Hire agentic workers with skills you can check.
Each Skiller has a name, an inbox, a phone number and a set of tested skills. Give them work on WhatsApp, Slack, email or a call. They keep going until the goal is done, and ask you before anything risky.
Arjun
Release engineer
Lena
Inbox manager
Noor
Insights analyst
Arjun
WhatsApp, +1 555 0142
Every task makes the skill better.
Skillers work in production. What they learn returns to development as a skill update, gets tested against real cases, and only a proven version goes back to work.
Hire by the skill, not the seat.
A Skiller is an identity plus a set of versioned skills. Start from a role, or put one together skill by skill. Every skill says what it does on its own and what it brings to you first.
An example team. Pick anyone to see what they handle alone and what they ask about.
Lena
Inbox manager
Handles on its own
- Sorts and labels new email
- Drafts replies in your voice
- Books meetings inside your hours
Asks you first
- Replying to someone new
- Anything that commits money or a date
Give work wherever you already talk.
Same Skiller, same memory, on every channel. Start a task on WhatsApp, follow up by email and review it on a call. Your team replies on the channel you used.
- Always on
- Skillers work through the night and pick up where they stopped.
- They come back to you
- When a task is done, or when they need a decision, they message you.
- Goals, not only tasks
- Give a goal and they keep working toward it until it is met or they run out of credits.
Dev
online
Can you get the Q3 board deck to investors by Friday?
Yes. Draft to you by Wednesday noon. After you approve, I’ll send it Thursday morning.
I’ll use last quarter’s template and the numbers Noor sent on Monday.
Perfect. Keep the appendix short.
One request: get the Q3 board deck to investors by Friday.
A name, an inbox, a number. And a persona you shape by talking.
Every Skiller gets its own email address and phone number, so people can reach them like any colleague. To change how one behaves, tell them. The persona updates and you can see exactly what changed.
Lena
Inbox manager
- [email protected]
- Phone
- +1 555 0193
- Same number
- Hours
- Always on
Persona
- Warm and direct. Writes in short paragraphs.
- Signs off as Lena from the support team.
- Offers a call when a thread passes three replies.
- Gives delivery estimates when asked.
Talking to Lena
Send an instruction below and watch the persona on the left change.
Autonomy is earned, one skill at a time.
During setup you answer a few questions about money, customers and production. That sets what counts as high risk. At first every task asks you. As a skill keeps passing its checks, you can let it run on its own and log off.
Your rules
Click a task to move it. Locked ones are set at setup.
Runs on its own
Reports back when done
Asks first
Waits for your yes
Always asks you
High risk, every time
draft-replies, version 3
You set it to run alone after 25 clean runs in a row.
19 of 25 clean runs
Until then it asks before sending. One failed check resets the count, and you can take autonomy back at any time.
Skills are distilled from real work, then proven.
Skiller runs on skilled, our open-source skill platform. It turns how work gets done into a skill, tests it against realistic cases, and only promotes a new version when it beats the old one without breaking anything.
# observation, reviewed and approved by you skill: refund-policy task: "Refund order 4471, returned on day 14" outcome: corrected feedback: "Shipping is never refunded. Round down." privacy: credentials redacted review: approved → becomes a test case
A real run, not a mockup.
A refund-policy skill refunded shipping, rounded up and denied valid returns after day 14. skilled caught it with a real Codex evaluation. Fixing the skill to match the policy turned five failures into passes without changing a single test.
Hover a case to see its name
13/18
known cases passed
–/10
unseen holdouts passed
One skill on one engine, recorded with skilled. A pass means the case’s script assertion passed; it is not an accuracy guarantee. See the full evidence
The skill layer is open source.
skilled is the platform every Skiller’s skills are built and tested on. It is free under Apache-2.0, so you can read how your team is evaluated, run the same tests yourself, and improve a skill for your own use.
DeadHead04/skillEd
- License
- Apache-2.0
- Runs on
- macOS, Linux, Windows
Tests skills on
$ curl -fsSL https://github.com/DeadHead04/skillEd/releases/latest/download/install.sh | bashWhat skilled tests with today, and where your Skillers work.
- Codex
- Claude Code
- OpenCode
- Qwen Code
- Qoder CLI
- Custom HTTP engines
- Docker isolation
- OpenSandbox
- MCP server
- GitHub Actions
- Rule checks
- Script judges
- Model judges
- Multi-turn cases
- Simulated users
- Holdout suites
- Strict compare
- JUnit and HTML reports
- Anthropic evals.json
- Slack
- Phone calls
- Video calls
- Approval requests
- Goal tracking
Questions and answers.
An agentic worker: an AI that owns tasks end to end, with its own name, email address, phone number and persona, plus a set of tested skills. You hire one or a whole team, then give them work on the channels you already use.
Hire your first Skiller.
Tell us the first job you want done. We set up your team with you and agree on the rules before anything runs.