@ilya-lesikov/pi-pi 0.4.0 → 0.6.0

package/extensions/orchestrator/usage-tracker.test.ts

@@ -0,0 +1,435 @@
+import { mkdirSync, mkdtempSync, readFileSync, rmSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { afterEach, describe, expect, it } from "vitest";
+import { createUsageTracker, dumpUsageSummary, loadUsageSummary } from "./usage-tracker.js";
+
+const tempDirs: string[] = [];
+
+function makeTempDir(): string {
+  const dir = mkdtempSync(join(tmpdir(), "pi-pi-usage-tracker-"));
+  tempDirs.push(dir);
+  return dir;
+}
+
+afterEach(() => {
+  delete process.env.PI_CODING_AGENT_DIR;
+  for (const dir of tempDirs.splice(0)) {
+    rmSync(dir, { recursive: true, force: true });
+  }
+});
+
+describe("usage-tracker", () => {
+  it("createUsageTracker starts with zero counts", () => {
+    const tracker = createUsageTracker();
+
+    expect(tracker.getMainInputTokens()).toBe(0);
+    expect(tracker.getMainOutputTokens()).toBe(0);
+    expect(tracker.getMainCacheReadTokens()).toBe(0);
+    expect(tracker.getMainCacheWriteTokens()).toBe(0);
+    expect(tracker.getMainCost()).toBe(0);
+    expect(tracker.getTotalInputTokens()).toBe(0);
+    expect(tracker.getTotalOutputTokens()).toBe(0);
+    expect(tracker.getTotalCost()).toBe(0);
+  });
+
+  it("loadFromSummary is idempotent and does not double-count subagents", () => {
+    const tracker = createUsageTracker();
+    const summary = {
+      subagents: [
+        { description: "a", agentType: "Explore", modelId: "m", inputTokens: 100, outputTokens: 50, cost: 0.5, toolUses: 1, durationMs: 10 },
+      ],
+    };
+
+    tracker.loadFromSummary(summary);
+    tracker.loadFromSummary(summary);
+
+    expect(tracker.getSubagentList()).toHaveLength(1);
+    const totals = tracker.getSubagentTotals();
+    expect(totals.inputTokens).toBe(100);
+    expect(totals.outputTokens).toBe(50);
+    expect(totals.cost).toBeCloseTo(0.5);
+  });
+
+  it("recordTurn accumulates input output and cache tokens", () => {
+    const tracker = createUsageTracker();
+
+    tracker.recordTurn("openai/gpt-5", "openai", 10, 20, 3, 4, 0.1, true);
+    tracker.recordTurn("openai/gpt-5", "openai", 5, 7, 2, 1, 0.2, true);
+
+    expect(tracker.getMainInputTokens()).toBe(15);
+    expect(tracker.getMainOutputTokens()).toBe(27);
+    expect(tracker.getMainCacheReadTokens()).toBe(5);
+    expect(tracker.getMainCacheWriteTokens()).toBe(5);
+  });
+
+  it("recordTurn accumulates cost", () => {
+    const tracker = createUsageTracker();
+
+    tracker.recordTurn("openai/gpt-5", "openai", 1, 1, 0, 0, 0.25, false);
+    tracker.recordTurn("openai/gpt-5", "openai", 1, 1, 0, 0, 0.5, false);
+
+    expect(tracker.getMainCost()).toBe(0.75);
+  });
+
+  it("recordTurn handles non-finite numeric input as zero", () => {
+    const tracker = createUsageTracker();
+
+    tracker.recordTurn("openai/gpt-5", "openai", Number.NaN, Number.POSITIVE_INFINITY, 0, 0, Number.NaN, false);
+
+    expect(tracker.getMainInputTokens()).toBe(0);
+    expect(tracker.getMainOutputTokens()).toBe(0);
+    expect(tracker.getMainCost()).toBe(0);
+  });
+
+  it("recordTurn tracks per-model usage", () => {
+    const tracker = createUsageTracker();
+
+    tracker.recordTurn("openai/gpt-5", "openai", 10, 5, 1, 0, 0.2, false);
+    tracker.recordTurn("anthropic/claude-opus-4-6", "anthropic", 7, 3, 0, 0, 0.3, false);
+
+    expect(tracker.getPerModelUsage()).toEqual({
+      "openai/gpt-5": {
+        inputTokens: 10,
+        outputTokens: 5,
+        cacheReadTokens: 1,
+        cacheWriteTokens: 0,
+        cacheSupported: false,
+        turns: 1,
+      },
+      "anthropic/claude-opus-4-6": {
+        inputTokens: 7,
+        outputTokens: 3,
+        cacheReadTokens: 0,
