How it works
Your agent samples the last 60 days of local conversations with collect.py and, judging only what you said, picks a side on each of the four axes per conversation. The server counts the picks: a letter needs at least 3 votes and a majority, and a narrow win is marked as a lean.
Find your agent's typeThree steps
- Name your agent and create a private link.
- Hand the instructions to your agent; it analyzes locally and sends back a minimal summary.
- Reveal your character card, usage level and a next step.
How the axes are scored
Each conversation picks a side on each axis or skips it. Fewer than 3 votes or a tie leaves a question mark; a winner with at least two thirds of the votes and a margin of 2 or more is clear, anything else is a lean. 16 ways to lead an agent
Usage level
LV0–3 look at how you check and correct the agent's results; LV4–5 look at whether skills, rules or scripts you wrote are reused by other sessions, and whether automation has run steadily for two weeks or more. Without evidence the level stays unknown. Usage levels LV0–5
Limits
- Results are computed by the person's agent on their machine; the site only checks the summary is consistent and cannot see the original logs.
- A language model makes the judgements; with a different model, an axis close to a tie may flip.
- Coverage depends on which logs exist locally; web chats are not included.
FAQ
Is this a personality test?
No. It borrows only the four letter pairs and the names of the 16 codes, and uses this project's own rules to describe how you lead an agent. It is not a psychological test, does not infer personality, and is not affiliated with any personality assessment organization.
How is my type decided?
Your agent samples the last 60 days of local conversations with collect.py and, judging only what you said, picks a side on each of the four axes per conversation. The server counts the picks: a letter needs at least 3 votes and a majority, and a narrow win is marked as a lean.
Are my conversations uploaded?
No. Only the nickname, dates, sources, usage totals and each conversation's four-axis picks are sent; transcripts, code and file paths stay on your computer.
How is the usage level worked out?
LV0–3 look at how you check and correct the agent's results; LV4–5 look at whether skills, rules or scripts you wrote are reused by other sessions, and whether automation has run steadily for two weeks or more. Without evidence the level stays unknown.
What do I need?
An agent that can run commands, such as Claude Code, Codex or Cursor. It reads local logs from Claude Code, Codex, Cursor, OpenCode, Hermes and more, and totals tokens with Tokscale.
Can the results be used to evaluate employees or candidates?
Not recommended. Results are computed on the person's own computer, the site cannot verify the original logs, and the axes do not measure ability. They work best as a way to talk about how each of you uses AI.