@zephyrdeng/pi-review 0.7.0 → 0.8.0

package/extensions/rv-interactive.test.ts

@@ -0,0 +1,562 @@
+import assert from "node:assert/strict";
+import { test } from "vitest";
+import type { ModelInfo } from "./rv-completions.js";
+import {
+  runRvInteractiveWizard,
+  shouldRunInteractiveWizard,
+  stripInteractiveToken,
+  type InteractiveUi,
+} from "./rv-interactive.js";
+import type { RvParsed } from "./rv-prompts.js";
+
+const models: ModelInfo[] = [
+  {
+    provider: "openai-codex",
+    id: "gpt-5.6-sol",
+    label: "openai-codex/gpt-5.6-sol",
+    name: "sol",
+    reasoning: true,
+    contextWindow: 200000,
+    thinkingLevels: ["high", "xhigh"],
+  },
+  {
+    provider: "anthropic",
+    id: "claude-sonnet-4-5",
+    label: "anthropic/claude-sonnet-4-5",
+    name: "sonnet",
+    reasoning: true,
+    contextWindow: 200000,
+    thinkingLevels: ["high", "xhigh"],
+  },
+];
+
+// 12 models across 3 providers (5/4/3) — above the MODEL_PICKER_THRESHOLD of 8,
+// so the wizard offers provider browse / search / ranked list / skip.
+// code-profile presets match claude-sonnet-4-5 and gpt-5.6-luna (ranked first).
+const manyModels: ModelInfo[] = [
+  // anthropic (5)
+  { provider: "anthropic", id: "claude-sonnet-4-5", label: "anthropic/claude-sonnet-4-5", name: "sonnet", reasoning: true, contextWindow: 200000, thinkingLevels: ["high", "xhigh"] },
+  { provider: "anthropic", id: "claude-opus-4", label: "anthropic/claude-opus-4", name: "opus", reasoning: true, contextWindow: 200000, thinkingLevels: ["high", "xhigh"] },
+  { provider: "anthropic", id: "claude-haiku-5", label: "anthropic/claude-haiku-5", name: "haiku", reasoning: false, contextWindow: 200000, thinkingLevels: [] },
+  { provider: "anthropic", id: "claude-fable-5", label: "anthropic/claude-fable-5", name: "fable", reasoning: true, contextWindow: 200000, thinkingLevels: ["high", "xhigh"] },
+  { provider: "anthropic", id: "claude-solomon", label: "anthropic/claude-solomon", name: "solomon", reasoning: true, contextWindow: 200000, thinkingLevels: ["high", "xhigh"] },
+  // openai (4)
+  { provider: "openai", id: "gpt-5.6-sol", label: "openai/gpt-5.6-sol", name: "sol", reasoning: true, contextWindow: 200000, thinkingLevels: ["high", "xhigh"] },
+  { provider: "openai", id: "gpt-5.6-luna", label: "openai/gpt-5.6-luna", name: "luna", reasoning: true, contextWindow: 200000, thinkingLevels: ["high", "xhigh"] },
+  { provider: "openai", id: "gpt-5.5", label: "openai/gpt-5.5", name: "gpt55", reasoning: true, contextWindow: 200000, thinkingLevels: ["high", "xhigh"] },
+  { provider: "openai", id: "o3", label: "openai/o3", name: "o3", reasoning: true, contextWindow: 200000, thinkingLevels: ["high"] },
+  // google (3)
+  { provider: "google", id: "gemini-3-pro", label: "google/gemini-3-pro", name: "pro", reasoning: true, contextWindow: 200000, thinkingLevels: ["high", "xhigh"] },
+  { provider: "google", id: "gemini-3-flash", label: "google/gemini-3-flash", name: "flash", reasoning: true, contextWindow: 200000, thinkingLevels: ["high", "medium"] },
+  { provider: "google", id: "gemma-3", label: "google/gemma-3", name: "gemma", reasoning: false, contextWindow: 200000, thinkingLevels: [] },
+];
+
+function seed(over: Partial<RvParsed> = {}): RvParsed {
+  return {
+    strategy: "loop",
+    mode: "code",
+    keepSession: false,
+    noStream: false,
+    modelsOnly: false,
+    target: "",
+    ...over,
+  };
+}
+
+function scriptedUi(answers: {
+  selects?: string[];
+  inputs?: string[];
+  confirms?: boolean[];
+  customModelPickerResults?: (string | "__skip__" | undefined)[];
+}): InteractiveUi & { log: string[] } {
+  const selects = [...(answers.selects ?? [])];
+  const inputs = [...(answers.inputs ?? [])];
+  const confirms = [...(answers.confirms ?? [])];
+  const pickerResults = [...(answers.customModelPickerResults ?? [])];
+  const log: string[] = [];
+  return {