+        cacheWriteTokens: 0,
+        cacheSupported: false,
+        turns: 1,
+      },
+    });
+  });
+
+  it("recordTurn cacheSupported sets model cache support", () => {
+    const tracker = createUsageTracker();
+
+    tracker.recordTurn("openai/gpt-5", "openai", 1, 1, 0, 0, 0, false);
+    tracker.recordTurn("openai/gpt-5", "openai", 1, 1, 0, 0, 0, true);
+
+    expect(tracker.getPerModelUsage()["openai/gpt-5"]?.cacheSupported).toBe(true);
+  });
+
+  it("recordTurn uses unknown-model key when model id is empty", () => {
+    const tracker = createUsageTracker();
+
+    tracker.recordTurn("", "openai", 3, 2, 0, 0, 0.1, false);
+
+    expect(tracker.getPerModelUsage()["unknown-model"]).toMatchObject({
+      inputTokens: 3,
+      outputTokens: 2,
+      turns: 1,
+    });
+  });
+
+  it("recordSubagentCompletion accumulates subagent tokens", () => {
+    const tracker = createUsageTracker();
+
+    tracker.recordSubagentCompletion({ input: 11, output: 4 } as any, 0.15);
+    tracker.recordSubagentCompletion({ input: 9, output: 6 }, 0.25);
+
+    expect(tracker.getSubagentTotals()).toEqual({
+      inputTokens: 20,
+      outputTokens: 10,
+      cost: 0.4,
+    });
+  });
+
+  it("recordSubagentCompletion defaults metadata fields when missing", () => {
+    const tracker = createUsageTracker();
+
+    tracker.recordSubagentCompletion({ input: 1, output: 2 } as any);
+
+    expect(tracker.getSubagentList()[0]).toMatchObject({
+      description: "unknown",
+      agentType: "unknown",
+      modelId: "unknown",
+      durationMs: 0,
+      toolUses: 0,
+    });
+  });
+
+  it("recordSubagentCompletion uses tokens.cost over fallback cost", () => {
+    const tracker = createUsageTracker();
+
+    tracker.recordSubagentCompletion({ input: 1, output: 1, cost: 0.6 } as any, 0.2);
+
+    expect(tracker.getSubagentTotals().cost).toBe(0.6);
+  });
+
+  it("recordSubagentCompletion stores subagent metadata", () => {
+    const tracker = createUsageTracker();
+
+    tracker.recordSubagentCompletion(
+      { input: 5, output: 2, cacheRead: 8, cacheWrite: 1 } as any,
+      0.2,
+      { description: "Planner", agentType: "planner", modelId: "openai/gpt-5", durationMs: 900, toolUses: 3 },
+    );
+
+    expect(tracker.getSubagentList()).toEqual([
+      {
+        description: "Planner",
+        agentType: "planner",
+        modelId: "openai/gpt-5",
+        inputTokens: 5,
+        outputTokens: 2,
+        cacheReadTokens: 8,
+        cacheWriteTokens: 1,
+        cacheSupported: true,
+        cost: 0.2,
+        durationMs: 900,
+        toolUses: 3,
+      },
+    ]);
+  });
+
+  it("recordSubagentCompletion uses total when input and output are zero", () => {
+    const tracker = createUsageTracker();
+
+    tracker.recordSubagentCompletion({ total: 77, input: 0, output: 0 });
+
+    expect(tracker.getSubagentTotals().inputTokens).toBe(77);
+    expect(tracker.getSubagentTotals().outputTokens).toBe(0);
+  });
+
+  it("getTotalInputTokens includes main and subagent input", () => {
+    const tracker = createUsageTracker();
+
+    tracker.recordTurn("openai/gpt-5", "openai", 10, 0, 0, 0, 0, false);
+    tracker.recordSubagentCompletion({ input: 4, output: 0 });
+
+    expect(tracker.getTotalInputTokens()).toBe(14);
+  });
+
+  it("getTotalOutputTokens includes main and subagent output", () => {
+    const tracker = createUsageTracker();
+
+    tracker.recordTurn("openai/gpt-5", "openai", 0, 7, 0, 0, 0, false);
+    tracker.recordSubagentCompletion({ input: 0, output: 3 } as any);
+
+    expect(tracker.getTotalOutputTokens()).toBe(10);
+  });
+
+  it("getTotalCost includes main and subagent cost", () => {
+    const tracker = createUsageTracker();
+
