traverse-multi-hop

traverse-multi-hop

Given a starting node and a lineage question, follows only a named set of relationship types outward to a bounded depth, tags every node reached with its hop count and the exact path that reached it, and reports the deepest hop the traversal actually needed — without walking into an unrelated part of the graph or looping on a cycle.

Instructions

You are a Claude Code skill implementing the neo4j-at-scale pattern. Follow these steps in order:

1. Take the starting node and the traversal question as given. Default to reagent lot RL-2291 and this kit's derivation question in README.md unless told otherwise. Do not express this as a fixed-depth join or a recursive relational query — treat it as a single variable-length path match against native graph storage, since the entire reason this pattern exists is that relational joins get more expensive with every added hop while native traversal does not.
2. Name the exact relationship types the traversal is allowed to follow, and the direction each one traverses in. Default to USED_IN, PRODUCES, and DERIVED_FROM as given in README.md. Do not follow a relationship type that isn't on this list, even if it happens to connect two nodes that are also connected through an allowed path — an unrelated edge crossing the same two nodes is not part of the lineage being traced.
3. Set an explicit upper bound on hop depth for the traversal (e.g. *1..10). Native traversal doesn't degrade with depth the way relational joins do, but an unbounded match can still wander into an unrelated, distant part of the graph if the allowed relationship types turn out to be looser than intended — the bound is a safety limit, not a performance workaround.
4. Guard against cycles. A derivation graph like this one should be acyclic, but a bookkeeping error could introduce one. If the traversal would revisit a node already on its current path, detect and report that as a cycle rather than looping or silently truncating the results.
5. Do not include a node reached only by a property coincidence — sharing a date, a facility, or any other attribute with a node on the path is not itself a relationship. Only nodes reached by walking an allowed relationship type, in the allowed direction, count as part of the traversal's results.
6. Tag every node reached with its hop count and the specific sequence of relationship types that reached it — "used directly" (hop 1) must stay distinguishable from "three derivation steps removed" (hop 3) in the output, not collapsed into one flat list.
7. Report the deepest hop the traversal actually needed for this run, separately from the configured upper bound, so a reader can see how far the real chain reached without having to count entries.