Opinionated multi-model delegation for very large repos.
Use before planning work in a large or unfamiliar repo: builds or refreshes a code map (structure, entry points, ownership, hotspots) so plan and execute work from facts, not guesses.
Install: npx skills add thunderock/thunderkit -s tk-map -g
A repo too large to hold in one context window cannot be planned from memory. tk-map builds a compact, durable code map so tk-plan and tk-execute reason about real structure. Route here first whenever the repo is large, unfamiliar, or hasn't been mapped this session.
Preferred model: Fable 5.1 (wide, cheap — recon fans across many files). See ../references/model-roster.md.
Write it to .thunderkit/MAP.md (committed, refreshable):
(codegraph, scout, or similar), use it — it's cheaper and more accurate than re-reading. Name which tool produced the map. If none is available, fall back to structured file/dir inspection and say so.
serial deep read. The map is breadth, not depth — depth is tk-plan's job per lane.
tk-plan will attach one to each lane and tk-review will run it.
.thunderkit/MAP.md and note the timestamp + the tool used. Stale maps mislead; tk-plan should refresh if the map is older than the working branch's base.
.thunderkit/MAP.md with the six sections above, each area carrying its verification command. This is what tk-plan consumes to cut disjoint, file-scoped lanes along real seams.
No code-graph tool? Say the map is inspection-based (lower fidelity) and recommend which tool to install. Repo too large to fully map in budget? Map the areas the requested change touches plus their immediate boundaries, and mark the rest unmapped rather than guessing.