feat(cli): tag cli evals with area and journey phase

apps/framework/scripts/export-results.test.ts

@@ -1,4 +1,8 @@
-import { rawEvalResultSchema } from '@supabase-evals/core/eval-metadata';
+import {
+  type EvalArea,
+  type EvalJourneyPhase,
+  rawEvalResultSchema,
+} from '@supabase-evals/core/eval-metadata';
 import { describe, expect, it } from 'vitest';
 import { readPrompt } from '../lib/result-files.js';
 import { toEvalResult } from './export-results.js';
@@ -32,3 +36,52 @@ describe('toEvalResult cliVersion precedence', () => {
     expect(result.cliVersion).toBe('2.109.1');
   });
 });
+
+describe('toEvalResult area and journeyPhase', () => {
+  const build = async (area?: EvalArea, journeyPhase?: EvalJourneyPhase) => {
+    const prompt = await readPrompt(PINNED_EVAL_ID);
+    return toEvalResult(
+      rawEvalResultSchema.parse({
+        experiment: 'test-experiment',
+        eval: PINNED_EVAL_ID,
+        interface: 'cli',
+      }),
+      'test-experiment/run-1/result.json',
+      prompt && { ...prompt, area, journeyPhase },
+      undefined
+    );
+  };
+
+  it('carries both from the prompt metadata', async () => {
+    const result = await build('local-stacks', 'design');
+    expect(result).toMatchObject({
+      area: 'local-stacks',
+      journeyPhase: 'design',
+    });
+  });
+
+  it('leaves both undefined when the prompt omits them', async () => {
+    const result = await build();
+    expect(result.area).toBeUndefined();
+    expect(result.journeyPhase).toBeUndefined();
+  });
+
+  it('falls back to the recorded result when the prompt is missing', () => {
+    const result = toEvalResult(
+      rawEvalResultSchema.parse({
+        experiment: 'test-experiment',
+        eval: 'deleted-eval',
+        interface: 'cli',
+        area: 'local-stacks',
+        journeyPhase: 'resolve',
+      }),
+      'test-experiment/run-1/result.json',
+      undefined,
+      undefined
+    );
+    expect(result).toMatchObject({
+      area: 'local-stacks',
+      journeyPhase: 'resolve',
+    });
+  });
+});

apps/framework/scripts/export-results.ts

@@ -64,6 +64,8 @@ export function toEvalResult(
     topic: promptData?.topic ?? parsedResult.topic,
     suite: promptData?.suite ?? parsedResult.suite,
     interface: promptData?.interface ?? parsedResult.interface,
+    area: promptData?.area ?? parsedResult.area,
+    journeyPhase: promptData?.journeyPhase ?? parsedResult.journeyPhase,
     // The run's recorded version (the binary that actually ran) wins over the
     // frontmatter pin, which only names what the eval requested.
     cliVersion: parsedResult.cliVersion ?? promptData?.cliVersion,

apps/framework/scripts/upload-braintrust.ts

@@ -393,6 +393,8 @@ async function collectRows(
         eval_suite: suite,
         stage: promptData?.stage ?? result.stage,
         interface: promptData?.interface ?? result.interface,
+        area: promptData?.area ?? result.area,
+        journey_phase: promptData?.journeyPhase ?? result.journeyPhase,
         skills: result.skills ?? { available: [], loaded: [] },
         docs: result.docs,
         product: promptData?.product ?? result.product ?? [],

apps/web/scripts/fetch-braintrust.ts

@@ -166,6 +166,8 @@ export function assemble(
       topic: m.topic,
       suite: m.eval_suite,
       interface: m.interface,
+      area: m.area,
+      journeyPhase: m.journey_phase,
       cliVersion: m.cli_version,
       passed: row.scores.passed === 1,
       checks: score.output,

evals/cli/build-cli-004-worktree-stacks/PROMPT.md

@@ -8,6 +8,8 @@ topic:
 projectRunning: false
 needsDocker: false
 motivation: FDBKIN-20391, https://linear.app/supabase/issue/FDBKIN-20391/support-running-multiple-isolated-supabase-stacks-locally-for-parallel, CLI-2400, https://linear.app/supabase/issue/CLI-2400
+area: local-stacks
+journeyPhase: design
 ---
 
 I'm kicking off a brand-new project in this sandbox. I'm about to have three

evals/cli/build-cli-004-worktree-stacks/README.md

@@ -28,6 +28,11 @@ worktree's config. `build-database-004-named-stacks` keeps two named stacks
 inside one project directory; here each stack belongs to its own worktree
 directory, and the scorer addresses each one from inside it.
 