+    tracker.recordTurn("openai/gpt-5", "openai", 0, 0, 0, 0, 1.2, false);
+    tracker.recordSubagentCompletion({ input: 0, output: 0 }, 0.8);
+
+    expect(tracker.getTotalCost()).toBe(2);
+  });
+
+  it("getCacheHitRate calculates ratio", () => {
+    const tracker = createUsageTracker();
+
+    tracker.recordTurn("openai/gpt-5", "openai", 30, 0, 10, 0, 0, true);
+
+    expect(tracker.getCacheHitRate()).toBeCloseTo(10 / 40);
+  });
+
+  it("getCacheHitRate returns zero when denominator is zero", () => {
+    const tracker = createUsageTracker();
+    expect(tracker.getCacheHitRate()).toBe(0);
+  });
+
+  it("isCacheSupported true when any model or subagent has cache", () => {
+    const tracker = createUsageTracker();
+    expect(tracker.isCacheSupported()).toBe(false);
+
+    tracker.recordTurn("openai/gpt-5", "openai", 1, 1, 0, 0, 0, true);
+    expect(tracker.isCacheSupported()).toBe(true);
+
+    tracker.reset();
+    tracker.recordSubagentCompletion({ input: 1, output: 1, cacheRead: 1 } as any);
+    expect(tracker.isCacheSupported()).toBe(true);
+  });
+
+  it("getPerModelUsage returns a copy", () => {
+    const tracker = createUsageTracker();
+    tracker.recordTurn("openai/gpt-5", "openai", 2, 3, 0, 0, 0, false);
+
+    const usage = tracker.getPerModelUsage();
+    usage["openai/gpt-5"]!.inputTokens = 999;
+
+    expect(tracker.getPerModelUsage()["openai/gpt-5"]!.inputTokens).toBe(2);
+  });
+
+  it("toSummary returns complete summary", () => {
+    const tracker = createUsageTracker();
+    tracker.recordTurn("openai/gpt-5", "openai", 10, 7, 3, 2, 0.1234567, true);
+    tracker.recordSubagentCompletion({ input: 4, output: 1 } as any, 0.2, { description: "Explore", agentType: "explore", modelId: "google/gemini-3.1-pro" });
+
+    const summary = tracker.toSummary() as any;
+
+    expect(summary.startedAt).toBeTypeOf("string");
+    expect(summary.endedAt).toBeTypeOf("string");
+    expect(summary.totals).toEqual({
+      inputTokens: 10,
+      outputTokens: 7,
+      cacheReadTokens: 3,
+      cacheWriteTokens: 2,
+      cost: 0.123457,
+      turns: 1,
+    });
+    expect(summary.subagents).toHaveLength(1);
+    expect(summary.models["openai/gpt-5"]).toMatchObject({ inputTokens: 10, outputTokens: 7, turns: 1 });
+  });
+
+  it("loadFromSummary restores state", () => {
+    const tracker = createUsageTracker();
+    tracker.loadFromSummary({
+      startedAt: "2026-01-01T00:00:00.000Z",
+      totals: {
+        inputTokens: 100,
+        outputTokens: 50,
+        cacheReadTokens: 20,
+        cacheWriteTokens: 5,
+        cost: 1.5,
+        turns: 3,
+      },
+      models: {
+        "openai/gpt-5": {
+          inputTokens: 100,
+          outputTokens: 50,
+          cacheReadTokens: 20,
+          cacheWriteTokens: 5,
+          cacheSupported: true,
+          turns: 3,
+        },
+      },
+      subagents: [
+        {
+          description: "Reviewer",
+          agentType: "reviewer",
+          modelId: "anthropic/claude-opus-4-6",
+          inputTokens: 7,
+          outputTokens: 2,
+          cacheReadTokens: 0,
+          cacheWriteTokens: 0,
+          cacheSupported: false,
+          cost: 0.4,
+          durationMs: 200,
+          toolUses: 2,
+        },
+      ],
+    });
+
+    expect(tracker.getMainInputTokens()).toBe(100);
+    expect(tracker.getMainOutputTokens()).toBe(50);
+    expect(tracker.getMainCacheReadTokens()).toBe(20);
+    expect(tracker.getMainCacheWriteTokens()).toBe(5);
+    expect(tracker.getMainCost()).toBe(1.5);
+    expect(tracker.getSubagentTotals()).toEqual({ inputTokens: 7, outputTokens: 2, cost: 0.4 });
+    expect(tracker.getPerModelUsage()["openai/gpt-5"]?.turns).toBe(3);
+  });
+
