Generate k8s manifests for recipes

[canercidam]

• New command builder-playground dir which can be used like cd $(builder-playground dir)/k8s to navigate to the k8s files of the latest run
• New --k8s flag for start. Usage:
  • Checkout to this branch, do go install . or do go build . and use local binary
  • Install kompose CLI
  • To try with L1 recipe, builder-playground start l1 --dry-run --k8s
  • cd $(builder-playground dir)/k8s
  • In one terminal, do ./minikube-mount.sh
  • On the second terminal, do minikube start if you haven't started yet and do kubectl apply -f .
  • See from minikube dashboard
  • On another terminal, kubectl port-forward service/el 8545:8545
  • And builder-playground test to verify that the deployed stack can build a new block with a new tx

I tested this with the opstack recipe. It doesn't work yet but we're close.

The outputted files are useful for kickstarting a k8s setup and iterating from there. Playground doesn't deploy to k8s yet.

[claude]

Claude finished @canercidam's task in 1m 49s —— View job

---

Review of PR #404: Generate k8s manifests for recipes

• Read all changed files
• Review code quality, bugs, security, and performance
• Post inline comments and summary

Summary

This PR adds k8s manifest generation by shelling out to kompose and then patching the output. The approach is reasonable for kickstarting a k8s setup, but there are several issues to address:

Bugs:

• Command injection in minikube-mount.sh — host paths are interpolated into the shell script without escaping. Paths with spaces or metacharacters will break or execute arbitrary commands.
• :ro/:rw volume suffixes not handled — SplitN(vol, ":", 2) leaves options like :ro attached to the container path, producing incorrect mount paths.
• os.Stat for volume classification is host-dependent — whether a volume is treated as a file or directory depends on the filesystem state of the machine running generation, making output non-deterministic across environments (local dev vs CI).

Design issues:

• --k8s silently implies --dry-run — the flag description doesn't mention this. Users may expect the Docker stack to start alongside manifest generation.
• Compose file parsed 3 times — parseNamedDockerVolumes and parseComposeVolumesPerService each independently read and unmarshal the same file. Parse once and share.
• No kompose availability check — a missing kompose binary produces a confusing error. exec.LookPath check with a helpful message would improve UX.

Suggestion:

• Use gopkg.in/yaml.v3 instead of v2 — since this is a new package, v3's map[string]interface{} default would eliminate the verbose map[interface{}]interface{} casting throughout.

[claude]

Claude finished @canercidam's task in 1m 54s —— View job

---

Review of PR #404 (updated — commit 35e36b0)

• Read all changed files
• Analyze code quality, bugs, and security
• Post inline comments and summary

Summary

The new commit (35e36b0) adds session-dir mount path normalization to /data and filters temp-dir volumes from the minikube mount script. These are good fixes.

Remaining issues:

Severity	Issue	Status
Bug	:ro/:rw volume suffixes not stripped — SplitN(vol, ":", 2) leaves options attached to containerPath, breaking mount path matching and producing invalid k8s manifests	Inline comment
Design	Session-dir /data normalization is done in two independent places (line 53-60 and line 211-222) that must stay in sync manually	Inline comment
Design	Compose file read+parsed twice independently	Inline comment
Noted	kompose availability check — author acknowledged, will fix before merge	—

The :ro/:rw bug is the most impactful — it will silently produce wrong manifests for any volume with access mode options.