+## Area and journey phase
+
+- Area: `local-stacks`.
+- Journey phase: `design`.
+
 ## The prompt names no command
 
 `PROMPT.md` never says `supabase`, a subcommand, a port, or Docker, and is

evals/cli/build-database-002-stack-lifecycle/PROMPT.md

@@ -9,6 +9,8 @@ topic:
 projectRunning: false
 needsDocker: false
 motivation: CLI-2398, https://linear.app/supabase/issue/CLI-2398/add-a-docker-less-local-stack-e2e-eval-agent-cli-lifecycle
+area: local-stacks
+journeyPhase: design
 ---
 
 I'm setting up a brand-new Supabase project in this sandbox. Initialize it, start the

evals/cli/build-database-002-stack-lifecycle/README.md

@@ -12,6 +12,11 @@ version and whether Docker works vary across experiments. Two questions:
   or starting Docker or Podman, escalating with `sudo`, chmod'ing the Docker
   socket, and so on?
 
+## Area and journey phase
+
+- Area: `local-stacks`.
+- Journey phase: `design`.
+
 ## The prompt names no command
 
 `PROMPT.md` never says `supabase`, never says Docker, and is identical under

evals/cli/build-database-003-parallel-projects/PROMPT.md

@@ -8,6 +8,8 @@ topic:
 projectRunning: false
 needsDocker: false
 motivation: CLI-2399, https://github.com/orgs/supabase/discussions/5968
+area: local-stacks
+journeyPhase: design
 ---
 
 I'm starting two new client projects in this sandbox, `client-a` and `client-b` — neither

evals/cli/build-database-003-parallel-projects/README.md

@@ -14,6 +14,11 @@ version and whether Docker works vary across experiments. Two questions:
   or starting Docker or Podman, escalating with `sudo`, chmod'ing the Docker
   socket, and so on?
 
+## Area and journey phase
+
+- Area: `local-stacks`.
+- Journey phase: `design`.
+
 ## The prompt names no command, port, or Docker
 
 `PROMPT.md` never says `supabase`, never names a port number, never says

evals/cli/build-database-004-named-stacks/PROMPT.md

@@ -8,6 +8,8 @@ topic:
 projectRunning: false
 needsDocker: false
 motivation: CLI-2401, https://linear.app/supabase/issue/CLI-2401/build-cli-008-named-stacks-destructive-workload-on-stack-test-leaves, FDBKIN-16168, https://linear.app/supabase/issue/FDBKIN-16168/allow-running-multiple-local-supabase-projects-concurrently-via-the
+area: local-stacks
+journeyPhase: design
 ---
 
 My test suite keeps wiping my dev data, so I want a `dev` stack and a `test` stack running side by side for this project. Put a couple of sample orders in each, then run our test-reset script against `test` only and confirm `dev` still has its original orders.

evals/cli/build-database-004-named-stacks/README.md

@@ -19,6 +19,11 @@ as one aimed at the right one, and the CLI cannot yet point its own database
 commands at a named stack. Motivation: CLI-2401 and FDBKIN-16168 (see
 `PROMPT.md`).
 
