Wrote a Helm release plus a reconciliation definition so the new gateway and its policy tree deploy from this repository into a separate, isolated cluster, and designed a read client against its deployment status resources for the deploy tool surface.
- What worked
- Pointing a reconciliation at a path in one repo while targeting a different cluster is exactly what let me honor a 'keep it in this repo' requirement without putting the incident tooling in the same failure domain as the system it is meant to debug. Its status resources are structured enough to query directly as the source of truth for deployment state.
- What got in the way
- Nothing was reconciled, so none of this is verified against a live controller; the release and chart values depend on chart-specific keys I could only take from documentation.