Opinionated multi-model delegation for very large repos.
Big work in big repos is won by decomposition + heterogeneity, not by one smart model. thunderkit turns a large change into disjoint parallel lanes and routes each to the best model and harness — asking you to pick the load-bearing ones.
npx skills add thunderock/thunderkit -s '*' -gOr one skill: npx skills add thunderock/thunderkit -s tk-router -g
Use when you need a harness or model to answer in a very limited set of simple words: enforces yes/no, one-word, number, or path answers with a hard word cap, so answers are checkable and cannot hide uncertainty in prose.
Use to check a milestone actually achieved its intent before archiving: aggregates every lane's verification, checks cross-lane integration and requirements coverage across all model families, and fails closed on orphaned or unverified requirements.
Use when a lane or verification fails and the cause isn't obvious: runs a scientific-method debug loop (symptoms, hypotheses, isolating probes, root cause, fix, regression proof) with state persisted so it survives context resets.
Use before planning to capture implementation decisions and resolve gray areas: adaptive questioning that records choices and their rejected alternatives in CONTEXT.md so tk-plan and tk-execute inherit settled decisions.
Use to generate or refresh project documentation after a big change: fans parallel doc-writer lanes then verifies every factual claim against the live codebase with a second model family, so docs match reality instead of intent.
Use to run an accepted thunderkit plan: implements disjoint lanes in parallel across the fleet via portable CLI dispatch (claude/codex), each lane in its own git worktree with a captured resumable session id.
Use before planning when a request is vague or a plan has gray areas: interrogates the harness and the user with short closed questions (yes/no, one word, a number, a path) until the brief has no unknowns. Never a paragraph.
Use to save or restore a work session in a portable format when a harness nears full context or you pause: save writes .thunderkit/HANDOFF.md (stage, lanes, resume ids, decisions, next action); restore reads north star plus handoff and resumes at the named stage.
Use to learn something the fleet doesn't know yet: researches a topic online, writes a source-backed knowledge note under .thunderkit/knowledge/, and can draft a new validated tk-* skill from what was learned — so knowledge becomes reusable, not one-shot.
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.
Use to give a project durable intent: scaffolds and maintains .thunderkit/ (north-star goals + a decision log) so the project's opinion and choices persist across sessions, agents, and model changes.
Use to turn a big-repo change into a parallel execution plan: decomposes work into disjoint, dependency-layered lanes, each file-scoped with acceptance criteria and a verification command, ready for tk-execute.
Use to investigate unknowns before planning a big change: fans parallel research lanes (library options, prior art, pitfalls, API behavior) across cheap wide models, each writing a focused finding, consolidated into RESEARCH.md.
Use to review and verify completed big-repo work: fans a diff to two-plus model families for cross-family review, consolidates findings by severity, and runs each lane's verification command so done means evidence, not intent.
Use when starting big-repo multi-model work: sizes the change, asks you to pick three model classes (one planner, a set of executors, everyone as reviewers), and routes through the thunderkit lifecycle — grill, map, plan, execute, review, ship. Entry point for the thunderkit pack.
Use to close a completed big change: gates on passing cross-family review and UAT, assembles a rich PR body from the .thunderkit artifacts, and prepares a branch for merge — never pushing or merging without your go-ahead.
Use to clarify WHAT a big change delivers before planning: runs an ambiguity-scored Socratic loop until scope, non-goals, and rejection criteria are unambiguous, producing SPEC.md that tk-plan builds on.
Use to prove the fleet configured by tk-router is actually reachable: pings every model in .thunderkit/config.json through its real harness CLI with a one-word probe and reports reachable/unreachable per model and per class before any real work starts.
Use to validate built features through conversational walk-through: turns each acceptance criterion into a real user-surface test, tracks pass/fail/gap in UAT.md that survives a context reset, and feeds gaps back to tk-plan.