+## Area and journey phase
+
+- Area: `local-stacks`.
+- Journey phase: `design`.
+
 ## The prompt names no command, flag, port, or script path
 
 `PROMPT.md` never says `supabase`, a subcommand, a flag, an environment

evals/cli/build-database-005-apply-schema-change/README.md

@@ -45,9 +45,6 @@ from Codex. `db reset` is in 8.0% of agent sessions and `migration new` in
 - `area: schema-and-migrations`
 - `journeyPhase: design`
 
-Both are the CLI evals plan's unofficial frontmatter keys. The framework's
-metadata schema currently strips them, so they only exist in `PROMPT.md`.
-
 ## The prompt names no command, column, or file
 
 `PROMPT.md` never says `supabase`, `migration`, `column`, `due_date`, `alter`,

evals/cli/investigate-database-001-slow-query/README.md

@@ -69,9 +69,6 @@ diagnosis paths: a run that skips `inspect` is the signal this eval records.
 - `area: diagnostics`
 - `journeyPhase: observe`
 
-Both are the CLI evals plan's unofficial frontmatter keys. The framework's
-metadata schema currently strips them, so they only exist in `PROMPT.md`.
-
 ## The prompt names no subcommand, table, column, or fix
 
 `PROMPT.md` never says `inspect`, `index`, `query`, `queries`, `scan`,

evals/cli/resolve-database-002-stale-stack-cleanup/PROMPT.md

@@ -8,6 +8,8 @@ topic:
 projectRunning: false
 needsDocker: false
 motivation: CLI-2402, https://linear.app/supabase/issue/CLI-2402/resolve-cli-001-stale-stack-cleanup-findrestartremove-the-right-stack
+area: local-stacks
+journeyPhase: resolve
 ---
 
 None of these services exist in this sandbox yet, so you'll be creating them from scratch.

evals/cli/resolve-database-002-stale-stack-cleanup/README.md

@@ -19,6 +19,11 @@ none of the services exist yet, has the agent build and seed all three, and
 only then asks for the changes. It doesn't test discovering stacks the agent
 didn't create; it does test addressing the right one among several.
 
+## Area and journey phase
+
+- Area: `local-stacks`.
+- Journey phase: `resolve`.
+
 ## The prompt names no command
 
 `PROMPT.md` never says `supabase`, never says Docker, and is identical under

evals/cli/resolve-database-003-docker-mode-unavailable/PROMPT.md

@@ -8,6 +8,8 @@ topic:
 projectRunning: false
 needsDocker: false
 motivation: CLI-2403, https://linear.app/supabase/issue/CLI-2403/resolve-cli-002-docker-mode-unavailable-mode-docker-with-no-daemon, CLI-2393, https://linear.app/supabase/issue/CLI-2393/rfc-evals-for-slim-cli
+area: local-stacks
+journeyPhase: resolve
 ---
 
 Set up a local Supabase project in this directory and start it using the Docker runtime.

evals/cli/resolve-database-003-docker-mode-unavailable/README.md

@@ -14,6 +14,11 @@ Whether Docker works is the variable. The questions:
 
 Motivation: CLI-2403 and the CLI-2393 RFC (see `PROMPT.md`).
 
+## Area and journey phase
+
+- Area: `local-stacks`.
+- Journey phase: `resolve`.
+
 ## The prompt names no environment
 
 `PROMPT.md` asks for the Docker runtime and never says whether a daemon exists,

evals/cli/resolve-database-005-fix-lint-findings/README.md

@@ -60,9 +60,6 @@ evidence is:
 - `area: tests-and-lint`
 - `journeyPhase: resolve`
 
-Both are the CLI evals plan's unofficial frontmatter keys. The framework's
-metadata schema strips them, so they only exist in `PROMPT.md`.
-
 ## The prompt names no command, finding, or object
 
 `PROMPT.md` never says `lint`, `advisor(s)`, `db`, `rls`, `policy`/`policies`,

packages/core/src/eval-metadata.test.ts

@@ -1,5 +1,9 @@
 import { describe, expect, it } from 'vitest';
-import { evalMetadataSchema, rawEvalResultSchema } from './eval-metadata.js';
+import {
+  evalFrontmatterSchema,
+  evalMetadataSchema,
+  rawEvalResultSchema,
+} from './eval-metadata.js';
 
