Add wrappers for common dev/infra commands (kubectl exec, uv run, sqlc, flutter, git checkout/remote, gh auth, npm install/test, npx jest --watch, env-prefixed go run)

Summary

After running rtk discover against ~30 days of Claude Code transcripts (~34k Bash calls across Go/TS/Python/Flutter projects + EKS/OKE infra work), here are commands that show up in high volume but have no rtk equivalent today. Filing one consolidated issue rather than fragmenting.

Counts (30d, single user)

Command	Count	Typical form
kubectl exec	87	kubectl exec deployment/api -n ns -- ...
uv run	89	uv run --extra dev pytest tests/...
sqlc generate	88	~/go/bin/sqlc generate
flutter run	55	flutter run -d chrome --web-port=4000
npm install	67	npm install 2>&1
npm test	103	npm test -- foo.test.tsx
npx jest --watch	(subset of 381 npx jest)	npx jest --watch — note: npx jest without --watch does rewrite
node (script runner)	80	node tools/lint-prisma-injection.mjs
go run (env-prefixed)	74	DATABASE_URL=postgres://... go run ./...
git checkout	336	git checkout -b scratch/...
git remote	126	git remote -v
gh auth	316	gh auth status
lsof	172	lsof -nP -iTCP:5173 -sTCP:LISTEN
pkill	104	pkill -f "mcp-chrome-..."

Notes

• npx jest --watch specifically exits 1 from rtk rewrite even though npx jest (no --watch) rewrites fine. Watch-mode flags should probably pass through with a rewrite, or at least short-circuit cleanly.
• git checkout / git remote are surprising omissions given git status/log/diff/add/commit/push/pull are all wrapped. Output for these is short and rarely needs filtering, but consistency matters — agents reach for rtk git ... and lose the rewrite.
• gh auth status output is small but called constantly in multi-account workflows. Low savings, high volume.
• kubectl exec is the big k8s gap — long output (logs, shells), called regularly during debugging.
• uv run pytest mirrors npx jest pattern — runner-prefix + test command.
• flutter run has noisy hot-reload chatter that would benefit from filtering.

Suggestion

Prioritize by token impact: kubectl exec > uv run pytest > flutter run > npm test/install > node <script> > git checkout/remote > gh auth > lsof/pkill.

Happy to test rewrites on real workloads if you ship a branch.

Side note (separate concern)

rtk discover's "Already using RTK: N (0%)" appears to count literal rtk strings in transcript tool_input.command, which stores the pre-hook command. When the PreToolUse hook rewrites via updatedInput, discover doesn't see it — so hit rate looks ~0% even when rtk gain shows 77.8% lifetime savings. Worth a separate issue if you want one filed.