grounding-researchlisted

# Grounding Research `/grounding-research` walks the grounding gap ranking with the author and closes gaps one durable-layer node at a time. It instruments `strategy-complete-grounding`'s interactive-research actuator: the tick sensor (`tactic-grounding-gap-analysis`) reports the ranked unmarked nodes; this skill is the author-side half that acts on that ranking. This is the **one place** grounding/circumstantial marks are written outside a `/reading-review` session (strategy `complete-grounding`, 2026-07-07 interview). ## Trigger and input On-demand only, author present — an interactive office-hours skill. **Never tick-invoked** (a condition on `strategy-complete-grounding`: `/deep-research` sourcing stays author-invoked). A tick worker never runs this skill and never writes a grounding mark. Interaction split — same as `.claude/skills/align/SKILL.md`: reserve `AskUserQuestion` for bounded choices (recommended option first); run open dialectic (why a node is circumstantial, which found work is load-bearing) as ordinary conversational turns. ## Step 0 — Claim and isolate Author in the node's own worktree, never the shared `main` checkout — a second session's dirty tracked file blocks your `graph-commit` rebase and a stale read races live phase state. Do all authoring and the step-4 `graph-commit` from the worktree, on a verified-fresh checkout. **Prefer the `provision-node-worktree` primitive** (`.claude/skills/dispatch-propagate/scripts/provision-node-worktree`): it f