+  it("loadFromSummary handles malformed summary entries", () => {
+    const tracker = createUsageTracker();
+
+    tracker.loadFromSummary({
+      startedAt: 123,
+      totals: "bad",
+      models: {
+        "openai/gpt-5": {
+          inputTokens: "x",
+          outputTokens: 2,
+          cacheSupported: "yes",
+          turns: "z",
+        },
+      },
+      subagents: [{ inputTokens: "bad", outputTokens: 4 }],
+    } as any);
+
+    expect(tracker.getMainInputTokens()).toBe(0);
+    expect(tracker.getMainOutputTokens()).toBe(0);
+    expect(tracker.getPerModelUsage()["openai/gpt-5"]).toMatchObject({
+      inputTokens: 0,
+      outputTokens: 2,
+      cacheSupported: false,
+      turns: 0,
+    });
+    expect(tracker.getSubagentList()[0]).toMatchObject({
+      inputTokens: 0,
+      outputTokens: 4,
+    });
+  });
+
+  it("loadFromSummary and toSummary preserve data round-trip", () => {
+    const source = createUsageTracker();
+    source.recordTurn("openai/gpt-5", "openai", 13, 8, 2, 1, 0.5, true);
+    source.recordSubagentCompletion({ input: 3, output: 1, cacheRead: 1 } as any, 0.1, {
+      description: "Librarian",
+      agentType: "librarian",
+      modelId: "google/gemini-3.1-pro",
+      durationMs: 321,
+      toolUses: 4,
+    });
+
+    const mid = source.toSummary() as Record<string, unknown>;
+    const restored = createUsageTracker();
+    restored.loadFromSummary(mid);
+    const roundTrip = restored.toSummary() as any;
+
+    expect(roundTrip.totals).toEqual((mid as any).totals);
+    expect(roundTrip.models).toEqual((mid as any).models);
+    expect(roundTrip.subagents).toEqual((mid as any).subagents);
+  });
+
+  it("reset clears all state", () => {
+    const tracker = createUsageTracker();
+    tracker.recordTurn("openai/gpt-5", "openai", 1, 2, 3, 4, 0.2, true);
+    tracker.recordSubagentCompletion({ input: 5, output: 6 } as any, 0.1);
+
+    tracker.reset();
+
+    expect(tracker.getMainInputTokens()).toBe(0);
+    expect(tracker.getMainOutputTokens()).toBe(0);
+    expect(tracker.getMainCacheReadTokens()).toBe(0);
+    expect(tracker.getMainCacheWriteTokens()).toBe(0);
+    expect(tracker.getMainCost()).toBe(0);
+    expect(tracker.getSubagentTotals()).toEqual({ inputTokens: 0, outputTokens: 0, cost: 0 });
+    expect(tracker.getPerModelUsage()).toEqual({});
+    expect(tracker.getSubagentList()).toEqual([]);
+  });
+
+  it("dumpUsageSummary and loadUsageSummary write and read JSON", () => {
+    const dir = makeTempDir();
+    process.env.PI_CODING_AGENT_DIR = dir;
+    const tracker = createUsageTracker();
+    tracker.recordTurn("openai/gpt-5", "openai", 11, 12, 1, 0, 0.9, true);
+
+    dumpUsageSummary(tracker, "session-1");
+    const loaded = loadUsageSummary("session-1") as any;
+
+    expect(loaded.sessionId).toBe("session-1");
+    expect(loaded.totals.inputTokens).toBe(11);
+    expect(loaded.totals.outputTokens).toBe(12);
+    expect(loaded.models["openai/gpt-5"].turns).toBe(1);
+
+    const filePath = join(dir, "extensions", "pp", "usage", "session-1.json");
+    const raw = readFileSync(filePath, "utf-8");
+    expect(raw.endsWith("\n")).toBe(true);
+  });
+
+  it("loadUsageSummary returns null for missing and invalid files", () => {
+    const dir = makeTempDir();
+    process.env.PI_CODING_AGENT_DIR = dir;
+
+    expect(loadUsageSummary("missing")).toBeNull();
+
+    const usageDir = join(dir, "extensions", "pp", "usage");
+    const invalidPath = join(usageDir, "broken.json");
+    rmSync(usageDir, { recursive: true, force: true });
+    mkdirSync(usageDir, { recursive: true });
+    writeFileSync(invalidPath, "{broken", "utf-8");
+
+    expect(loadUsageSummary("broken")).toBeNull();
+  });
+});

