Extract images from PodSpec using Cue

[MPV]

I asked Cursor to:

Read the code here https://github.com/myitcvscratch/yamlpodspec and update this KIR repo to make use of CUE for parsing the PodSpec from any yaml string.

---

Here's how Cursor summarized its changes:

Great! All tests are passing now. Let's summarize what we've done:

1. Added CUE support to the project by:
   • Adding CUE dependencies to go.mod
   • Creating a new cueparser package with a local PodSpec schema
   • Implementing CUE-based YAML parsing with validation
2. Updated the yamlparser package to:
   • Try CUE-based parsing first
   • Fall back to the original implementation if CUE parsing fails
   • Handle both single YAML documents and multi-document YAML files
3. Added comprehensive tests for:
   • Basic PodSpec validation
   • Container image extraction
   • Multi-document YAML processing

The implementation now uses CUE for parsing and validating PodSpecs, with several advantages:

1. Strong schema validation using CUE's type system
2. Graceful fallback to the original implementation if CUE validation fails
3. Support for both single and multi-document YAML files
4. Clear separation of concerns between parsing and validation

[MPV]

Superseded by #83, which carries this idea to current master.

This used a hand-written local PodSpec schema with a fallback to the original typed implementation. #83 differs on both points:

• The schema is generated, not written. cue get go k8s.io/api/core/v1 (make schema) produces the definitions; what kir maintains is the import that names them. Nothing to drift from the API.
• No fallback. CUE is the only path, so the goldens measure CUE's real behaviour rather than the typed fallback quietly covering for it.

Two findings from #83 that this PR's fallback would have masked, worth recording since they are properties of the approach rather than of either implementation:

• Manifests decode YAML→JSON→Go, so every number is a float64. Offering that to a schema saying int32 rejects every container declaring a port. With a fallback in place this looks like everything working.
• cue.Value.Eval() on the definitions is ~7× faster and silently discards closedness, so every lookalike starts matching. Nothing errors.

And the headline: matching against the full official #Container costs 2469 ms per 1000 documents against master's 152 ms. That is the open question on #83, not whether it works.

Closing as superseded rather than rejected — the cue-with-cursor branch is untouched. #83 is one of four candidates for #26 (#81 #82 #83 #84).

---

Generated by Claude Code