/os-done-or-not
A short report and a straight yes or no.
"All done?"
Skill pack for coding agents
Your agent speaks engineer.
Read it in your own words.
Open Steps is a free pack of agent skills for Claude Code, Codex, Cursor and Gemini CLI. It turns jargon into plain language and gives steps you can follow. For the person running the build, not reading the code.
The same session, in plain language
On the left, what the agent said. On the right, what it meant. Illustrative example.
> All done?
● Hotfix deployed: session TTL misconfig in auth middleware caused 401 cascades after key rotation; patched the refresh path, invalidated stale JWTs, redeployed api+web.
Root cause: env drift after the 09-14 rollout. SESSION_TTL resolved to 30s instead of 30m because the staging overlay shadowed the prod values file.
Timeline: first 401 spike 14:02, auth deploy rolled back 14:19, patch merged 14:41, canary at 10% 14:52, full rollout 15:06. p95 back to 180ms from 2.4s.
Verification: 412/414 e2e green, 2 quarantined (already known, tracked). No schema change, no migration. Refresh-token reuse counter flat for 30m post-deploy. Follow-ups filed: values-file lint in CI, alert on SESSION_TTL below 5m. Neither blocking.
> All done?
● Skill: os-done-or-not
Fully done?Yes
Anything needed from you?No
New debt?Two small ones, written down
Safe to close?Yes
Preface, by the author
I am not an engineer. I am a market-led builder: I look for where demand already exists and the infrastructure does not, then build the missing piece. Twenty years of building web and software products, always from the product side, and more than 50 developers at my company today.
Apart from the company, I started building a product on my own, just vibe coding. Partly to stay ahead of where all this is going, partly because I think the moment has arrived for people like me: you can now imagine a serious product and build it yourself, with no engineers or very few of them.
Then I ran straight into a wall. The agent does excellent work, then tells me about it in commit hashes and jargon, and I genuinely cannot tell whether we are done. Not because the work is unclear. Because nobody taught the agent to talk to someone who does not speak engineering.
So I built this pack, and that is the whole idea behind the name: every step of building with an agent, kept open to the person doing the building. Right now it changes what the agent tells you. Where I want to take it next is the work itself, not just the reporting.
Pavlo Kharmanskyi
A glossary of moments
Say the ordinary sentence, or call the skill by name. In Claude Code, type a slash and the name, as shown here. Other tools have their own way to call a skill by name.
How often sentences like these switch the skill on is measured on Claude Code, Codex CLI and Gemini CLI. The numbers are in the repository.
/os-ask-simple
Plain words, and one marked recommendation.
"Ask me simply."
/os-whats-next
One next task, and the reason it comes first.
"What now?"
/os-check-work
Every claim in another session's report, checked.
"Check the other session."
/os-step-by-step
Numbered steps for the part that needs you.
"What do I do?"
/os-say-simple
Any text again in plain words, with nothing left out.
"Say it simply."
/os-what-could-go-wrong
A premortem before anything hard to undo, ending on one verdict.
"What could go wrong?"
/os-big-picture
Where the whole project stands, written down in one file.
"Where are we?"
How to install it
The steps differ by tool. Find yours below.
In Claude Code
Ask Claude Code to do it
The shortest way, and it needs no terminal knowledge: paste this to Claude Code and let it install the pack for you. The skills and both hooks come wired with the plugin, so there is no second step.
Install the Open Steps skill pack for me: clone https://github.com/kharmanskyi/open-steps, add it as a plugin marketplace and install the plugin, then append its docs/routing-block.md to my ~/.claude/CLAUDE.md. When you are done, tell me in plain words what changed.Or install it yourself
In Claude Code the pack installs as a plugin. Clone the repository, then run the second command from the folder that holds open-steps/. The skills and both hooks are wired for you.
git clone https://github.com/kharmanskyi/open-steps.gitclaude plugin marketplace add ./open-steps && claude plugin install open-steps@open-stepsMake the routing a rule
The routing block is a short list of moments and the skill each one needs. In Claude Code, adding it to your own CLAUDE.md turns the skills from a choice into a rule that survives long conversations. Safe to re-run.
