@ilya-lesikov/pi-pi 0.7.0 → 0.9.0

package/extensions/orchestrator/flant-infra.test.ts

@@ -427,6 +427,7 @@ describe("flant-infra", () => {
       enabled: false,
       autoUpdate: true,
       cacheTTLDays: 7,
+      switchBackIntervalMinutes: 30,
       subscription: false,
       lastUpdated: null,
       cachedFlantModels: null,
@@ -465,6 +466,7 @@ describe("flant-infra", () => {
       enabled: true,
       autoUpdate: false,
       cacheTTLDays: 3,
+      switchBackIntervalMinutes: 30,
       subscription: false,
       lastUpdated: "2026-01-02T03:04:05.000Z",
       cachedFlantModels: ["gpt-5-4", "claude-opus-4-6"],
@@ -488,6 +490,7 @@ describe("flant-infra", () => {
       enabled: true,
       autoUpdate: true,
       cacheTTLDays: 14,
+      switchBackIntervalMinutes: 30,
       subscription: true,
       lastUpdated: "2026-02-01T00:00:00.000Z",
       cachedFlantModels: ["claude-opus-4-6", "gpt-5-4"],
@@ -508,4 +511,26 @@ describe("flant-infra", () => {
     expect(loaded).toEqual(settings);
     expect(existsSync(join(dir, "extensions", "pp", "cache", "flant-models.json"))).toBe(true);
   });
+
+  it("normalizes switchBackIntervalMinutes: default when missing, parsed, floored to >=1", async () => {
+    const dir = makeTempDir();
+    const mod = await loadFlantInfraModule(dir);
+    const settingsDir = join(dir, "extensions", "pp", "cache");
+    const settingsPath = join(settingsDir, "flant-models.json");
+    mkdirSync(settingsDir, { recursive: true });
+
+    // Missing -> default 30.
+    writeFileSync(settingsPath, JSON.stringify({ enabled: true }), "utf-8");
+    expect(mod.loadFlantSettings().switchBackIntervalMinutes).toBe(30);
+
+    // String numeric -> parsed and rounded.
+    writeFileSync(settingsPath, JSON.stringify({ enabled: true, switchBackIntervalMinutes: "45" }), "utf-8");
+    expect(mod.loadFlantSettings().switchBackIntervalMinutes).toBe(45);
+
+    // Invalid / <1 -> floored to 1.
+    writeFileSync(settingsPath, JSON.stringify({ enabled: true, switchBackIntervalMinutes: 0 }), "utf-8");
+    expect(mod.loadFlantSettings().switchBackIntervalMinutes).toBe(1);
+    writeFileSync(settingsPath, JSON.stringify({ enabled: true, switchBackIntervalMinutes: "nonsense" }), "utf-8");
+    expect(mod.loadFlantSettings().switchBackIntervalMinutes).toBe(30);
+  });
 });

package/extensions/orchestrator/flant-infra.ts

@@ -35,6 +35,13 @@ export interface FlantSettings {
    * Claude OAuth token (billed against their personal Claude subscription).
    */
   subscription: boolean;
+  /**
+   * Minutes between out-of-band "is the subscription limit cleared yet?" probes
+   * while the sub→non-sub rate-limit fallback is active. On each interval a
+   * cheap probe hits the sub model; on success the user is asked to switch back.
+   * Default 30.
+   */
+  switchBackIntervalMinutes: number;
 }
 
 const GEMINI_MAP: Record<string, string> = {
@@ -64,6 +71,7 @@ const DEFAULT_SETTINGS: FlantSettings = {
   cachedFlantModels: null,
   cachedOpenRouterData: null,
   subscription: false,
+  switchBackIntervalMinutes: 30,
 };
 
 /** Provider name for the personal-subscription Claude routing. */
@@ -245,10 +253,15 @@ function normalizeSettings(raw: unknown): FlantSettings {
   if (!raw || typeof raw !== "object") return { ...DEFAULT_SETTINGS };
   const value = raw as Record<string, unknown>;
   const cacheTTLDays = Math.max(1, Math.round(toNumber(value.cacheTTLDays, DEFAULT_SETTINGS.cacheTTLDays)));
+  const switchBackIntervalMinutes = Math.max(
+    1,
+    Math.round(toNumber(value.switchBackIntervalMinutes, DEFAULT_SETTINGS.switchBackIntervalMinutes)),
+  );
   return {
     enabled: !!value.enabled,
     autoUpdate: value.autoUpdate === undefined ? true : !!value.autoUpdate,
     cacheTTLDays,
+    switchBackIntervalMinutes,
     subscription: !!value.subscription,
     lastUpdated: typeof value.lastUpdated === "string" ? value.lastUpdated : null,
     cachedFlantModels: Array.isArray(value.cachedFlantModels)
@@ -347,6 +360,81 @@ function mapFlantToOpenRouterId(modelId: string): string | null {
   return null;
 }
 
