docs(dashboard-agent): say that the health verbs rewrite the environment's queue metrics

Author: kathiekiwi

apps/webapp/test/dashboardAgentTranscriptStore.test.ts

@@ -77,10 +77,9 @@ function textMessage(id: string, text = id) {
 
 // Compile-time: the insert reads `role` off the body and throws without one, so a
 // message that satisfies the signature must never be able to lack it.
-() => {
+const _roleIsRequired = (message: { id: string }) =>
   // @ts-expect-error a message with no role is not appendable
-  void appendChatMessageOnceByChatId(agentDb, { chatId: "chat_x", message: { id: "m1" } });
-};
+  appendChatMessageOnceByChatId(agentDb, { chatId: "chat_x", message });
 
 function toolMessage(id: string, state: "input-available" | "output-available") {
   return {

internal-packages/dashboard-agent/GUIDEBOOK.md

@@ -185,8 +185,11 @@ verb rather than hand-run Redis and ClickHouse surgery:
 pnpm --filter webapp run scenarios:watch -- --help
 ```
 
-It targets any local project and environment, is idempotent, wipes nothing, and
-prints the next dashboard step itself. Node 20 is the version it is run on.
+It targets any local project and environment, is idempotent, and prints the next
+dashboard step itself. Node 20 is the version it is run on. Everything but the
+`health:` verbs is additive; those two rewrite the environment's recent queue
+metrics in ClickHouse, so point them at an environment whose history you don't
+need.
 
 Every command takes `--project <ref-or-slug>` and `--env <dev|staging|prod|slug>`
 (default `dev`). The walkthroughs below assume a shell alias so the flags aren't

internal-packages/dashboard-agent/src/eval-error-category.test.ts

@@ -139,10 +139,9 @@ describe("the derived error category", () => {
   it("does not label a run that simply carries an error field", () => {
     // `curateRun` always emits the key, undefined when the run succeeded, and a run that
     // failed is still a tool call that worked.
-    const succeeded = redactEvalToolValue(curateRun({ id: "run_1", status: "COMPLETED" })) as Record<
-      string,
-      unknown
-    >;
+    const succeeded = redactEvalToolValue(
+      curateRun({ id: "run_1", status: "COMPLETED" })
+    ) as Record<string, unknown>;
     expect(succeeded).not.toHaveProperty("errorCategory");
     expect(toolResultErrored(succeeded)).toBe(false);
 

internal-packages/dashboard-agent/src/eval-redaction.test.ts

@@ -194,7 +194,9 @@ describe("an errored tool result", () => {
   });
 
   it("reads the derived category once the message is gone", () => {
-    expect(toolResultErrored({ errorCategory: "timeout", error: { redacted: "error" } })).toBe(true);
+    expect(toolResultErrored({ errorCategory: "timeout", error: { redacted: "error" } })).toBe(
+      true
+    );
     expect(toolResultErrored({ error: { name: "TimeoutError" } })).toBe(false);
     expect(toolResultErrored({ id: "run_1" })).toBe(false);
     expect(toolResultErrored("boom")).toBe(false);