@monotykamary/pi-tps 1.3.5 → 1.3.7

package/README.md

@@ -53,6 +53,8 @@ Or install from GitHub:
 pi install https://github.com/monotykamary/pi-tps
 ```
 
+The published npm package is intentionally slim: only `extensions/pi-tps/index.ts`, `README.md`, and `LICENSE`. It has **no runtime `dependencies`** and does **not** ship `npm-shrinkwrap.json` / lockfiles, so installs do not pull the multi-hundred-MB developer toolchain into `~/.pi/agent/npm`.
+
 <details>
 <summary>Manual install</summary>
 
@@ -234,18 +236,23 @@ When `cost` is unavailable (provider doesn't report it), the entire `cost` objec
 
 ## Testing
 
+Dev tooling (vitest, typescript, hooks) stays in the git repo only — use `pnpm` for local development.
+
 ```bash
-# Install dependencies
-npm install
+# Install dev dependencies (repo checkout only)
+pnpm install
 
 # Run tests
-npm test
+pnpm test
 
 # Run tests with coverage
-npm run test:coverage
+pnpm test:coverage
 
 # Type check
-npm run typecheck
+pnpm typecheck
+
+# Verify published tarball contents (should be ~3 runtime files)
+pnpm pack:check
 ```
 
 ---

package/package.json

@@ -1,11 +1,16 @@
 {
   "name": "@monotykamary/pi-tps",
-  "version": "1.3.5",
+  "version": "1.3.7",
   "description": "Tokens-per-second tracker for pi — see your LLM generation speed after every agent turn",
   "keywords": [
     "pi-package"
   ],
   "license": "MIT",
+  "files": [
+    "LICENSE",
+    "README.md",
+    "extensions/pi-tps/index.ts"
+  ],
   "pi": {
     "extensions": [
       "./extensions"
@@ -24,8 +29,8 @@
   "devDependencies": {
     "@commitlint/cli": "21.0.1",
     "@commitlint/config-conventional": "21.0.1",
-    "@earendil-works/pi-ai": "0.84.0",
-    "@earendil-works/pi-coding-agent": "0.84.0",
+    "@earendil-works/pi-ai": "0.84.2",
+    "@earendil-works/pi-coding-agent": "0.84.2",
     "@types/node": "25.9.1",
     "@vitest/coverage-v8": "4.1.7",
     "knip": "6.14.1",
@@ -51,6 +56,6 @@
     "test:coverage": "vitest run --coverage",
     "typecheck": "tsc --noEmit",
     "lint:dead": "knip --no-gitignore",
-    "postinstall": "simple-git-hooks 2>/dev/null || true"
+    "pack:check": "npm pack --dry-run"
   }
 }

package/.github/FUNDING.yml

@@ -1,4 +0,0 @@
-github: monotykamary
-ko_fi: monotykamary
-buy_me_a_coffee: monotykamary
-polar: monotykamary

package/.github/workflows/test.yml

@@ -1,57 +0,0 @@
-name: Test
-
-on:
-  push:
-    branches: [main]
-  pull_request:
-    branches: [main]
-
-jobs:
-  typecheck:
-    runs-on: ubuntu-latest
-
-    steps:
-      - uses: actions/checkout@v4
-
-      - uses: pnpm/action-setup@v4
-      - name: Setup Node.js
-        uses: actions/setup-node@v4
-        with:
-          node-version: '22'
-          cache: 'pnpm'
-
-      - name: Install dependencies
-        run: pnpm install --frozen-lockfile
-
-      - name: Type check
-        run: pnpm typecheck
-
-  test:
-    runs-on: ubuntu-latest
-    needs: typecheck
-
-    steps:
-      - uses: actions/checkout@v4
-
-      - uses: pnpm/action-setup@v4
-      - name: Setup Node.js
-        uses: actions/setup-node@v4
-        with:
-          node-version: '22'
-          cache: 'pnpm'
-
-      - name: Install dependencies
-        run: pnpm install --frozen-lockfile
-
-      - name: Run tests
-        run: pnpm test
-
-      - name: Run tests with coverage
-        run: pnpm test:coverage
-
-      - name: Upload coverage
-        uses: actions/upload-artifact@v4
-        with:
-          name: coverage
-          path: coverage/
-        if: always()

package/.pi/autoresearch/session-id

@@ -1 +0,0 @@
-019e1a55-4f8b-7797-849c-eb2b21f4fc7a

package/.pi/fabric/mesh/actors/actors.json

@@ -1,4 +0,0 @@
-{
-  "format": 1,
-  "actors": []
-}

package/.prettierrc

@@ -1,7 +0,0 @@
-{
-  "semi": true,
-  "singleQuote": true,
-  "trailingComma": "es5",
-  "tabWidth": 2,
-  "printWidth": 100
-}

package/commitlint.config.cjs

@@ -1 +0,0 @@
-module.exports = { extends: ['@commitlint/config-conventional'] };

package/extensions/pi-tps/__tests__/cost-rate.test.ts

@@ -1,503 +0,0 @@
-import { describe, it, expect, beforeEach, afterEach } from 'vitest';
-import type { AssistantMessage } from '@earendil-works/pi-ai';
-import { createTestFixture, activateExtension, tick, makeAssistantMessage } from './helpers';
-
-/**
- * Coverage for the blended $/M-tokens notification field:
- *   rateUsdPerMTokens = effectiveCost / (tokens.total / 1_000_000)
- *
- * effectiveCost is:
- *   - the Neuralwatt billed cost when stashed via the `neuralwatt:turn-energy`
- *     event (energy-based, what the user actually pays)
- *   - otherwise the list-price compute cost from message.usage.cost.total
- *
- * Only one source contributes per turn (no double-counting): when both are
- * present the billed cost wins outright.