+/**
+ * Out-of-band probe: is the personal Claude subscription limit cleared yet?
+ * Sends a tiny, fully throwaway request (NOT part of the session, so it cannot
+ * pollute conversation/context/cache) to the gateway's Anthropic endpoint using
+ * the personal Claude OAuth token. Returns:
+ *   "ok"          — 200: capacity is back (offer switch-back)
+ *   "rate_limited" — 429: still limited (stay on non-sub, retry later)
+ *   "error"       — credentials missing / network / other status (treat as not-back)
+ *
+ * The OAuth token is refreshed first so a failure is a genuine 429 and not an
+ * expired-token false negative. `max_tokens: 1` + an explicit "just respond hi"
+ * instruction keep output at ~1 token.
+ */
+// Derive the gateway probe model id (`sub/<bare-claude-id>`) from any stored
+// form: `pp-flant-anthropic-sub/sub/<m>`, `sub/<m>`, or a bare `<m>`. Exported
+// for testing the derivation without a network call.
+export function subProbeModelId(modelId: string): string {
+  let bare = modelId;
+  if (bare.startsWith(`${SUB_PROVIDER}/`)) bare = bare.slice(`${SUB_PROVIDER}/`.length);
+  if (bare.startsWith(SUB_MODEL_PREFIX)) bare = bare.slice(SUB_MODEL_PREFIX.length);
+  return `${SUB_MODEL_PREFIX}${bare}`;
+}
+
+export async function probeSubscriptionCleared(
+  modelId: string,
+): Promise<"ok" | "rate_limited" | "error"> {
+  const log = getLogger();
+  await refreshClaudeOAuthToken();
+  const oauthToken = readClaudeOAuthToken();
+  const gatewayKey = readGatewayApiKey();
+  if (!oauthToken || !gatewayKey) {
+    log.debug({ s: "flant", hasOAuth: !!oauthToken, hasGatewayKey: !!gatewayKey }, "probe skipped: missing credentials");
+    return "error";
+  }
+  // The gateway expects the bare claude-* id under the `sub/` prefix.
+  const probeModel = subProbeModelId(modelId);
+  try {
+    const res = await fetch("https://llm-api.flant.ru/v1/messages", {
+      method: "POST",
+      headers: {
+        "content-type": "application/json",
+        "anthropic-version": "2023-06-01",
+        // Claude Code OAuth identity headers — the gateway requires these for
+        // subscription (sub/) routing; without them it can reject the probe and
+        // we would never detect that the limit cleared. Mirrors doctor.ts.
+        "anthropic-beta": "claude-code-20250219,oauth-2025-04-20",
+        "user-agent": "claude-cli/1.0.0",
+        "x-app": "cli",
+        Authorization: `Bearer ${oauthToken}`,
+        "x-litellm-api-key": `Bearer ${gatewayKey}`,
+      },
+      body: JSON.stringify({
+        model: probeModel,
+        max_tokens: 1,
+        temperature: 0,
+        messages: [{ role: "user", content: "just respond hi" }],
+      }),
+      signal: AbortSignal.timeout(30000),
+    });
+    if (res.status === 429) {
+      log.debug({ s: "flant", model: probeModel }, "probe: still rate limited");
+      return "rate_limited";
+    }
+    if (res.ok) {
+      log.debug({ s: "flant", model: probeModel }, "probe: capacity back");
+      return "ok";
+    }
+    log.debug({ s: "flant", model: probeModel, status: res.status }, "probe: unexpected status");
+    return "error";
+  } catch (err: any) {
+    log.debug({ s: "flant", err: err?.message }, "probe failed");
+    return "error";
+  }
+}
+
 export async function discoverFlantModels(apiKey: string): Promise<string[]> {
   const res = await fetch("https://llm-api.flant.ru/v1/models", {
     headers: { Authorization: `Bearer ${apiKey}` },
@@ -655,6 +743,9 @@ export function generateFlantConfig(models: string[], subscriptionActive = false
           explore: { model: modelSpec(fastModel, sub), thinking: "low" },
           librarian: { model: modelSpec(fastModel, sub), thinking: "medium" },
           task: { model: modelSpec(taskModel, sub), thinking: "medium" },
+          advisor: { model: modelSpec(latestGpt ?? fallback, sub), thinking: "high" },
+          "deep-debugger": { model: modelSpec(latestGpt ?? fallback, sub), thinking: "high" },
+          reviewer: { model: modelSpec(latestGpt ?? fallback, sub), thinking: "high" },
         },
         presetGroups: {
           planners: buildPresetGroup({

package/extensions/orchestrator/index.ts

@@ -65,7 +65,7 @@ function registerSubagentTools(pi: ExtensionAPI): void {
             ...event.content,
             {
               type: "text" as const,
-              text: `\n\n<validation-error>\nPlan structure is invalid:\n${result.errors.map((e) => `- ${e}`).join("\n")}\n\nFix immediately. Required structure:\n# Plan\n## Scope\n<2-4 lines>\n## Checklist\n- [ ] <outcome> — Done when: <observable condition>\n## Blockers (optional)\n<issues>\n\nRewrite the file now.\n</validation-error>`,
+              text: `\n\n<validation-error>\nPlan structure is invalid:\n${result.errors.map((e) => `- ${e}`).join("\n")}\n\nFix immediately. Required structure:\n# Plan\n## Scope\n<2-4 lines>\n## Checklist\n- [ ] <outcome> — Done when: <observable condition>\n## Pattern constraints (optional; include when adding a type/function/user-facing value)\n<closest existing analog + conventions to mirror>\n## Blockers (optional)\n<issues>\n\nRewrite the file now.\n</validation-error>`,
             },
           ],
         };