graphify-link

graphify-link

Intent

Use this skill only when the user asks to link/init graphify, add coverage, start watch for an existing graph, or unlink/remove graphify.

• Unlink words: unlink, ยกเลิก, cancel, remove, ลบ, ถอด
• Add words: add, เพิ่ม, append, include, ใส่
• A bare path means "use this as the root" unless it follows an add word, then it means "add this submodule"

Workflow

1. Resolve intent and root

   • Root defaults to cwd unless args provide a root path.
   • If unlink intent, remove the selected root from ~/.claude/graphify-roots.json, stop graphify watch, delete ~/.claude/graphify.marker, then report one line.
   • If graphify.marker already exists for this root, report that it is already linked and stop.

2. Decide what work is needed

   • If **/graphify-out/GRAPH_REPORT.md exists, prefer using the existing graph.
   • For add intent, run /graphify on the provided submodule path, then graphify merge-graphs <root>/graphify-out/graph.json <path>/graphify-out/graph.json --out <root>/graphify-out/graph.json, then graphify cluster-only <root>.
   • For re-init intent, ask which paths to extract; default to ..
   • For an existing graph with no add/re-init request, skip extraction and just start watch.
   • For no existing graph, ask once for confirmation and run /graphify on . unless the user provided specific paths.

3. Register and install fast path

   • Add root to graphify-roots.json if missing, using a case-insensitive duplicate check.
   • Copy graphify-build.py and graphify-update.py from this skill directory into the project root only when missing.
   • Do not overwrite project copies during initial link. Version upgrades are handled by reset-graphify.ps1 using the # graphify-driver-version: N stamp, which must stay near the top of each driver.

4. Start watch and finish

   • Read the highest-level GRAPH_REPORT.md so the session has graph context.
   • Stop any existing graphify watch, then start graphify watch <root> in the background.
   • Write ~/.claude/graphify.marker as:

     {"root":"<root>","timestamp":"<ISO timestamp>"}
     

   • Confirm with one line: report loaded, root registered, watch started.

Guardrails

• /graphify is the slow semantic extraction; graphify watch is the fast AST daemon.
• If extraction fails for one selected path, continue with the rest and summarize failures at the end.
• If merge fails, stop before clustering.
• If clustering fails after a successful merge, warn that graph.json was updated but clustering needs a retry.
• Runtime hook: point settings.json at hooks/intercept-graphify-skill.py (cross-platform, no PowerShell needed). The .ps1 shim in the same directory still works for existing configs.