- */
-
-function makeMessageWithCost(opts: {
-  input: number;
-  output: number;
-  costTotal: number;
-  provider?: string;
-  model?: string;
-}): AssistantMessage {
-  const { input, output, costTotal, provider = 'openai', model = 'gpt-4' } = opts;
-  return {
-    role: 'assistant',
-    content: [{ type: 'text', text: 'Hello' }],
-    api: 'openai-completions',
-    provider,
-    model,
-    usage: {
-      input,
-      output,
-      cacheRead: 0,
-      cacheWrite: 0,
-      totalTokens: input + output,
-      cost: {
-        input: 0,
-        output: 0,
-        cacheRead: 0,
-        cacheWrite: 0,
-        total: costTotal,
-      },
-    },
-    stopReason: 'stop',
-    timestamp: Date.now(),
-  };
-}
-
-/** Drive a minimal turn that yields primary-branch null TPS (burst) but valid telemetry. */
-async function runBurstTurn(
-  fixture: ReturnType<typeof createTestFixture>,
-  message: AssistantMessage,
-  turnIndex = 0,
-  beforeTurnEnd?: () => void
-) {
-  const { handlers, mockCtx } = fixture;
-  handlers['turn_start']?.({ type: 'turn_start', turnIndex, timestamp: Date.now() });
-  await tick(50);
-  handlers['message_start']?.({ type: 'message_start', message });
-  await tick(50);
-  handlers['message_update']?.({
-    type: 'message_update',
-    message,
-    assistantMessageEvent: { type: 'text_delta', delta: 't' },
-  });
-  handlers['message_end']?.({ type: 'message_end', message });
-  beforeTurnEnd?.();
-  handlers['turn_end']?.({ type: 'turn_end', turnIndex, message, toolResults: [] }, mockCtx);
-}
-
-describe('pi-tps extension — blended $/M-tokens rate', () => {
-  let fixture: ReturnType<typeof createTestFixture>;
-
-  beforeEach(async () => {
-    fixture = createTestFixture();
-    await activateExtension(fixture);
-  });
-
-  afterEach(() => {
-    vi.restoreAllMocks();
-  });
-
-  it('shows list-price $/M rate in the banner from message.usage.cost.total', async () => {
-    // 1000 in + 1000 out = 2000 tokens; cost.total = $0.008 → $4.00/M
-    const message = makeMessageWithCost({ input: 1000, output: 1000, costTotal: 0.008 });
-
-    await runBurstTurn(fixture, message);
-
-    const { notifySpy, appendEntrySpy } = fixture;
-    expect(notifySpy).toHaveBeenCalledOnce();
-    const banner = notifySpy.mock.calls[0][0] as string;
-    expect(banner).toContain('$4.00/M');
-    expect(banner).not.toMatch(/\$.*\/M.*\$.*\/M/); // exactly one rate segment
-
-    const [, data] = appendEntrySpy.mock.calls[0];
-    expect(data.rateUsdPerMTokens).toBe(4.0);
-  });
-
-  it('uses Neuralwatt billed cost over list-price when the energy event fires first', async () => {
-    // Same token volume, but billed cost differs from list price.
-    // 2000 tokens; list cost.total = $0.008 ($4.00/M); billed = $0.006 ($3.00/M)
-    const message = makeMessageWithCost({
-      input: 1000,
-      output: 1000,
-      costTotal: 0.008,
-      provider: 'neuralwatt',
-      model: 'moonshotai/Kimi-K2.5',
-    });
-
-    // Simulate the neuralwatt provider's turn_end running BEFORE ours: it emits
-    // the per-turn energy event after streaming but before pi-tps handles turn_end.
