Authoritative JSON Schemas and greppable field indexes for core Kubernetes, OpenShift, the Flux ecosystem, and a broad set of CNCF projects — an LLM-friendly kubectl explain for manifest work, no cluster required.
Five read-only tools with clear verb-noun naming (get_, list_, search_). All have descriptions (avg 95 chars) and documented input schemas with typed parameters. Output schemas are not explicitly documented in the source code provided. Parameter descriptions are present but minimal (avg 45 chars). No error handling guidance visible. Tools follow single-responsibility principle well. Naming is consistent and action-oriented, supporting LLM intent inference.
Retrieve the field index (.fields.txt) for a resource, listing all available fields and their types
Retrieve JSON Schema for a Kubernetes or CNCF resource by group, kind, and version
List all available resource kinds in the catalog, optionally filtered by API group or project
List all projects in the catalog with their metadata, versions, and resource counts
Search the catalog for resources by keyword, returning ranked results
Output schemas not documented in source. LLMs cannot plan downstream calls or extract fields without knowing response structure (e.g., does list_kinds return array of objects? What fields in each object?).
Parameter descriptions are minimal (avg 45 chars). 'Optional: filter by API group' lacks context on format (e.g., 'apps', 'batch.keda.sh'). Descriptions should explain expected format, constraints, and when to use each parameter.
No error handling guidance. If get_schema is called with invalid group/kind/version, what error is returned? How should the LLM recover? No recovery hints provided.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 68 | 2026-07-28+ | v2 |
search_schemas 'limit' parameter lacks bounds. No min/max specified. LLMs could pass 0, negative, or 999999 values. Should constrain to e.g. 1-100 with default 20.
Tool descriptions lack 'when to use' guidance. E.g., when should an LLM call list_kinds vs search_schemas? When is get_fields preferred over get_schema? Ambiguity forces LLM guessing.