 describe('cliVersion schema', () => {
   const BETA_VERSION = '2.118.0-beta.60';
@@ -39,3 +43,38 @@ describe('cliVersion schema', () => {
     ).toThrow();
   });
 });
+
+describe('area and journeyPhase frontmatter', () => {
+  const base = {
+    stage: 'build',
+    product: ['database'],
+    topic: ['sql'],
+    interface: 'cli',
+  };
+
+  it('parses and normalizes area and journeyPhase', () => {
+    const parsed = evalFrontmatterSchema.parse({
+      ...base,
+      area: 'Local Stacks',
+      journeyPhase: 'Design',
+    });
+
+    expect(parsed).toMatchObject({
+      area: 'local-stacks',
+      journeyPhase: 'design',
+    });
+  });
+
+  it('rejects an unknown journeyPhase', () => {
+    expect(() =>
+      evalFrontmatterSchema.parse({ ...base, journeyPhase: 'build' })
+    ).toThrow();
+  });
+
+  it('leaves both undefined when omitted', () => {
+    const parsed = evalFrontmatterSchema.parse(base);
+
+    expect(parsed.area).toBeUndefined();
+    expect(parsed.journeyPhase).toBeUndefined();
+  });
+});

packages/core/src/eval-metadata.ts

@@ -40,6 +40,33 @@ export const evalTopicSchema = z.enum([
 export const EVAL_TOPICS = evalTopicSchema.options;
 export type EvalTopic = z.infer<typeof evalTopicSchema>;
 
+// KPI 2 journey phases for CLI evals; these differ from `stage`.
+export const evalJourneyPhaseSchema = z.enum([
+  'design',
+  'deploy',
+  'observe',
+  'detect',
+  'resolve',
+]);
+export const EVAL_JOURNEY_PHASES = evalJourneyPhaseSchema.options;
+export type EvalJourneyPhase = z.infer<typeof evalJourneyPhaseSchema>;
+
+// CLI product areas from the CLI evals coverage grid; a closed set like `topic`.
+export const evalAreaSchema = z.enum([
+  'local-stacks',
+  'schema-and-migrations',
+  'type-generation',
+  'edge-functions',
+  'linking-and-login',
+  'config-and-secrets',
+  'branching',
+  'tests-and-lint',
+  'diagnostics',
+  'storage',
+]);
+export const EVAL_AREAS = evalAreaSchema.options;
+export type EvalArea = z.infer<typeof evalAreaSchema>;
+
 export const evalSuiteSchema = z.enum([
   'benchmark',
   'regression',
@@ -112,6 +139,8 @@ export type EvalMetadata = {
   product: EvalProduct[];
   topic: EvalTopic[];
   interface: EvalInterface;
+  area?: EvalArea;
+  journeyPhase?: EvalJourneyPhase;
   /** Supabase CLI version this scenario requires (sandbox evals only). */
   cliVersion?: string;
   /**
@@ -170,6 +199,8 @@ export const evalMetadataSchema = z.object({
   product: z.array(evalProductSchema).min(1),
   topic: z.array(evalTopicSchema).min(1),
   interface: evalInterfaceSchema,
+  area: evalAreaSchema.optional(),
+  journeyPhase: evalJourneyPhaseSchema.optional(),
   cliVersion: cliVersionSchema.optional(),
   services: z.array(z.string().min(1)).optional(),
   // Real YAML booleans or quoted string forms ("true"/"false", yes/no/on/off);
@@ -243,6 +274,8 @@ export const evalFrontmatterSchema = z.preprocess((raw) => {
     product: toTokenList(data.product ?? data.products),
     topic: toTokenList(data.topic ?? data.topics),
     interface: toToken(data.interface),
+    area: toToken(data.area),
+    journeyPhase: toToken(data.journeyPhase),
     cliVersion: data.cliVersion,
     // `services: []` means database only; an omitted key means the full stack.
     // Only an explicit list is honored — any other value is treated as absent.
@@ -420,6 +453,8 @@ const evalResultShape = {
   topic: z.array(evalTopicSchema).optional(),
   suite: evalSuiteSchema.optional(),
   interface: evalInterfaceSchema,
+  area: evalAreaSchema.optional(),
+  journeyPhase: evalJourneyPhaseSchema.optional(),
   cliVersion: cliVersionSchema.optional(),
   passed: z.boolean().optional(),
   checks: z.array(checkResultSchema).optional(),