package/extensions/orchestrator/usage-tracker.ts

@@ -7,11 +7,12 @@ export interface ModelUsage {
   outputTokens: number;
   cacheReadTokens: number;
   cacheWriteTokens: number;
+  cacheSupported: boolean;
   turns: number;
 }
 
 export interface UsageTracker {
-  recordTurn(modelId: string, provider: string, input: number, output: number, cacheRead: number, cacheWrite: number, cost: number): void;
+  recordTurn(modelId: string, provider: string, input: number, output: number, cacheRead: number, cacheWrite: number, cost: number, cacheSupported?: boolean): void;
   recordSubagentCompletion(tokens: { input?: number; output?: number; total?: number }, cost?: number, meta?: { description?: string; agentType?: string; modelId?: string; durationMs?: number; toolUses?: number }): void;
   loadFromSummary(summary: Record<string, unknown>): void;
   getTotalInputTokens(): number;
@@ -25,6 +26,7 @@ export interface UsageTracker {
   getMainCacheWriteTokens(): number;
   getMainCost(): number;
   getCacheHitRate(): number;
+  isCacheSupported(): boolean;
   getPerModelUsage(): Record<string, ModelUsage>;
   getSubagentTotals(): { inputTokens: number; outputTokens: number; cost: number };
   getSubagentList(): SubagentUsage[];
@@ -40,6 +42,7 @@ export interface SubagentUsage {
   outputTokens: number;
   cacheReadTokens: number;
   cacheWriteTokens: number;
+  cacheSupported: boolean;
   cost: number;
   durationMs: number;
   toolUses: number;
@@ -96,7 +99,7 @@ export function createUsageTracker(): UsageTracker {
   const state = createInitialState();
 
   return {
-    recordTurn(modelId: string, _provider: string, input: number, output: number, cacheRead: number, cacheWrite: number, cost: number): void {
+    recordTurn(modelId: string, _provider: string, input: number, output: number, cacheRead: number, cacheWrite: number, cost: number, cacheSupported?: boolean): void {
       const safeInput = toFiniteNumber(input);
       const safeOutput = toFiniteNumber(output);
       const safeCacheRead = toFiniteNumber(cacheRead);
@@ -116,6 +119,7 @@ export function createUsageTracker(): UsageTracker {
         outputTokens: 0,
         cacheReadTokens: 0,
         cacheWriteTokens: 0,
+        cacheSupported: false,
         turns: 0,
       };
 
@@ -123,6 +127,7 @@ export function createUsageTracker(): UsageTracker {
       usage.outputTokens += safeOutput;
       usage.cacheReadTokens += safeCacheRead;
       usage.cacheWriteTokens += safeCacheWrite;
+      if (cacheSupported) usage.cacheSupported = true;
       usage.turns += 1;
       state.models.set(key, usage);
     },
@@ -145,6 +150,8 @@ export function createUsageTracker(): UsageTracker {
       state.subagentOutputTokens += safeOutput;
       state.subagentCost += safeCost;
 
+      const cacheSupported = typeof tokens.cacheRead === "number" || typeof tokens.cacheWrite === "number";
+
       state.subagents.push({
         description: meta?.description ?? "unknown",
         agentType: meta?.agentType ?? "unknown",
@@ -153,6 +160,7 @@ export function createUsageTracker(): UsageTracker {
         outputTokens: safeOutput,
         cacheReadTokens: safeCacheRead,
         cacheWriteTokens: safeCacheWrite,
+        cacheSupported,
         cost: safeCost,
         durationMs: toFiniteNumber(meta?.durationMs),
         toolUses: toFiniteNumber(meta?.toolUses),
@@ -172,6 +180,10 @@ export function createUsageTracker(): UsageTracker {
         state.totalTurns = toFiniteNumber(totals.turns);
       }
       if (Array.isArray(summary.subagents)) {
+        state.subagents = [];
+        state.subagentInputTokens = 0;
+        state.subagentOutputTokens = 0;
+        state.subagentCost = 0;
         for (const sa of summary.subagents as Record<string, unknown>[]) {
           const entry: SubagentUsage = {
             description: typeof sa.description === "string" ? sa.description : "unknown",
@@ -181,6 +193,7 @@ export function createUsageTracker(): UsageTracker {
             outputTokens: toFiniteNumber(sa.outputTokens),
             cacheReadTokens: toFiniteNumber(sa.cacheReadTokens),
             cacheWriteTokens: toFiniteNumber(sa.cacheWriteTokens),
+            cacheSupported: sa.cacheSupported === true,
             cost: toFiniteNumber(sa.cost),
             durationMs: toFiniteNumber(sa.durationMs),
             toolUses: toFiniteNumber(sa.toolUses),
@@ -198,6 +211,7 @@ export function createUsageTracker(): UsageTracker {
             outputTokens: toFiniteNumber(usage.outputTokens),
             cacheReadTokens: toFiniteNumber(usage.cacheReadTokens),
             cacheWriteTokens: toFiniteNumber(usage.cacheWriteTokens),
+            cacheSupported: (usage as any).cacheSupported === true,
             turns: toFiniteNumber(usage.turns),
           });
         }
@@ -255,6 +269,13 @@ export function createUsageTracker(): UsageTracker {
       return totalCacheRead / denominator;
     },
 