grep -q 'os-done-or-not' ~/.claude/CLAUDE.md 2>/dev/null || cat open-steps/docs/routing-block.md >> ~/.claude/CLAUDE.mdIn Codex, Cursor or Gemini CLI
Install the skills
Clone the repository. Then, from the folder that holds open-steps/, copy the skills into ~/.agents/skills/, the folder Codex CLI, Cursor CLI and Gemini CLI read. One copy installs them for all three. It is a copy: after a git pull, run it again from the same folder.
git clone https://github.com/kharmanskyi/open-steps.gitmkdir -p ~/.agents/skills && cp -R open-steps/skills/os-* ~/.agents/skills/Add the routing block
The routing block is a short list of moments and the skill each one needs. It turns the skills from a choice into a rule. Put it in the file your tool reads as standing instructions: ~/.codex/AGENTS.md for Codex, ~/.gemini/GEMINI.md for Gemini CLI, and AGENTS.md in the project root for Cursor. The one-line command for each is in docs/other-agents.md.
Wire the two hooks
Each tool wires the two hooks in its own settings file. The steps for each tool are in docs/other-agents.md.
Questions before you install
The things people ask first, answered plainly. Pick a question to read the answer.
Open Steps is a free, MIT-licensed pack of skills for coding agents. It changes how the agent talks to you. Instead of commit hashes and jargon, you get a short report in plain language, a straight verdict on whether the work is finished, and numbered steps whenever something genuinely needs your hands.
Install Open Steps and ask the ordinary question: “all done?” The pack answers in plain English instead of commit hashes and file paths: what changed, in words a non-engineer reads, followed by a straight verdict and anything that still needs you. On Claude Code with Sonnet 5.5 and Opus 5.5, the skills switched on by themselves in 100% of the pack’s test phrases. On those models you rarely need to type a command.
There are two routes: one for Claude Code, and one for Codex, Cursor and Gemini CLI. Both start by cloning the repository.
In Claude Code, run claude plugin marketplace add ./open-steps followed by claude plugin install open-steps@open-steps from the folder that holds open-steps/. The skills and both hooks are wired for you. Optionally append the routing block to your own CLAUDE.md so the skills fire as a rule rather than as a choice the model makes.
On Codex CLI, Cursor CLI and Gemini CLI, copy the skills from the folder that holds open-steps/ with mkdir -p ~/.agents/skills && cp -R open-steps/skills/os-* ~/.agents/skills/ and run it again from there after a git pull. Each of these tools wires the routing block and the two hooks its own way. The file docs/other-agents.md in the repository has the steps.
Yes. MIT licensed, free to take, use and change, including commercially. No account, no key, no paid tier. The pack itself makes no network calls.
For the person running the build rather than reading the code: non-technical founders, product people and vibe coders who direct an agent every day but do not read diffs. If your agent regularly tells you it is done and you cannot tell whether that is true, this pack was written for you.
An agent saying “done” is a claim, not proof. Open Steps turns it into an explicit verdict with four answers: is it fully done, is anything needed from you, was any new debt created, and is it safe to close. When a claim cannot be verified, it says “not checked” rather than quietly implying success.
Yes, that is the point. The pack describes the change by what it means for the people using your product, not by which files moved. “People can sign in again” instead of a diff. If you do want the technical detail it is still there; it is simply no longer the only thing on offer.
Plain-English skills rewrite prose style: shorter sentences, fewer clichés. Open Steps works one level up: it translates the agent’s account of its own session. It answers whether the work is finished, what still needs you, and what debt was just created. A style rewriter never touches those questions, because they are not questions about writing.
Open Steps is for Claude Code, Codex, Cursor and Gemini CLI. In Claude Code, one plugin install wires the skills and both hooks. Codex CLI, Cursor CLI and Gemini CLI read ~/.agents/skills/, so one copy command puts the skills in all three. The file docs/other-agents.md in the repository has the steps for the routing block and the two hooks on each tool.
How often the skills switch on is measured on Claude Code, Codex CLI and Gemini CLI. Answer quality and the premortem’s fresh agent are measured on Claude Code. The numbers are in the repository.
Pavlo Kharmanskyi created Open Steps. He is a technology entrepreneur and the founder of OnlyMonster.ai, and for twenty years he has built web and software products from the product side. Today he leads a company of more than fifty developers. While building a product of his own with coding agents, he saw them do excellent work, then report it in a language he did not read. He wrote Open Steps to close that gap.