4.2 KiB
4.2 KiB
Per AI Blitz Plan §global. Layer: 1. Repo: devops-infra-helm-charts.
Escalation Matrix
Use this table when a task crosses a Sanctity/Boundary line, when blast radius exceeds the agent's authority, or when a request belongs in a different repo. Always escalate before writing the change, not after.
Primary owner: siddharth.pal@meesho.com Secondary owner: samarth.nag@meesho.com
Owner data is sourced from the repo-root repository.yaml. If owners change, update that file via the registry-bootstrap flow; do not edit repository.yaml by hand. See SANCTITY_RULES.md for the don't-touch list and AGENT_BOUNDARIES.md for layer classification.
Situation → owner → channel
| # | Situation | Who | Channel | Notes |
|---|---|---|---|---|
| 1 | Edit suspected to silently fork an upstream chart (touching helm-templates/<chart>/templates/ or values.yaml of a vanilla-pulled chart) |
siddharth.pal | Slack #devops-infra + PR review |
Document the intentional fork in the chart's README.md. See procedures/fork-upstream-chart.md. |
| 2 | Cross-cluster cleanup or "normalization" requested in the same PR as a feature change | siddharth.pal | Slack #devops-infra |
Refuse in-PR; request a separate cleanup PR. See SANCTITY_RULES.md §surgical-edits. |
| 3 | TruffleHog flagged a real secret in a staged commit | siddharth.pal + secret owner (BU on-call) | Slack #sec-incidents (private) + revoke pipeline |
NEVER bypass with --no-verify. Rotate the credential, then move it to External Secrets Operator. |
| 4 | Edit to repository.yaml requested |
siddharth.pal | Slack #devops-infra |
This file is owned by registry-bootstrap automation. Refuse and redirect to that pipeline. |
| 5 | Edit to an Argo CD Application / ApplicationSet manifest requested |
siddharth.pal | PR against sister repo Meesho/devops-infra-argo-config |
This repo does not own routing manifests. See coding-guidelines/argocd.md. |
| 6 | Schedule fields (nodeSelector, tolerations, computeClass) copy-pasted from one cluster's override to another without rewriting |
siddharth.pal | PR review block | Re-derive from per-cluster node-pool topology. See contour-nodeselector-tolerations-summary.md and runbooks/pod-pending-scheduling.md. |
| 7 | Chart dep version bump where upstream changelog flags breaking template changes | siddharth.pal | Slack #devops-infra + PR review |
Run helm dependency update, refresh Chart.lock, dry-run helm template against every consumer cluster. See procedures/update-chart-version.md. |
| 8 | Vault HA write-path failure (seal status, Raft peer loss, unsealed standby) | siddharth.pal + platform on-call | Slack #sec-incidents + PagerDuty |
Production secret-store outage. Do not modify Vault overrides without on-call ack. |
| 9 | Edit to manifests/storageclass/*.yaml or manifests/priorityclass/<cluster>/*.yaml |
siddharth.pal + cluster BU owner | PR review (two reviewers) | Cluster-wide singleton; affects every PVC / scheduling priority. See schemas/storageclass-priorityclass-schema.md. |
| 10 | Deletion of a versioned-sibling chart (-green, -vX.Y.Z, -latest, -old) |
siddharth.pal | Slack #devops-infra + grep sister repo |
Confirm zero references in Meesho/devops-infra-argo-config before deletion. See analyses/ADR-A2-blue-green-sibling-pattern.md. |
Refusal language
When refusing, state:
- The Sanctity rule or Layer constraint that's tripped.
- Which owner to ping (from the table above).
- The repo or pipeline the request actually belongs in (sister repo,
registry-bootstrap, Vault on-call, etc.).
See also
- AGENT_BOUNDARIES.md
- SANCTITY_RULES.md
- agent-operations-guide.md
- Repo-root
CLAUDE.mdNEVER-DO list