-    await runBurstTurn(fixture, message, 0, () => {
-      fixture.emitEvent('neuralwatt:turn-energy', {
-        costUsd: 0.006,
-        energyJoules: 21.6,
-        turnIndex: 0,
-      });
-    });
-
-    const { notifySpy, appendEntrySpy } = fixture;
-    expect(notifySpy).toHaveBeenCalledOnce();
-    const banner = notifySpy.mock.calls[0][0] as string;
-    // Billed ($3.00/M) wins over list-price ($4.00/M)
-    expect(banner).toContain('$3.00/M');
-    expect(banner).not.toContain('$4.00/M');
-
-    const [, data] = appendEntrySpy.mock.calls[0];
-    expect(data.rateUsdPerMTokens).toBe(3.0);
-    // The cost block still carries the list-price compute cost separately.
-    expect(data.cost).toEqual(expect.objectContaining({ total: 0.008 }));
-  });
-
-  it('null rate when totalTokens is zero (degenerate)', async () => {
-    const message = makeMessageWithCost({ input: 0, output: 1000, costTotal: 0.008 });
-    message.usage.totalTokens = 0;
-
-    await runBurstTurn(fixture, message);
-
-    const { appendEntrySpy, notifySpy } = fixture;
-    const banner = notifySpy.mock.calls[0][0] as string;
-    expect(banner).not.toMatch(/\$.*\/M/);
-    const [, data] = appendEntrySpy.mock.calls[0];
-    expect(data.rateUsdPerMTokens).toBeNull();
-  });
-
-  it('null rate when cost is unavailable', async () => {
-    // No usage.cost at all → effectiveCost null → rate null.
-    const message = makeAssistantMessage({ output: 500, input: 500 });
-    // Strip the cost block that makeAssistantMessage adds.
-    (message.usage as any).cost = null;
-    // Allow the isAssistantMessage guard to still pass: it only checks input/output.
-    await runBurstTurn(fixture, message);
-
-    const { appendEntrySpy } = fixture;
-    const [, data] = appendEntrySpy.mock.calls[0];
-    expect(data.rateUsdPerMTokens).toBeNull();
-    expect(data.cost).toBeNull();
-  });
-
-  it('omits the rate when pi reports an all-zero cost block for an unpriced model', async () => {
-    const message = makeMessageWithCost({
-      input: 500,
-      output: 500,
-      costTotal: 0,
-      provider: 'makora',
-      model: 'zai-org/GLM-5.2-NVFP4',
-    });
-
-    await runBurstTurn(fixture, message);
-
-    const { appendEntrySpy, notifySpy } = fixture;
-    const [, data] = appendEntrySpy.mock.calls[0];
-    expect(data.cost).toBeNull();
-    expect(data.rateUsdPerMTokens).toBeNull();
-    expect(notifySpy.mock.calls[0][0]).not.toMatch(/\$.*\/M/);
-  });
-
-  it('does not reuse a late billed cost after /tree when a turn index repeats', async () => {
-    const neuralwattMessage = makeMessageWithCost({
-      input: 1178,
-      output: 1235,
-      costTotal: 0.0102352,
-      provider: 'neuralwatt',
-      model: 'glm-5.2-short',
-    });
-    neuralwattMessage.usage.cacheRead = 8192;
-    neuralwattMessage.usage.totalTokens = 10605;
-
-    await runBurstTurn(fixture, neuralwattMessage, 2);
-    fixture.emitEvent('neuralwatt:turn-energy', {
-      costUsd: 0.00116,
-      energyJoules: 835.2,
-      turnIndex: 2,
-    });
-
-    fixture.handlers['session_tree']?.(
-      { type: 'session_tree', newLeafId: null, oldLeafId: 'old-leaf' },
-      fixture.mockCtx
-    );
-
-    const unpricedMessage = makeMessageWithCost({
-      input: 8460,
-      output: 717,
-      costTotal: 0,
-      provider: 'makora',
-      model: 'zai-org/GLM-5.2-NVFP4',
-    });
-    unpricedMessage.usage.cacheRead = 1216;
-    unpricedMessage.usage.totalTokens = 10393;
-
-    await runBurstTurn(fixture, unpricedMessage, 2);
-
-    const latestTelemetry = fixture.appendEntrySpy.mock.calls.at(-1)![1];
-    expect(latestTelemetry.model).toEqual({
-      provider: 'makora',
-      modelId: 'zai-org/GLM-5.2-NVFP4',
-    });
-    expect(latestTelemetry.cost).toBeNull();
-    expect(latestTelemetry.rateUsdPerMTokens).toBeNull();
-
-    const latestBanner = fixture.notifySpy.mock.calls.at(-1)![0] as string;
-    expect(latestBanner).not.toContain('$0.11/M');