+    log,
+    async select(title, options) {
+      log.push(`select:${title} :: ${options.slice(0, 3).join(" | ")}`);
+      return selects.shift() ?? options[0];
+    },
+    async input(title) {
+      log.push(`input:${title}`);
+      return inputs.shift() ?? "@src";
+    },
+    async confirm(title, message) {
+      log.push(`confirm:${title}`);
+      log.push(message.split("\n")[0] ?? "");
+      return confirms.shift() ?? true;
+    },
+    notify(message) {
+      log.push(`notify:${message}`);
+    },
+    ...(answers.customModelPickerResults
+      ? {
+          async customModelPicker(input: { title: string; ranked: unknown[]; allowSkip: boolean }) {
+            log.push(`customModelPicker:${input.title} :: ranked=${input.ranked.length} skip=${input.allowSkip}`);
+            return pickerResults.shift();
+          },
+        }
+      : {}),
+  };
+}
+
+test("shouldRunInteractiveWizard triggers on empty or --interactive", () => {
+  assert.equal(shouldRunInteractiveWizard("", seed()), true);
+  assert.equal(shouldRunInteractiveWizard("interactive", seed()), true);
+  assert.equal(shouldRunInteractiveWizard("--interactive @src", seed({ target: "@src" })), true);
+  assert.equal(shouldRunInteractiveWizard("-i @src", seed({ target: "@src" })), true);
+  assert.equal(shouldRunInteractiveWizard("@src -i", seed({ target: "@src" })), true);
+  assert.equal(shouldRunInteractiveWizard("--reviewers 3 @src", seed({ target: "@src", reviewers: 3 })), false);
+});
+
+test("stripInteractiveToken removes the trigger token", () => {
+  assert.equal(stripInteractiveToken("interactive @src"), "@src");
+  assert.equal(stripInteractiveToken("--interactive"), "");
+  assert.equal(stripInteractiveToken("@src --interactive"), "@src");
+  assert.equal(stripInteractiveToken("-i @src"), "@src");
+  assert.equal(stripInteractiveToken("@src -i"), "@src");
+  assert.equal(stripInteractiveToken("--reviewers 3 -i @src"), "--reviewers 3 @src");
+});
+
+test("wizard chooses mode before asking for a missing natural-language target", async () => {
+  const ui = scriptedUi({
+    selects: ["plan review (plan)"],
+    inputs: ["review the auth design"],
+    confirms: [true],
+  });
+
+  const result = await runRvInteractiveWizard(ui, {
+    strategy: "panel",
+    seed: seed({ strategy: "panel", reviewers: 1, model: "openai-codex/gpt-5.6-sol" }),
+    models,
+    locale: "en",
+  });
+
+  assert.ok(result);
+  assert.equal(result!.mode, "plan");
+  assert.equal(result!.target, "review the auth design");
+  const modeStep = ui.log.findIndex((entry) => entry === "select:Review mode :: code review (code) | plan review (plan) | challenge review (challenge)");
+  const targetStep = ui.log.findIndex((entry) => entry === "input:Review target (natural language or @path)");
+  assert.ok(modeStep >= 0 && targetStep >= 0 && modeStep < targetStep, ui.log.join("\n"));
+});
+
+test("wizard assigns per-reviewer models through select dialogs", async () => {
+  const ui = scriptedUi({
+    inputs: ["@src"],
+    selects: [
+      "code review (code)",
+      "Single gate (fix then re-run /rv-loop) · recommended",
+      "1 · host fix point (recommended)",
+      "Custom reviewer count 2–8 (r1..rN, pick models)",
+      "3",
+      "Pick separately for each reviewer",
+      "openai-codex/gpt-5.6-sol",
+      "xhigh",
+      "anthropic/claude-sonnet-4-5",
+      "high",
+      "openai-codex/gpt-5.6-sol",
+      "xhigh",
+      "quorum · at least min-agree agree (default)",
+      "2",
+    ],
+    confirms: [true, true],
+  });
+
+  const result = await runRvInteractiveWizard(ui, {
+    strategy: "loop",
+    seed: seed(),
+    models,
+    locale: "en",
+  });
+
+  assert.ok(result);
+  assert.equal(result!.target, "@src");
+  assert.equal(result!.reviewers, 3);
+  assert.equal(result!.maxRounds, 1); // wizard always asks; first option kept
