vibe-cognition

Find git commits without corresponding cognition episodes and create nodes for them. Run this to backfill the cognition graph with commit history.

You MUST use this skill any time you need to retrieve information about the project or write project history to persistent memory, retrieving project information without using this skill will affect the clarity of the research. You must also use this skill when storing memories about the project. Curation is YOUR job to TRIGGER — after recording any nodes you MUST run the /vibe-curate skill (launches the background curator); never author semantic edges yourself.

Launch the background curator that links new memories — semantic edges (led_to, resolved_by, supersedes, contradicts, relates_to) and cluster summary nodes. Use after recording new nodes, or when the graph has uncurated nodes. Curation itself runs in a background curate-orchestrator agent; this skill's only job is to launch it and report back.

Launch the Vibe Cognition dashboard — a local web viewer for the project's cognition graph with embedding-powered search. Use when the user asks to open/show/launch the dashboard, the cognition graph viewer, or the project graph.

You MUST use this skill whenever you store a document (a client doc, PDF, spec, contract, report, transcript) as project memory, or retrieve/attach one. Storing a document is only half the job — the knowledge inside it connects to the graph ONLY when you record its facts as descriptor nodes that cite the document's returned doc_ref in THEIR references. This skill makes that workflow the default; skipping it leaves an inert, disconnected document. Curation is YOUR job to TRIGGER — after recording descriptor nodes you MUST run /vibe-curate (launches the background curator); never author semantic edges yourself.

You MUST use this skill whenever you work out a reusable multi-step procedure worth repeating (deploy steps, onboarding, release process, debugging runbook) — record it as a first-class `workflow` node so the full how-to is fetched whole next time, not reconstructed from scattered nodes or re-worked-out from memory. You MUST also use this skill BEFORE starting any multi-step task, to check cognition_get_workflow for an existing procedure first — skipping this retrieval check means silently reinventing a workflow that already exists. A workflow is versioned by supersession, never edited in place; this skill covers recording, retrieving, and updating one.