fix(init): gate unsupported --features and tag dry-run runs (#1535)

Two small `init` hardening fixes surfaced by a wizard-run triage.

## 1. Gate unsupported `--features` (source maps, metrics, attachments)

`sourcemaps`, `metrics`, and `attachments` are non-selector products —
the interactive selector never offers them, but `--features` still
accepted them, leaving a half-configured integration (e.g. source-map
upload needs an auth token this wizard does not provision). Removed from
`FEATURE_ALIASES`, `SUPPORTED_FEATURE_NAMES`, and the flag help, so they
now fail loudly (`Unknown init feature "sourcemaps"`) until their setup
flow is automated.

## 2. Tag `wizard.dry_run` in telemetry

Dry runs (`--dry-run`) apply and install nothing, so the CLI's
post-apply `verify-setup` runtime check runs the app's dev command
against an **unmodified** project and reports **`init verification
failed`** (CLI-2HP) — a false positive. Set `wizard.dry_run` once at
`runWizard` start (global scope, so every CLI event inherits it,
including that capture). `wizard.dry_run:false` then isolates real runs
from dry-run noise.

Companion server change tags the same `wizard.dry_run` for the
server-side `Verification found issues` (CLI-SERVER-22) — see below.

## Verification

- `tsc --noEmit`: no new errors in the changed files.
- `test/commands/init.test.ts`: 50/50 — updated the two normalization
tests + the "Supported features:" message, and added a `test.each`
locking in the rejection of the three gated features.
- `interactive.test.ts` + `clack-utils.test.ts`: 45/45 — the interactive
selector's filtering and the display labels are separate code paths,
unaffected by the flag change.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

Author: betegon

packages/cli/plugins/sentry-cli/skills/sentry-cli/references/init.md

@@ -18,7 +18,7 @@ Initialize Sentry in your project (experimental)
 **Flags:**
 - `-y, --yes - Accept non-interactive defaults (requires --features outside a TTY)`
 - `-n, --dry-run - Show what would happen without making changes`
-- `--features <value>... - Features to enable: errors,tracing,logs,replay,metrics,profiling,sourcemaps,crons,attachments,agent-tracing,mcp-observability`
+- `--features <value>... - Features to enable: errors,tracing,logs,replay,profiling,crons,agent-tracing,mcp-observability`
 - `-t, --team <value> - Team slug to create the project under`
 - `--app <value> - App to initialize in a monorepo (required with --yes when multiple apps are detected)`
 - `--tui - Use the Ink-based interactive UI (default). Pass --no-tui to fall back to plain log output.`

packages/cli/src/commands/init.ts

@@ -42,6 +42,12 @@ const FEATURE_DELIMITER = /[,+ ]+/;
 const NON_INTERACTIVE_USAGE_HINT =
   "sentry init --yes --features errors,tracing,replay [target] [directory]";
 
+// Only features backed by a Sentry SDK selector product are accepted here.
+// Non-selector products (source maps, metrics, attachments) are intentionally
+// not exposed via --features yet: their setup isn't fully automated — e.g.
+// source-map upload needs an auth token this wizard does not provision — so
+// accepting them would leave a half-configured integration. Re-add an alias
+// (and its SUPPORTED_FEATURE_NAMES entry) once that flow is complete.
 const FEATURE_ALIASES = {
   errors: "errorMonitoring",
   errorMonitoring: "errorMonitoring",
@@ -50,12 +56,8 @@ const FEATURE_ALIASES = {
   logs: "logs",
   replay: "sessionReplay",
   sessionReplay: "sessionReplay",
-  metrics: "metrics",
   profiling: "profiling",
-  sourcemaps: "sourceMaps",
-  sourceMaps: "sourceMaps",
   crons: "crons",
-  attachments: "attachments",
   aiMonitoring: "aiMonitoring",
   "agent-tracing": "aiMonitoring",
   agentTracing: "aiMonitoring",
@@ -68,11 +70,8 @@ const SUPPORTED_FEATURE_NAMES = [
   "tracing",
   "logs",
   "replay",
-  "metrics",
   "profiling",
-  "sourcemaps",
   "crons",
-  "attachments",
   "agent-tracing",
   "mcp-observability",
 ] as const;
@@ -339,7 +338,7 @@ export const initCommand = buildCommand<
         kind: "parsed",
         parse: String,
         brief:
-          "Features to enable: errors,tracing,logs,replay,metrics,profiling,sourcemaps,crons,attachments,agent-tracing,mcp-observability",
+          "Features to enable: errors,tracing,logs,replay,profiling,crons,agent-tracing,mcp-observability",
         variadic: true,
         optional: true,
       },

