qa-e2e-security

QA — E2E Security (live cluster)

Overview

security-audit and security-audit-platform are static reviews — diff, code paths, scanner output. This skill is the dynamic counterpart: it fires real requests at the live cluster and confirms the security invariants actually hold at runtime. It catches policy-materialization regressions, header middleware regressions, and audit-event regressions that static review and unit tests miss.

Threat profiles in scope:

• anonymous → user (no key, bad key, expired session)
• user → admin (UI key vs admin key vs ingest-only key)
• agent A → agent B (grant scoping by humanID/agentID/serverRef)
• low trust → high trust tool (consent gating)
• internal pod → secret (reach API/proxy/operator secrets unprivileged) → for cluster RBAC/PSS coverage, defer to k8s-hardening-audit.

Regression evidence contract: for each touched security surface, capture at least one live allow path and one live deny path, plus audit/log/header evidence where that invariant applies. Static code review can explain a finding, but it cannot close this skill as passed.

Step 1 — Confirm precondition

kubectl config current-context | grep -qx kind-mcp-runtime \
  || { echo "Run qa-cluster-bringup first"; exit 1; }
./bin/mcp-runtime cluster doctor

Pull the three key types up front; never echo them into the report.

UI_KEY="$(kubectl get secret mcp-sentinel-secrets -n mcp-sentinel \
  -o jsonpath='{.data.UI_API_KEY}' | base64 -d)"
test -n "$UI_KEY" || { echo "Failed to retrieve UI_KEY"; exit 1; }
INGEST_KEY="$(kubectl get secret mcp-sentinel-secrets -n mcp-sentinel \
  -o jsonpath='{.data.INGEST_API_KEYS}' | base64 -d | cut -d, -f1)"
test -n "$INGEST_KEY" || { echo "Failed to retrieve INGEST_KEY"; exit 1; }

Step 2 — Choose mode

• head-only. Run the full backend + UI security matrix.
• git-range (BASE=<merge-base>, default origin/main). Trim by diff.

Sub-suites by changed paths: