refactor(provider): drop meta provider narration

Author: TheGreatAxios

src/auth/meta/constants.ts

@@ -6,7 +6,5 @@ import {
 
 export { META_BASE_URL, META_DEFAULT_MODELS, META_API_KEY_LIFETIME_MS };
 
-// Meta's minted Model API key is valid about 24h. The refresh path re-mints
-// from the stored identity token, so the skew just needs to stay comfortably
-// below the hard 24h lifetime to avoid last-second races.
+// Minted keys last ~24h; skew stays inside that to avoid an expiry race.
 export const META_REFRESH_SKEW_MS = 5 * 60 * 1000;

src/auth/meta/login.test.ts

@@ -103,7 +103,6 @@ describe("startMetaLogin (device flow)", () => {
           expires_in: 1800,
         });
       }
-      // Token poll parks until the caller aborts, then rejects like real fetch.
       return new Promise<Response>((_resolve, reject) => {
         init?.signal?.addEventListener(
           "abort",

src/auth/meta/login.ts

@@ -3,10 +3,6 @@ import { metaDeviceLogin, type MetaOAuthTokens } from "@corbits/meta-provider";
 import type { AuthProfile } from "../store.js";
 import { saveMetaProfile } from "../../config/oauth-stores.js";
 
-// Meta's device flow is a direct RFC 8628 exchange: there is no PKCE loopback
-// server and no authorization-code round trip. `startMetaLogin` adapts the
-// self-contained `metaDeviceLogin` into the TUI's `OAuthLoginHandle` shape so
-// the sign-in step can wait on a single promise.
 export type MetaLoginHandle = {
   authorizeUrl: string;
   completed: Promise<{
@@ -20,34 +16,22 @@ export type StartMetaLoginOptions = {
   profile: string;
   signal: AbortSignal;
   home?: string;
-  /** Injectable fetch for tests (device authorize / token poll / mint). */
   fetchImpl?: typeof fetch;
-  /** Injectable sleep so tests skip real wall-clock waits. */
   sleep?: (ms: number, signal: AbortSignal) => Promise<void>;
-  /** Injectable clock for expiry math. */
   now?: () => number;
-  /** Called with the device-code event so the TUI can surface verificationUri + userCode. */
   notify?: (event: {
     type: "device_code";
     verificationUri: string;
     userCode: string;
   }) => void;
 };
 
-/**
- * Runs Meta's device login. `metaDeviceLogin` owns the authorize → poll →
- * mint sequence; `notify` surfaces the verification URI + user code so the
- * caller can print them for the operator (Meta has no loopback callback).
- * The returned handle resolves with the minted tokens once the flow
- * completes, and `cancel` aborts the in-flight request.
- */
 export function startMetaLogin(opts: StartMetaLoginOptions): MetaLoginHandle {
   const controller = new AbortController();
   const abort = (): void => {
     controller.abort();
   };
-  // The TUI's cancel is a soft abandon: it lets the underlying login reject
-  // with the cancel message (the same shape the caller matches on).
+  // Abort only; the login promise rejects with the message the TUI matches.
   opts.signal.addEventListener("abort", abort, { once: true });
 
   let committed: Promise<void> | undefined;
@@ -92,10 +76,7 @@ export function startMetaLogin(opts: StartMetaLoginOptions): MetaLoginHandle {
   })();
 
   return {
-    // The verification URI is surfaced via `notify` (device flow); the
-    // authorizeUrl slot is unused for Meta, but the TUI reads it for the
-    // "open this url" login box. Keep it empty — the device code box shows
-    // the real instructions instead.
+    // Empty: notify carries the device URI; this slot is the open-url box.
     authorizeUrl: "",
     completed,
     cancel: abort,

src/auth/meta/session.test.ts

@@ -20,7 +20,6 @@ describe("isMetaTokenExpired", () => {
   });
 
   test("expired within the refresh skew window", () => {
-    // 4 minutes before expiry is inside the 5-minute skew, so treated as expired.
     expect(
       isMetaTokenExpired(
         { access: "a", refresh: "r", expiresAt: 1_000_000 },

src/auth/meta/session.ts

@@ -20,10 +20,6 @@ import {
 } from "../token-session-boundary.js";
 import { META_REFRESH_SKEW_MS } from "./constants.js";
 
-// Raised when a Meta profile cannot yield a usable access token: it is gone,
-// or its identity token has been revoked/expired (401/403 on re-mint). Carries
-// the profile name so the TUI can name the affected profile in a re-login
-// prompt.
 export class MetaAuthError extends Error {
   readonly profile: string;
   readonly reason: "missing" | "refresh-failed";

src/config/index.ts

@@ -342,8 +342,7 @@ export type ProviderCatalogEntry = Omit<
   // xAI/Grok OAuth profile. Still routes through openai-compatible; the
   // marker only controls token refresh and persistence.
   xaiProfile?: string;
-  // Set when this entry is a Meta (Muse) OAuth profile. Routes to the
-  // "meta-responses" adapter; the marker controls token refresh/persistence.
+  // Routes to the meta-responses adapter, not openai-compatible.
   metaProfile?: string;
   // Bifrost virtual key: sources use provider "bifrost" so the adapter
   // injects the x-bf-vk header; also enables /models auto-discovery scoped
@@ -431,11 +430,6 @@ export function buildXaiSource(fields: {
   };
 }
 
