Docs · Overview

How Agentry fits together

Agentry is an adaptive agentic layer for Claude Code and OpenCode. It routes a task to the least process that wins, conducts the work with SDLC specialists, and gates the result with you — so the bar is held instead of hoped for.

Every task enters through one front door — /agentry:go — and gets routed to the least process that wins: a clear, reversible change runs one-shot; an under-specified goal is shaped spec-first before any code; a multi-file, structurally non-trivial change is run decompose+verify — sliced into bounded tasks, built by specialists, and checked by an independent verifier. Agentry conducts that work with SDLC specialists (product, architect, implementer, verifier, librarian) and gates the result with you at the decision points that matter.

What's in the box

  • 9 agents, each with one job, called only when the work earns them.
  • 15 skills that carry the craft the agents load: architecting, planning, implementing, testing, reviewing, designing and more.
  • Your board, as lanes. /agentry:exec runs a task whose contract is already written in its own worktree, with a builder, an independent verifier and your gate; /agentry:merge lands it. 17 board rules mirror the checks you already run. See Lanes.
  • A local memory, agentry mem. Decisions, gotchas and repo facts as plain files, recalled when a task needs them. See Memory.
  • One CLI, agentry. Every move the slash commands make is a command you can script.

New here? Start with the Quickstart, or install Agentry and run your first task through /agentry:go.