diff --git a/.ai/spec/what/agent-modes.md b/.ai/spec/what/agent-modes.md
index 2df31913a..e50df58db 100644
--- a/.ai/spec/what/agent-modes.md
+++ b/.ai/spec/what/agent-modes.md
@@ -165,6 +165,22 @@ behavior; other specs reference it rather than redefining mode rules.
    mode. TROUBLESHOOTING mode uses its own dedicated agent instructions
    regardless of model family.
 
+## E2E Test Coverage
+
+### Troubleshooting Scenario Tests
+
+Scenario-based e2e tests verify that `mode=troubleshooting` produces structurally correct responses when diagnosing real cluster problems. Tests are gated behind the `cluster_scenario` pytest marker and require `oc` access to a live OpenShift cluster.
+
+Each test:
+1. Injects a broken cluster state via setup script
+2. Queries OLS with `mode=troubleshooting`
+3. Asserts HTTP 200, non-empty response, and presence of problem-domain keywords
+4. Cleans up cluster state
+
+Scenarios are defined under `eval/troubleshooting/scenarios/`. The test module is `tests/e2e/test_troubleshooting_scenarios.py`.
+
+[PLANNED] LLM-graded response quality evaluation via `lightspeed-eval` will layer on top of these structural tests.
+
 ## Planned Changes
 
 | Jira Key | Summary |