+  assert.equal(result!.until, undefined);
+  assert.equal(result!.consensus, "quorum");
+  assert.equal(result!.minAgree, 2);
+  assert.deepEqual(result!.reviewerModels, [
+    "r1=openai-codex/gpt-5.6-sol:xhigh",
+    "r2=anthropic/claude-sonnet-4-5:high",
+    "r3=openai-codex/gpt-5.6-sol:xhigh",
+  ]);
+});
+
+test("wizard keeps an explicit single-reviewer non-panel choice", async () => {
+  const ui = scriptedUi({
+    inputs: ["@src"],
+    selects: [
+      "code review (code)",
+      "Single reviewer non-panel (no consensus; shell single review)",
+      "Skip (Pi default)",
+    ],
+    confirms: [true],
+  });
+
+  const result = await runRvInteractiveWizard(ui, {
+    strategy: "panel",
+    seed: seed({ strategy: "panel" }),
+    models,
+    locale: "en",
+  });
+
+  assert.ok(result);
+  assert.equal(result!.reviewers, 1);
+  assert.equal(result!.panel, undefined);
+  assert.equal(result!.consensus, undefined);
+});
+
+test("wizard can use code-experts preset and shared model", async () => {
+  const ui = scriptedUi({
+    inputs: ["review auth"],
+    selects: [
+      "code review (code)",
+      "Single gate (fix then re-run /rv-loop) · recommended",
+      "1 · host fix point (recommended)",
+      "Preset code-experts (correctness/security/testing)",
+      "Same model for all reviewers",
+      "openai-codex/gpt-5.6-sol",
+      "xhigh",
+      "quorum · at least min-agree agree (default)",
+      "2",
+    ],
+    confirms: [true, true],
+  });
+
+  const result = await runRvInteractiveWizard(ui, {
+    strategy: "loop",
+    seed: seed(),
+    models,
+    locale: "en",
+  });
+
+  assert.ok(result);
+  assert.equal(result!.panel, "code-experts");
+  assert.deepEqual(result!.reviewerModels, [
+    "correctness=openai-codex/gpt-5.6-sol:xhigh",
+    "security=openai-codex/gpt-5.6-sol:xhigh",
+    "testing=openai-codex/gpt-5.6-sol:xhigh",
+  ]);
+});
+
+test("wizard can select until-clean with a hard budget", async () => {
+  const ui = scriptedUi({
+    inputs: ["@src"],
+    selects: [
+      "code review (code)",
+      "until clean · review→fix→re-review until clean (hard budget)",
+      "10 · default",
+      "Preset code-experts (correctness/security/testing)",
+      "Same model for all reviewers",
+      "openai-codex/gpt-5.6-sol",
+      "xhigh",
+      "quorum · at least min-agree agree (default)",
+      "2",
+    ],
+    confirms: [true, true],
+  });
+
+  const result = await runRvInteractiveWizard(ui, {
+    strategy: "loop",
+    seed: seed(),
+    models,
+    locale: "en",
+  });
+
+  assert.ok(result);
+  assert.equal(result!.until, "clean");
+  assert.equal(result!.maxRounds, 10);
+  assert.equal(result!.panel, "code-experts");
+});
+
+// --- Model picker with a large catalog (> MODEL_PICKER_THRESHOLD) ---
+// These exercise the provider browse / search / ranked-list / skip chooser that
+// replaces the old flat `labels.slice(0, 40)` list. With <= 8 models the wizard
+// still uses the flat list (covered by the tests above with the 2-model fixture).
+
+test("wizard model picker browses by provider for a large catalog", async () => {
+  const ui = scriptedUi({
+    inputs: ["@src"],
+    selects: [
+      "code review (code)",
+      "Single reviewer non-panel (no consensus; shell single review)",
+      "Browse by provider (arrows switch provider, enter drills in)",
+      "▸ anthropic (5)",
+      "anthropic/claude-sonnet-4-5",
+      "xhigh",
+    ],
+    confirms: [true],
+  });
+
+  const result = await runRvInteractiveWizard(ui, {
+    strategy: "panel",
+    seed: seed({ strategy: "panel" }),
+    models: manyModels,
+    locale: "en",
+  });
+
+  assert.ok(result);
+  assert.equal(result!.reviewers, 1);
+  assert.equal(result!.panel, undefined);
+  assert.equal(result!.model, "anthropic/claude-sonnet-4-5");
+  assert.equal(result!.thinking, "xhigh");
+  // The provider picker was actually visited.
