Install

Install the plugin

Agentry installs as a Claude Code plugin from its GitHub repo. From a Claude Code session in your repo, add it, install it, and let /agentry:init draft its process file from what it finds:

$ /plugin marketplace add silentthreadstudio/agentry#v0.4.2
$ /plugin install agentry@agentry-dev
$ /agentry:init

That's the whole setup. The conducting front door (/agentry:go), the SDLC specialists, and the durable memory layer all run from here — entirely from the terminal, no account and no build step. Head to Quickstart to hand it a first task.

The #v0.4.2 suffix pins the install to the 0.4.2 release; the repo's main branch moves with development. To move to a later release, remove the marketplace and add it again with the new tag. Each release is also on GitHub Releases as a zip.

OpenCode

What you are installing

One npm package, agentry-opencode, used in two ways:

• OpenCode loads it as a plugin. opencode plugin add writes an entry in opencode.jsonc, and OpenCode fetches its own copy into its own cache. That copy is not on your PATH.
• You install or run it as the CLI. npm install -g agentry-opencode (or npx agentry-opencode) gives you the agentry command, which the agents run in the shell. The same package carries the agents, commands and skills that agentry init writes for OpenCode to read.

So the plugin and the CLI are the same package installed twice, for two readers: OpenCode, and your shell.

The plugin gives the agents native tools, with structured input and output and OpenCode's permission prompt on writes:

Tool	What it does
agentry_board	Reads the board: status, check, explain, contract, next-id, commits, scan, draft list, instructions.
agentry_board_write	Changes the board: add, set, fix, draft. Previews unless --apply is passed.
agentry_lane	Reads lanes: session start, the sweep plan, a lane's ledger (exec show).
agentry_lane_drive	Drives lanes: exec steps, merge, sweep --apply, note, spike, session close.
agentry_mem	Reads memory: recall, search, stats.
agentry_mem_write	Changes memory: write, update, forget, feedback, episode.

Without the plugin the agents fall back to running agentry ... in the shell, so everything still works. The CLI is for the same board, lane and memory operations from your terminal or the agents' shell, plus init.

An OpenCode setup has three pieces. The package, agentry-opencode, is on npm and needs OpenCode V2 and Node 24 or newer.

Piece	What it is
CLI	agentry, which the agents run in the shell.
Agents, commands and skills	Files OpenCode reads.
Plugin	The native agentry_board, agentry_lane and agentry_mem tools, loaded from OpenCode's config. Recommended; without it, the agents fall back to the CLI.

Pick one install.

Global install

Recommended for one person across all repos. Run these in this order:

$ npm install -g agentry-opencode
$ opencode plugin add agentry-opencode@0.4.2
$ opencode service restart
$ agentry init --host opencode --global --apply

Then, in each repo, run /agentry/init in OpenCode to learn or scaffold the board, and run agentry init --host opencode --apply to add the AGENTS.md block. Run that repo init without --vendor; its payload overrides the global one. --global writes only agents, commands and skills, never touches a repo, and never removes your own files there.

Per-project install

Pins the version in the repo and can be shared by committing .opencode/:

$ opencode plugin add agentry-opencode@0.4.2
$ opencode service restart
$ npx agentry-opencode@0.4.2 init --host opencode --vendor --apply

Then run /agentry/init in OpenCode, and run the same npx agentry-opencode@0.4.2 init --host opencode --vendor --apply again to add the AGENTS.md block. --vendor copies the CLI into the repo's .agentry/, so nothing global is needed.

Pin the plugin version

Write agentry-opencode@<version> in opencode plugin add, never @latest: OpenCode caches the version it first fetched for @latest and does not refetch it, so an update silently does nothing. Keep exactly one agentry-opencode entry in opencode.jsonc, because running plugin add again adds a second line.

Updating

Global:

1. Edit the version in the existing "agentry-opencode@<version>" line in ~/.config/opencode/opencode.jsonc, or run opencode plugin add agentry-opencode@<new> and delete the old line.
2. opencode service restart
3. npm install -g agentry-opencode@<new>
4. agentry init --host opencode --global --apply

Per project: update the plugin line and restart (steps 1 and 2), then rerun npx agentry-opencode@<new> init --host opencode --vendor --apply in the repo. Init previews changes, rewrites only Agentry's generated files, keeps your own files, and does nothing when already in sync.

Check it worked

• agentry --version prints 0.4.2.
• opencode debug agents lists agentry/conductor and the 8 specialists.
• In OpenCode, typing /agentry shows the commands. Commands are namespaced: /agentry/init, not /agentry:init.

Without --apply, init previews. The AGENTS.md block waits for a learned process, so a first run in a repo skips it and says so.