+    isCacheSupported(): boolean {
+      for (const usage of state.models.values()) {
+        if (usage.cacheSupported) return true;
+      }
+      return state.subagents.some((sa) => sa.cacheSupported);
+    },
+
     getPerModelUsage(): Record<string, ModelUsage> {
       const out: Record<string, ModelUsage> = {};
       for (const [model, usage] of state.models.entries()) {

package/extensions/orchestrator/validate-artifacts.test.ts

@@ -1,5 +1,5 @@
 import { describe, expect, it } from "vitest";
-import { validatePlan, validateUserRequest } from "./validate-artifacts.js";
+import { validateArtifact, validatePlan, validateResearch, validateUserRequest } from "./validate-artifacts.js";
 
 describe("validateUserRequest", () => {
   it("rejects placeholder distillation and constraints content", () => {
@@ -86,3 +86,85 @@ Ship minimal fix.
     expect(result.errors).toContain("Section ## Scope is empty. Expected 2-4 lines summarizing scope and constraints.");
   });
 });
+
+describe("validateResearch", () => {
+  it("accepts valid research content", () => {
+    const content = `## Affected Code
+src/main.ts:run — entry point
+
+## Architecture Context
+- Main flow calls run and dispatches handlers
+
+## Constraints & Edge Cases
+- MUST: Keep behavior backward compatible
+- RISK: Regression in startup path
+
+## Open Questions
+Need confirmation about deprecated flag
+`;
+
+    const result = validateResearch(content);
+
+    expect(result.ok).toBe(true);
+  });
+
+  it("fails when required sections are missing", () => {
+    const content = `## Affected Code
+src/main.ts:run — entry point
+
+## Constraints & Edge Cases
+- MUST: Keep behavior backward compatible
+`;
+
+    const result = validateResearch(content);
+
+    expect(result.ok).toBe(false);
+    if (result.ok) return;
+    expect(result.errors.some((error) => error.includes("Missing required section: ## Architecture Context"))).toBe(true);
+  });
+
+  it("fails when affected code section is empty", () => {
+    const content = `## Affected Code
+...
+
+## Architecture Context
+- Main flow calls run and dispatches handlers
+
+## Constraints & Edge Cases
+- MUST: Keep behavior backward compatible
+`;
+
+    const result = validateResearch(content);
+
+    expect(result.ok).toBe(false);
+    if (result.ok) return;
+    expect(result.errors).toContain("Section ## Affected Code is empty. Expected non-empty content.");
+  });
+});
+
+describe("validateArtifact", () => {
+  it("accepts valid artifact with top-level title", () => {
+    const content = `# Risk Analysis
+
+- First risk
+- Second risk
+`;
+
+    const result = validateArtifact(content);
+
+    expect(result.ok).toBe(true);
+  });
+
+  it("fails when top-level title heading is missing", () => {
+    const content = `## Risk Analysis
+
+- First risk
+`;
+
+    const result = validateArtifact(content);
+
+    expect(result.ok).toBe(false);
+    if (result.ok) return;
+    expect(result.errors.some((error) => error.includes("Expected a top-level heading"))).toBe(true);
+  });
+});

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ilya-lesikov/pi-pi",
-  "version": "0.4.0",
+  "version": "0.6.0",
   "description": "Pi-Pi Coding Agent: based on Pi Coding Agent, but twice as good",
   "keywords": [
     "pi-package",
@@ -52,6 +52,7 @@
     "@pierre/diffs": "^1.2.0",
     "effect": "^3.0.0",
     "minimatch": "^10.0.0",
+    "pino": "^10.3.1",
     "proper-lockfile": "^4.1.2",
     "turndown": "^7.2.0",
     "typebox": "^1.0.0"