CLI: coordinate and deliver
Choose agent sessions, prepare task workspaces and verify Git delivery. External agents implement, test and commit; native merge launches no models or test suites.
Getting started
Evaluate the system, install one product, or operate an existing project. You do not need to adopt every layer at once.
Evaluating
See what each product owns—and what it does not.
Installing
Copy exact stable versions, then verify installed help.
Already using YYLO
Find current lifecycle, Record and evaluation guidance.
Choose agent sessions, prepare task workspaces and verify Git delivery. External agents implement, test and commit; native merge launches no models or test suites.
Own tasks, dependencies, wiki, workflow definitions and artifact Records. The website explains the system; it does not duplicate or override your Ledger state.
Review cases, run independent attempts and evaluate retained outputs with checks or judges. Frozen criteria and deterministic loss do not turn an experiment into production authorization.
Eight independently versioned procedures for CLI, Ledger and Benchmark. Review canonical instructions before installing them into an agent environment.
Stable channels checked 2026-10-07. Node 20.10+ and Git are required for CLI/Benchmark; Ledger requires Python 3.8+. Benchmark execution additionally requires a POSIX host, tar and the selected harness. Provider credentials and costs are separate.
CLI 0.2.11 requires Ledger 0.4.0 and Benchmark 0.2.1 exactly, and Skills ^2.1.1. These are independently installed packages. Use a Python virtual environment or your approved tool environment for Ledger; ensure its executable is on PATH when using the CLI delegate.
Skills installation uses the network and changes local agent instruction directories. Customized files require review; do not force replacement just to clear a warning. Installing or upgrading the CLI alone does not update existing skills. Fresh Simple initialization installs skills by default; --no-skills opts out.
Choose Simple or Advanced workspace mode · Review changes before upgrading.
Start a hydrated worktree, implement/test/commit, finish, then have the target owner land it. Preserve interrupted work and ownership.
Retrieve by immutable ID without guessing its kind. Distinguish payload bytes from metadata and inspect omissions.
Freeze criteria before execution or explicitly record a later revision. Unknown evidence stays unscored.
This public hub owns explanations and reviewed release guidance. Installed help defines your executable surface; Ledger owns your Records; Benchmark retains experiment evidence; the Skills repository owns complete agent instructions. Publication, merge and deployment are separate permissions.