Make your AI coding agent work like it already knows your repo.

The plugin gives Claude Code, Cursor, Codex CLI, and GitHub Copilot CLI access to the architectural context already in your repository: decisions, specs, team rules, patterns, and plans — so the agent edits code with the same constraints your team works under, not its best guess from a flat instruction file.

One install wires up architecture-aware /archcore slash commands inside your agent, four of them, no subcommands: /archcore:init, /archcore:plan, /archcore:document, /archcore:review. Everyday context needs no command at all: hooks inject the rules and specs that apply to the file being edited, and each session opens with a recap of what is decided and in progress. Use /archcore:document to record a finalized decision as an ADR (optionally codified as a team rule) or to document a module that only lives in someone's head, /archcore:plan to break a feature into a requirements cascade and an implementation plan, and /archcore:review before merge to check changed code against the documents that claim it.

Capture decisions, standards, and plans without leaving chat. The plugin uses the underlying CLI for execution, so you also get MCP tools (list, get, create, update) for browsing and editing .archcore/ documents, and session hooks that inject the relevant context automatically at the start of a conversation.

Plugin hosts: Claude Code (production), plus Cursor 2.5+, Codex CLI 0.117+, and GitHub Copilot CLI (implemented). On Copilot the plugin ships no MCP server, so wiring the project once with archcore init --agent copilot is required rather than optional — without it the agent has no document tools. The plugin needs a host with a plugin runtime; every other MCP-aware agent reads the same .archcore/ directory through the CLI, which supports eight agents over MCP and session hooks for five of them. Open source, fully local — no servers, no accounts, no telemetry. Everything is stored in .archcore/ inside your repo and versioned with your code.

Already have instruction files? You don't start over. /archcore:init detects your repo's scale and composes a full first-day seed: stack rule, run guide, data model, entry points, and specs for hotspot modules — shown in one preview and written on a single confirm. It also imports the ones you already wrote (CLAUDE.md, AGENTS.md, .cursorrules, .cursor/rules/*) as typed documents: conventions become rules, the reasoning behind them becomes ADRs, and prose already sitting in docs/ becomes guides, specs, and plans. Path-scoped instruction files keep their scope but gain status, timestamps, and relations, so you can see what is still accepted and what a later decision superseded.

Common tracks: record an ADR for a finalized decision, propose an RFC when the team needs to weigh in, document a module nobody has written down, or extend a feature plan after scope changes. The slash commands are tuned for these patterns, and the document graph means relevant rules and specs surface automatically when an agent reads a file under their scope. Reviewers can audit decisions and standards in code review like any other diff, since the documents live in Git.