select an output file in an application, using the actual button names and explaining how to recognise a successful selection.
- Read the target and expected result before acting. Consequences of a hard-to-reverse change should be clear. Keep passwords and private keys out of chat; their entry destination must be separate and explicit.
- Perform the action and report what you see. Let the agent check within its actual access; if the step failed, ask for that correction rather than the whole procedure again.
FAQ
Source
Made by Pavlo Kharmanskyi. Original skill: os-step-by-step, published on GitHub under the MIT licence.
View on GitHub