Agentic platform engineering on entirely free and open-source tooling: golden path, Score, Crossplane-shaped platform API, OPA policy gates, and an authz-gated MCP server.
12 tools with complete input schemas and descriptions, but quality is uneven. Tool names follow verb_noun convention (propose_, evaluate_, list_, get_, remediate_, parse_, check_, render_, detect_, get_policy_, list_tools, get_tool_schema). Descriptions range 40 - 180 chars, meeting the 10 - 1024 baseline. However, descriptions are generic and lack WHEN/WHY context. Parameters have types and descriptions, but lack constraints (enums, ranges, patterns). No output schemas documented. Error handling absent, tools do not guide recovery. Security concerns: check_authorization and remediate_findings accept complex objects without validation guidance. No idempotency hints for write operations (propose_manifest, remediate_findings). Composition is reasonable, tools chain via manifest/spec objects, but no pagination, no result limits documented, no chaining IDs guaranteed in responses.
Check if an identity is authorized to perform an action
Detect drift between desired and actual state
Evaluate a Kubernetes manifest against the policy gate
Retrieve the policy rules that apply to a manifest
Get details of a specific service template
Get the input and output schema for a specific tool
List available service templates
List all available tools for the current identity
No output schemas documented. LLMs cannot plan downstream calls or extract required fields. Responses are unstructured.
No error handling guidance. Tools do not indicate retryability, user-fixable vs fatal errors, or recovery steps. Agents cannot self-correct.
Descriptions lack WHEN/WHY context. Generic 40 - 60 char descriptions do not explain when to call each tool vs alternatives. LLMs cannot disambiguate.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 50 | 2026-07-28+ | v2 |
Parse natural language intent into a structured service request
Propose a Kubernetes manifest for a service based on a Score specification
Generate remediation recommendations for policy findings
Render a Kubernetes manifest from a Score specification
No input constraints (enums, ranges, patterns). Parameters like 'scenario' in evaluate_manifest declare enum values, but most tools lack constraints. LLMs hallucinate invalid values.
No idempotency or confirmation hints for write operations. propose_manifest and remediate_findings modify state but do not indicate retry safety or dry-run support.