+  assert.ok(ui.log.some((e) => e.startsWith("select:Model · pick provider")), ui.log.join("\n"));
+});
+
+test("wizard model picker can switch providers via the back entry", async () => {
+  const ui = scriptedUi({
+    inputs: ["@src"],
+    selects: [
+      "code review (code)",
+      "Single reviewer non-panel (no consensus; shell single review)",
+      "Browse by provider (arrows switch provider, enter drills in)",
+      "▸ anthropic (5)",
+      "← Switch provider",
+      "  openai (4)",
+      "openai/gpt-5.6-luna",
+      "xhigh",
+    ],
+    confirms: [true],
+  });
+
+  const result = await runRvInteractiveWizard(ui, {
+    strategy: "panel",
+    seed: seed({ strategy: "panel" }),
+    models: manyModels,
+    locale: "en",
+  });
+
+  assert.ok(result);
+  assert.equal(result!.model, "openai/gpt-5.6-luna");
+  assert.equal(result!.thinking, "xhigh");
+  // Provider list was shown twice (initial + after switching back).
+  const providerPicks = ui.log.filter((e) => e.startsWith("select:Model · pick provider"));
+  assert.equal(providerPicks.length, 2, ui.log.join("\n"));
+});
+
+test("wizard model picker searches by keyword", async () => {
+  const ui = scriptedUi({
+    inputs: ["@src", "claude"],
+    selects: [
+      "code review (code)",
+      "Single reviewer non-panel (no consensus; shell single review)",
+      "Search models (type a provider or model keyword)",
+      "anthropic/claude-opus-4",
+      "high",
+    ],
+    confirms: [true],
+  });
+
+  const result = await runRvInteractiveWizard(ui, {
+    strategy: "panel",
+    seed: seed({ strategy: "panel" }),
+    models: manyModels,
+    locale: "en",
+  });
+
+  assert.ok(result);
+  assert.equal(result!.model, "anthropic/claude-opus-4");
+  assert.equal(result!.thinking, "high");
+  assert.ok(ui.log.some((e) => e.startsWith("input:Model · search")), ui.log.join("\n"));
+  assert.ok(ui.log.some((e) => e.startsWith("select:Model · results (5)")), ui.log.join("\n"));
+});
+
+test("wizard model picker re-prompts when a search has no matches", async () => {
+  const ui = scriptedUi({
+    inputs: ["@src", "zzz", "claude"],
+    selects: [
+      "code review (code)",
+      "Single reviewer non-panel (no consensus; shell single review)",
+      "Search models (type a provider or model keyword)",
+      "anthropic/claude-sonnet-4-5",
+      "xhigh",
+    ],
+    confirms: [true],
+  });
+
+  const result = await runRvInteractiveWizard(ui, {
+    strategy: "panel",
+    seed: seed({ strategy: "panel" }),
+    models: manyModels,
+    locale: "en",
+  });
+
+  assert.ok(result);
+  assert.equal(result!.model, "anthropic/claude-sonnet-4-5");
+  assert.ok(ui.log.some((e) => e.includes(`No models match "zzz"`)), ui.log.join("\n"));
+  // Search input was called twice (first miss + retry).
+  const searchInputs = ui.log.filter((e) => e.startsWith("input:Model · search"));
+  assert.equal(searchInputs.length, 2, ui.log.join("\n"));
+});
+
+test("wizard model picker can use the full ranked list", async () => {
+  const ui = scriptedUi({
+    inputs: ["@src"],
+    selects: [
+      "code review (code)",
+      "Single reviewer non-panel (no consensus; shell single review)",
+      "Ranked list (all models)",
+      "openai/gpt-5.6-luna",
+      "xhigh",
+    ],
+    confirms: [true],
+  });
+
+  const result = await runRvInteractiveWizard(ui, {
+    strategy: "panel",
+    seed: seed({ strategy: "panel" }),
+    models: manyModels,
+    locale: "en",
+  });
+
+  assert.ok(result);
+  assert.equal(result!.model, "openai/gpt-5.6-luna");
+  assert.equal(result!.thinking, "xhigh");
+});
+
+test("wizard model picker supports skip for a large catalog", async () => {
+  const ui = scriptedUi({
+    inputs: ["@src"],
+    selects: [
+      "code review (code)",
+      "Single reviewer non-panel (no consensus; shell single review)",
+      "Skip (Pi default)",
+    ],
+    confirms: [true],
+  });
+
+  const result = await runRvInteractiveWizard(ui, {
+    strategy: "panel",
+    seed: seed({ strategy: "panel" }),
+    models: manyModels,
+    locale: "en",
+  });
+
+  assert.ok(result);
+  assert.equal(result!.model, undefined);
+  assert.equal(result!.thinking, undefined);
+});
+
+// --- Inline searchable model picker (custom component, TUI path) ---
+// When the host provides `customModelPicker`, the wizard uses the inline
+// searchable picker for every catalog size (it supersedes the select-based
+// flow). These tests stub `customModelPicker` and verify the wizard maps its
+// result to model / skip / cancel correctly.
