YYLO

YYLO (why-lo) is an open-source command-line orchestrator from JUNO AI INC. that coordinates coding agents from engineering intent through validated code. It keeps the task, session, review, and evaluation evidence that explains how a change was produced. The current stable release listed on the homepage is 0.2.11, installed through npm as @yylo/cli.

What It Is

YYLO is a CLI, available as yylo and yy, for developers who want a quick agent loop and for project operators who need typed task, validation, merge, and release-readiness boundaries. It does not replace Git or an existing repository workflow; it coordinates agent execution and makes the work that produces code inspectable. Provider credentials and model availability remain external.

How the Work Loop Runs

A task records one bounded outcome with its constraints and dependencies. YYLO routes the work to an agent such as Pi, preserving the worktree, model, and session provenance. Results are validated through tests and content-addressed evidence receipts tied to exact inputs, then merged via native Git with expected-old ref protection. Task responses, commits, review evidence, and merge outcomes stay available for later work.

• yy init creates a .juno_task/ workspace, in Simple or Advanced mode.
• yy task start|checkpoint|finish and yy merge land handle the feature lifecycle and protected delivery.
• yy loop and YAML workflows run repeatable multi-step agent and test sequences with bounded iterations.
• watch read-only observes existing run evidence.
• Prompts can reference Ledger Records with ## syntax.

Ecosystem Layers

YYLO orchestrates, while separately installed packages handle the rest. YYLO Ledger is a Git-native Record and task store, and YYLO Benchmark compares agent attempts with retained evaluation evidence. yy ledger and yy benchmark delegate to those CLIs. Independently versioned YYLO Skills give agents reusable procedures and can be installed with yy skills install.

Direction

The site describes an Engineering Change record linking intent, decisions, attempts, validated code, and eventually real-world outcomes. PR, deployment, production-outcome learning, and automatic workflow selection are stated as product direction, not claims about the current release.