-// Build the InferenceSource for a Meta (Muse) OAuth profile. Routes to the
-// "meta" Responses adapter (Meta's Model API speaks the Responses protocol
-// natively at META_BASE_URL). The minted key is registered in the credential
-// cell under the source id; refresh re-mints it from the stored identity
-// token (see getValidMetaToken).
 export function buildMetaSource(fields: {
   id: string;
   profile: string;
@@ -462,11 +456,6 @@ export function buildMetaSource(fields: {
   };
 }
 
-// Build the InferenceSource for a Meta Model API key (the Meta chooser's
-// "API key" path, a hand-named meta/<slug>, or a bare `meta` settings row).
-// The endpoint speaks the Responses protocol, so the source rides the same
-// "meta" adapter as the OAuth profiles — only the credential provenance
-// differs: a plain key is api-key provenance, never an OAuth profile.
 export function buildMetaApiKeySource(fields: {
   id: string;
   apiKey?: string;

src/config/inference-sources.ts

@@ -131,11 +131,7 @@ export function buildInferenceSourceForRef(
       ...(effort !== undefined ? { reasoningEffort: effort } : {}),
     });
   }
-  // Meta Model API key rows (the Meta chooser's "API key" path, a hand-named
-  // meta/<slug>, or a bare `meta` row) also speak the Responses protocol at
-  // META_BASE_URL — never the openai-compatible /chat/completions fallback
-  // below. They have no metaProfile marker, so route them to the same
-  // meta-responses adapter as the OAuth profiles.
+  // No metaProfile on API-key rows; do not fall through to chat completions.
   if (
     ref.provider === "meta" ||
     (entry?.metaProfile === undefined && isMetaProviderName(ref.provider))

src/provider/meta-responses.ts

@@ -1,9 +1,5 @@
 import { createMetaResponsesAdapter as createPackageMetaResponsesAdapter } from "@corbits/meta-provider";
 import type { AdapterFactory } from "@intx/inference";
 
-// The packaged Meta Responses adapter is host-agnostic; Corbits Code only
-// needs the provider id wired into the adapter registry. No host quirks or
-// option remaps are required — Meta's Model API speaks the Responses protocol
-// natively and the package factory bakes the host-specifics in.
 export const createMetaResponsesAdapter: AdapterFactory =
   createPackageMetaResponsesAdapter;

src/tui/provider-setup.test.ts

@@ -98,7 +98,6 @@ function stagedLoginStarter(
   opts: {
     onStart?: (input: StagedLoginInput) => void;
     deniedStarts?: number;
-    /** Emit a device_code event via `notify` on the first start (Meta flow). */
     emitDeviceCode?: boolean;
   } = {},
 ): {
@@ -358,8 +357,6 @@ describe("provider setup pure helpers", () => {
       { name: "codex/default" },
     ]);
     expect(rows.map((r) => r.id)).toContain(CUSTOM_CHOICE_ID);
-    // Alt+A rows are keyed by the pick-list row id (two Meta rows stay
-    // distinct), which mirrors the onboarding list.
     expect(rows.map((r) => r.id)).toEqual(choices.map((c) => c.rowId));
     const openai = rows.find((r) => r.id === "openai");
     const codex = rows.find((r) => r.id === "codex");
@@ -496,7 +493,6 @@ async function mountConfig(
   return { done, harness };
 }
 
-/** Recording clipboard port for the Meta device-flow copy test. */
 function recordingClipboard(): {
   clipboard: { writeText: (t: string) => void; writes: string[] };
 } {
@@ -1258,7 +1254,6 @@ describe("runProviderSetup sign-in", () => {
     await pickRow(harness, PROVIDER_IDS, "meta/api");
     await flush(harness);
     expect(harness.captureCharFrame()).toContain("step 2 of 4");
-    // Instance name step, then key — not the login step.
     harness.pressKey("Enter");
     await harness.renderOnce();
     type(harness, "LLM|test-key");

src/tui/provider/choices.ts

@@ -207,10 +207,7 @@ export function providerChoices(): readonly ProviderChoice[] {
         if (seeded === undefined) continue;
         const choice = choiceFromDef(seeded);
         if (choice !== null) {
-          // A chooser whose paths resolve to the same catalog kind (Meta Sign
-          // In vs Meta API key, both `meta`) would otherwise put two rows
-          // under one id — the second row becomes unreachable with arrows.
-          // Give it a distinct pick-list row id while keeping the catalog id.
+          // A shared catalog id would hide the second row from arrow keys.
           const rowId = pushedRowIds.has(choice.rowId)
             ? `${def.id}/${path.id}`
             : choice.rowId;
@@ -317,8 +314,6 @@ export function addProviderSelectorChoices(
     const connected =
       choice.oauth !== null && !choice.custom && accountCount > 0;
     return {
-      // The pick-list row id so two chooser rows sharing a catalog kind stay
-      // distinct in the Alt+A selector too.
       id: choice.rowId,
       label: connected
         ? `${choice.label.split(" — ")[0]} · ${accountCount} connected`