Walk Users Through Actions, Not Walls: os-step-by-step

Where this fits

You are mid-task, your agent needs a database password, a browser approval, or a shell command that only a human can run. The user says "I don't know what to do." You could dump a wall of jargon, or you could hand them a single, copy-pasteable line. That is exactly the gap os-step-by-step fills: it converts any agent request into a sequence of one-action-per-step instructions a non-technical person can follow without asking a single follow-up question.

Why agents benefit

• Earns the ask first: the skill forces the agent to try alternative routes, shrink the task to the part only the user can do, and only then asks for action.
• One command per step: each step is a single line that prompts for input—secrets typed hidden, values confirmed in plain words—so the user never has to parse a multi-line script.
• Jargon-free by design: commands are labelled by what they touch (e.g., "your browser", "your terminal") and verified after each step, reducing error and rework.
• No follow-up needed: the skill's trigger phrases cover "step by step", "walk me through", "what do I do" in any language, so it activates exactly when the user is lost.
• Post-step verification: after the user acts, the agent checks the result and moves on—no blind trust, no silent failure.

Practical scenarios

Onboarding a non-technical teammate to run a local script

Your agent needs the user to run a Python setup script. Instead of pasting a block of pip install and export commands, os-step-by-step produces: "Open your terminal (the black window). Paste this line: python setup.py --token — it will ask for your API key. Type it (it won't show), press Enter." The user does it, the agent verifies the output, and you're done.

Handling a one-time approval in a CI/CD flow

A deployment pipeline pauses for a manual approval. The agent knows the user must click a button in a web console. The skill turns that into: "Go to https://deploy.console (opens in your browser). Click the green 'Approve' button. Tell me when it's done." The agent then polls the API to confirm—no guessing.

Collecting a secret without leaking it into chat history

The user needs to provide a database password. os-step-by-step instructs them to run a command that prompts for the value with hidden input, so the secret never appears in the conversation log. The agent then uses the value in a secure environment variable.

Add it to your agent workflow

Install the skill with:

npx skills add kharmanskyi/open-steps --skill os-step-by-step

Then, in your agent's instructions, add a trigger: whenever you need the user to act and they seem unsure, call this skill. A minimal usage example:

When the user says "what do I do?", invoke os-step-by-step. It will generate a single-command step, wait for user confirmation, and verify the result before proceeding.

For a quick reference, see the skill's page on OpenAgentSkill to check its exact triggers and allowed-tools config.

Compare before adopting

Before committing, compare os-step-by-step with alternatives like os-whats-next (which picks the next task) or generic prompting templates. Look at: quality signals (stars, issues, recent commits), maintenance freshness (last update, active maintainer), and how well it fits your existing toolchain—especially if you already use other skills from the open-steps family.

Why it is worth tracking

With 198 stars and an active single-author repository, os-step-by-step shows solid community momentum for a niche but painful problem. It's worth tracking because the pattern—earn the ask, one line per step, verify after—is reusable beyond this skill. Evaluate it when you find yourself writing long, multi-command instructions for users, or when your agent's "please do X" messages routinely get "I don't understand" replies. Tracking it now means you'll be ready to adopt it the moment your workflow hits that friction.