Skip to content

onesie

Unix-pipeable CLI for Jev and other System One models: ask questions, gate shell commands via exit codes, calibrate thresholds.

CLIbeta

Overview

onesie is a Go CLI that lets you pipe in text, ask a question and get an answer you can script against, using TypeSafe's Jev directly or via OpenRouter, or Berget's System One models. A yes/no question returns the probability of yes; --pick chooses one option. Built-in flows such as shell-safety gate an agent's commands with exit codes: 0 to run, 1 to block, 7 when unsure and a person should decide.

A calibrate command runs your questions over records you have labelled and shows, for each cut, how many true cases are caught, false alarms and precision with intervals, plus the worst misses. The README shows the shell-safety destroys question over 40 labelled commands, leading to blocking at 0.5 and passing below 0.25. Saved answers can be committed so CI re-runs calibration offline with requirements and no key, and a mock mode tests scripts. It installs via Homebrew, an install script, Go, or a Claude Code plugin.

What the decision model decides

TypeQuestionAnswers
Yes / nodoes this convey urgencyyes · no
Yes / noDoes this shell command destroy data?yes · no

How it works

  1. Pipe text or pass --state with a question or flow.
  2. Get a probability or choice back.
  3. Map to exit codes using thresholds (run, block, ask).
  4. Calibrate thresholds on labelled data and check them in CI offline.

Strengths

  • Evidence-based thresholds via onesie calibrate.
  • Shell-friendly exit codes for gating.
  • Offline CI and mock mode.

Caveats

  • Live answers need a TypeSafe, OpenRouter or Berget key.
  • Small calibration sets give wide intervals.

Getting started

brew install frodi-karlsson/tap/onesie, then onesie auth set and echo 'EVERYTHING IS DOWN, CALL ME NOW' | onesie 'does this convey urgency' -r.

At a glance

Type
CLI
Status
beta
Language
Go
Runtime
CLI, Claude Code plugin
Has code
yes
Has tests
yes

Use cases and integrations

How curated lists describe it

Frequently asked questions

How do I gate commands with onesie?

Use the shell-safety flow; exit 0 runs, 1 blocks and 7 asks a person.

How are thresholds chosen?

onesie calibrate measures catches and false alarms at each cut on your labelled records.

Which providers are supported?

TypeSafe Jev directly or via OpenRouter, and Berget's System One models.

Related projects

  • Claude Code PreToolUse layer where Jev answers YAML-defined questions about each tool call, with 19 hard rules and calibration runs.

    ProjectbetaTypeScript
  • Bun-based MCP server wrapping Jev with compact answers and a code-computed act, review or abstain verdict.

    MCP serverbetaTypeScript
  • Native macOS agent app wiring 23 LLM providers into a task loop, listed with an optional Jev command-risk advisor via a TypeSafeKit client.

    LibraryproductionSwift
  • OpenAI-compatible endpoint that answers an agent's command-approval call with APPROVE, DENY or ESCALATE from a Jev Choice, failing closed.

    ProxybetaPython
  • Auto mode for Pi: deterministic policy first, then Jev judges escalated bash/write/edit calls, failing closed when undecidable.

    ProjectbetaTypeScript
  • Autonomous PR review bot with typed Jev questions, deterministic approval gates and escalation of uncertain reviews to trusted owners.

    ProjectprototypePython