packages/cli/src/lib/init/wizard-runner.ts

@@ -986,6 +986,13 @@ export async function runWizard(initialOptions: WizardOptions): Promise<void> {
 
   const { directory, yes, dryRun, features, forceLegacyUi } = initialOptions;
 
+  // Tag the whole run's telemetry with dry-run mode. Dry runs plan changes but
+  // apply and install nothing, so post-apply verification (both this CLI's
+  // `verify-setup` runtime check and the server's `verify-changes` step) sees an
+  // unmodified project and reports expected "issues". This tag lets that
+  // expected noise be filtered out (`wizard.dry_run:false` isolates real runs).
+  setTag("wizard.dry_run", dryRun === true);
+
   // Construct the UI once for the entire run; tear down on every exit
   // path via `await using`. The factory picks `InkUI` for interactive
   // runs and `LoggingUI` for CI / `--yes` / `--no-tui`.

packages/cli/test/commands/init.test.ts

@@ -183,16 +183,12 @@ describe("init command func", () => {
       const ctx = makeContext();
       await func.call(ctx, {
         ...DEFAULT_FLAGS,
-        features: [
-          "errors,tracing,replay,sourcemaps,attachments,agent-tracing,mcp-observability",
-        ],
+        features: ["errors,tracing,replay,agent-tracing,mcp-observability"],
       });
       expect(capturedArgs?.features).toEqual([
         "errorMonitoring",
         "performanceMonitoring",
         "sessionReplay",
-        "sourceMaps",
-        "attachments",
         "aiMonitoring",
         "mcpObservability",
       ]);
@@ -203,15 +199,13 @@ describe("init command func", () => {
       await func.call(ctx, {
         ...DEFAULT_FLAGS,
         features: [
-          "errorMonitoring,performanceMonitoring,sessionReplay,sourceMaps,attachments,aiMonitoring,mcpObservability",
+          "errorMonitoring,performanceMonitoring,sessionReplay,aiMonitoring,mcpObservability",
         ],
       });
       expect(capturedArgs?.features).toEqual([
         "errorMonitoring",
         "performanceMonitoring",
         "sessionReplay",
-        "sourceMaps",
-        "attachments",
         "aiMonitoring",
         "mcpObservability",
       ]);
@@ -225,7 +219,7 @@ describe("init command func", () => {
       });
       await expect(promise).rejects.toThrow(ValidationError);
       await expect(promise).rejects.toThrow(
-        "Supported features: errors, tracing, logs, replay, metrics, profiling, sourcemaps, crons, attachments, agent-tracing, mcp-observability"
+        "Supported features: errors, tracing, logs, replay, profiling, crons, agent-tracing, mcp-observability"
       );
       expect(runWizardSpy).not.toHaveBeenCalled();
       expect(findProjectsSpy).not.toHaveBeenCalled();
@@ -248,6 +242,23 @@ describe("init command func", () => {
       expect(runWizardSpy).not.toHaveBeenCalled();
     });
 
+    test.each([
+      "metrics",
+      "sourcemaps",
+      "attachments",
+    ])("rejects %s because init does not yet automate its setup", async (feature) => {
+      const ctx = makeContext();
+      const promise = func.call(ctx, {
+        ...DEFAULT_FLAGS,
+        features: [feature],
+      });
+      await expect(promise).rejects.toThrow(ValidationError);
+      await expect(promise).rejects.toThrow(
+        `Unknown init feature "${feature}"`
+      );
+      expect(runWizardSpy).not.toHaveBeenCalled();
+    });
+
     test("passes undefined when features not provided", async () => {
       const ctx = makeContext();
       await func.call(ctx, DEFAULT_FLAGS);