Getting Started · Quickstart
Quickstart
From nothing to your first conducted run. One line installs Agentry; one front door — /agentry:go — takes a task and picks the least process that wins.
Agentry installs into Claude Code as a plugin and adds a single front door:
/agentry:go <task>. You don't pick a workflow — you hand it a task and it
routes to the least process that wins. Three steps:
- Install the plugin. From a Claude Code session, add Agentry from its
GitHub repo:
No account and no build step — see Install for the channels-enabled launch and the full sequence./plugin marketplace add Codestz/agentry - Hand it a task. Describe what you want in plain language:
/agentry:go "add pagination to the users endpoint" - Stay in the loop. Agentry conducts the run and gates the forks with you — you approve the shape and the decisions, it does the work.
What to expect
/agentry:go doesn't run a fixed pipeline. It sizes the process to the task and
escalates only on evidence:
- It recalls memory first. Before acting, it pulls what your earlier runs taught it — the conventions, gotchas, and decisions that apply to this task — so the work starts warmer than from cold.
- It right-sizes the shape. A clear, reversible change goes one-shot — a single bounded edit, then verify. An ask where "done = X" isn't clear yet goes spec-first. Several coupled parts or a real design fork earn decompose + verify — spec, plan, split, build, verify, assemble.
- It gates with you. At each fork the conductor pauses for your call rather than guessing — you stay in the loop the whole way.
Requires Node ≥ 24 (the memory engine uses Node's built-in
node:sqlite) and Claude Code. Next: the full Install guide — including the channels launch — or Why Agentry if you want the case for it first.