How I onboard into an unfamiliar repo with the knowledge graph instead of grep

Show & Tell
1

The ladder I use every time I have to make a targeted change in a codebase I don’t know well:

  1. cos_graph_communities — one call, the subsystem map: which clusters exist and how they connect. Replaces an afternoon of directory spelunking.
  2. cos_graph_context on the file I plan to touch — its neighbors, callers, and contracts at depth 1.
  3. cos_graph_impact on the function I want to change — the honest blast radius, several levels deep. This is the step that surfaces callers grep can’t connect: dynamic dispatch and indirection mean there’s no literal symbol match to find, but the graph resolved the reference at index time.
  4. Make the change, run the targeted tests, done.

A typical session: three structural queries, a handful of files actually opened. The step-3 callers are the ones that would otherwise become production incidents.

The habit shift that matters: ask “who depends on this?” before “where is this?” — the second question is grep-shaped, the first one is graph-shaped, and the first one is what safe changes actually need.