Plugins

The archgate CLI enforces governance deterministically — it reads ADRs and runs checks anywhere, fully offline. But the CLI is a set of passive capabilities: it can tell an agent which ADRs apply and whether code passes, yet nothing in the CLI tells an AI agent when to read decisions or how to validate its own work. That orchestration is what the editor plugins add.

A plugin wires the governance loop into your AI coding agent. With a plugin installed, your agent reads the applicable ADRs before writing code, validates after, and captures new patterns — without you copy-pasting rules into prompts or remembering to run checks.

Six editors, one governance model

Archgate ships plugins for six AI coding surfaces — Claude Code, Cursor, VS Code, Copilot CLI, Claude Desktop, and OpenCode. The governance model is the same across all of them; each plugin adapts the invocation mechanism to its host editor. The plugins overview compares them side by side.

The plugin is the workflow, the CLI is the engine

The plugins do not replace the CLI — they direct it. The deterministic checks still run in the archgate CLI; the plugin adds the agent and skills that decide when to invoke them. The CLI works fully on its own in CI and pre-commit; the plugin adds the AI-assisted half of the loop on top.

What a plugin installs

Every Archgate plugin installs the same two kinds of governance roles:

• Agents — the orchestrators. The developer agent drives the full workflow (read ADRs → write code → validate → capture) and refuses tasks that would violate an ADR. The read-only planner agent designs ADR-compliant approaches without writing code.
• Skills — focused capabilities the agent invokes at specific points: reviewer validates, lessons-learned captures, adr-author writes ADRs, onboard bootstraps a project, and cli-reference loads authoritative command syntax on demand.

The session-context model

Plugins connect to Archgate through CLI commands that supply the agent with exactly the context it needs, when it needs it. Two commands are central:

• archgate review-context — the prevention and detection input. It returns condensed ADR briefings (the Decision and Do’s and Don’ts of each applicable ADR), grouped by domain and scoped to the files being changed. Instead of loading every 100–400 line ADR into the agent’s context, the agent receives a focused ~10–20 line briefing per relevant decision. It can also include automated check results in the same call.
• archgate session-context — the learning input. It reads the editor’s own session transcript from disk so the lessons-learned skill can review what actually happened during a session — the requests, errors, and changes that earlier context compaction may have dropped — and codify durable lessons from it.

This session-context model is why the loop survives long sessions and sub-agent boundaries: the agent always recovers the full, accurate picture from the CLI rather than relying on whatever happens to remain in its conversation window. All of these commands run locally and read directly from your project — no governance data leaves your machine.

Command and protocol detail lives in the CLI reference

The exact flags for archgate review-context and archgate session-context (per-editor subcommands, —staged, —base, —run-checks, —skip), along with installation commands and authentication, are documented authoritatively at cli.archgate.dev. This page covers the model; the CLI docs cover the interface.

Installing and getting access

Plugins are installed through the CLI (archgate init wires them in during project setup, or archgate plugin install adds them to an existing project) and are distributed privately through the Archgate plugin service rather than a public package registry. The plugins overview and the per-editor guides walk through setup, and the authoritative install commands live in the CLI documentation.

Related concepts

Plugins overview Compare the six editor plugins feature by feature.

Skills The capabilities a plugin installs into your agent.

Agents The developer and planner agents that orchestrate the workflow.

The Governance Loop What the plugins wire into your AI agent.