+
+test("inline picker: chosen label sets model + thinking", async () => {
+  const ui = scriptedUi({
+    inputs: ["@src"],
+    selects: [
+      "code review (code)",
+      "Single reviewer non-panel (no consensus; shell single review)",
+      // thinking select still happens after the picker resolves to a label
+      "xhigh",
+    ],
+    confirms: [true],
+    customModelPickerResults: ["anthropic/claude-sonnet-4-5"],
+  });
+
+  const result = await runRvInteractiveWizard(ui, {
+    strategy: "panel",
+    seed: seed({ strategy: "panel" }),
+    models: manyModels,
+    locale: "en",
+  });
+
+  assert.ok(result);
+  assert.equal(result!.model, "anthropic/claude-sonnet-4-5");
+  assert.equal(result!.thinking, "xhigh");
+  assert.ok(ui.log.some((e) => e.startsWith("customModelPicker:Model ")), ui.log.join("\n"));
+});
+
+test("inline picker: skip sentinel leaves model unset", async () => {
+  const ui = scriptedUi({
+    inputs: ["@src"],
+    selects: [
+      "code review (code)",
+      "Single reviewer non-panel (no consensus; shell single review)",
+    ],
+    confirms: [true],
+    customModelPickerResults: ["__skip__"],
+  });
+
+  const result = await runRvInteractiveWizard(ui, {
+    strategy: "panel",
+    seed: seed({ strategy: "panel" }),
+    models: manyModels,
+    locale: "en",
+  });
+
+  assert.ok(result);
+  assert.equal(result!.model, undefined);
+  assert.equal(result!.thinking, undefined);
+});
+
+test("inline picker: cancel (undefined) cancels the whole wizard", async () => {
+  const ui = scriptedUi({
+    inputs: ["@src"],
+    selects: [
+      "code review (code)",
+      "Single reviewer non-panel (no consensus; shell single review)",
+    ],
+    confirms: [true],
+    customModelPickerResults: [undefined],
+  });
+
+  const result = await runRvInteractiveWizard(ui, {
+    strategy: "panel",
+    seed: seed({ strategy: "panel" }),
+    models: manyModels,
+    locale: "en",
+  });
+
+  assert.equal(result, undefined);
+});
+
+test("inline picker: per-reviewer models resolve through repeated picker calls", async () => {
+  const uiPanel = scriptedUi({
+    inputs: ["@src"],
+    selects: [
+      "code review (code)",
+      "Single gate (fix then re-run /rv-loop) · recommended",
+      "1 · host fix point (recommended)",
+      "Custom reviewer count 2–8 (r1..rN, pick models)",
+      "3",
+      "Pick separately for each reviewer",
+      // thinking selects for r1, r2, r3
+      "xhigh",
+      "high",
+      "xhigh",
+      "quorum · at least min-agree agree (default)",
+      "2",
+    ],
+    confirms: [true, true],
+    customModelPickerResults: [
+      "openai/gpt-5.6-luna",
+      "anthropic/claude-opus-4",
+      "google/gemini-3-pro",
+    ],
+  });
+
+  const result = await runRvInteractiveWizard(uiPanel, {
+    strategy: "loop",
+    seed: seed(),
+    models: manyModels,
+    locale: "en",
+  });
+
+  assert.ok(result);
+  assert.deepEqual(result!.reviewerModels, [
+    "r1=openai/gpt-5.6-luna:xhigh",
+    "r2=anthropic/claude-opus-4:high",
+    "r3=google/gemini-3-pro:xhigh",
+  ]);
+  // The picker was invoked once per reviewer.
+  const pickerCalls = uiPanel.log.filter((e) => e.startsWith("customModelPicker:"));
+  assert.equal(pickerCalls.length, 3, uiPanel.log.join("\n"));
+});