Architecture: per-tool integrations as separate Claude plugins that depend on graphifyy?

[serithemage]

Hi — I'm working on a three-tier Claude Code integration on top of graphify:

1. Passive — existing ## graphify section in CLAUDE.md
2. Auto — a new UserPromptSubmit hook (python3 -m graphify.auto_context) that extracts keywords from every user prompt, queries graph.json, and injects a <graphify-context> subgraph block. Always exits 0 so a missing/broken graph never blocks the prompt.
3. Active — MCP server auto-registered in .mcp.json (not just settings.json).

I have a working implementation + tests. Before opening a PR against v5, I wanted to ask about direction.

Observation. _PLATFORM_CONFIG now covers 9+ tools (Claude, Codex, OpenCode, Cursor, Gemini CLI, Copilot CLI, Claw, Droid, Trae, Antigravity), and __main__.py is ~65 KB. Recent issues like #522 (bash-only hook fails on PowerShell) suggest the per-platform install logic is starting to show stress at the seams.

Question. Is the roadmap to keep merging per-tool integrations into graphify/__main__.py, or would you prefer to push new tool integrations into separate Claude plugins that declare graphifyy as a dependency? Claude plugins bundle slash commands + hooks + skills + .mcp.json in one manifest, which matches the three-tier shape exactly and would let graphify stay focused on the graph itself.

I'm happy to go either way — either open a PR against v5, or extract this as a standalone graphify-claude plugin and just link to it from the README. Just want to avoid burning review cycles on a direction you don't want.

Thanks!

[serithemage]

Closing as duplicate of #146 — same proposal with more user signal already collected there. Continuing the conversation on the issue: #146 (comment)