-    expect(latestBanner).not.toMatch(/\$.*\/M/);
-  });
-
-  it('does not apply an early current-turn event to a previous run with the same index', async () => {
-    const previousMessage = makeMessageWithCost({
-      input: 1000,
-      output: 1000,
-      costTotal: 0.008,
-    });
-    await runBurstTurn(fixture, previousMessage, 0);
-
-    const currentMessage = makeMessageWithCost({
-      input: 1000,
-      output: 1000,
-      costTotal: 0.01,
-      provider: 'neuralwatt',
-      model: 'glm-5.2',
-    });
-
-    await runBurstTurn(fixture, currentMessage, 0, () => {
-      fixture.emitEvent('neuralwatt:turn-energy', {
-        costUsd: 0.006,
-        energyJoules: 21.6,
-        turnIndex: 0,
-      });
-      expect(fixture.appendEntrySpy).toHaveBeenCalledTimes(1);
-    });
-
-    expect(fixture.appendEntrySpy).toHaveBeenCalledTimes(2);
-    const latestTelemetry = fixture.appendEntrySpy.mock.calls.at(-1)![1];
-    expect(latestTelemetry.rateUsdPerMTokens).toBe(3);
-  });
-
-  it('falls back to list-price rate when billed-cost event misses (out-of-order load)', async () => {
-    // Neuralwatt turn but the energy event never arrives (provider loaded after
-    // us). Must not block or crash — falls back to the list-price compute rate.
-    const message = makeMessageWithCost({
-      input: 500,
-      output: 500,
-      costTotal: 0.004, // $4.00/M for 1000 tokens
-      provider: 'neuralwatt',
-      model: 'moonshotai/Kimi-K2.5',
-    });
-
-    await runBurstTurn(fixture, message);
-
-    const { notifySpy } = fixture;
-    const banner = notifySpy.mock.calls[0][0] as string;
-    expect(banner).toContain('$4.00/M');
-  });
-
-  it('corrects the banner + persisted entry when the energy event arrives late', async () => {
-    // Provider loads AFTER pi-tps: our turn_end runs first on the list-price
-    // fallback, then the provider's turn_end emits the energy event, which
-    // must retroactively correct the rate without a second turn_end.
-    const message = makeMessageWithCost({
-      input: 1000,
-      output: 1000,
-      costTotal: 0.008, // list price: $4.00/M; billed: $3.00/M
-      provider: 'neuralwatt',
-      model: 'moonshotai/Kimi-K2.5',
-    });
-
-    await runBurstTurn(fixture, message, 0);
-
-    const { notifySpy, appendEntrySpy } = fixture;
-    // Initially committed on the list-price fallback.
-    expect(appendEntrySpy.mock.calls[0][0]).toBe('tps');
-    expect(appendEntrySpy.mock.calls[0][1].rateUsdPerMTokens).toBe(4.0);
-    expect(notifySpy.mock.calls[0][0]).toContain('$4.00/M');
-
-    // Provider's turn_end fires the energy event after ours.
-    fixture.emitEvent('neuralwatt:turn-energy', {
-      costUsd: 0.006, // $3.00/M for 2000 tokens
-      energyJoules: 21.6,
-      turnIndex: 0,
-    });
-
-    // A corrected `tps` entry is appended in place of the list-price one.
-    expect(appendEntrySpy.mock.calls).toHaveLength(2);
-    expect(appendEntrySpy.mock.calls[1][0]).toBe('tps');
-    expect(appendEntrySpy.mock.calls[1][1].rateUsdPerMTokens).toBe(3.0);
-
-    // Banner is refreshed to the billed rate.
-    const lastBanner = notifySpy.mock.calls.at(-1)![0] as string;
-    expect(lastBanner).toContain('$3.00/M');
-    expect(lastBanner).not.toContain('$4.00/M');
-  });
-
-  it('does not double-correct a turn already billed at commit time', async () => {
-    // Provider loads BEFORE pi-tps: the event fires first, our turn_end reads
-    // it from the live cache (billedApplied=true), so no late correction.
-    const message = makeMessageWithCost({
-      input: 1000,
-      output: 1000,
-      costTotal: 0.008,
-      provider: 'neuralwatt',
-      model: 'moonshotai/Kimi-K2.5',
-    });
-
-    await runBurstTurn(fixture, message, 0, () => {
-      fixture.emitEvent('neuralwatt:turn-energy', {
-        costUsd: 0.006,
-        energyJoules: 21.6,
-        turnIndex: 0,
-      });
-    });
-
-    // A stray duplicate event must not append a second corrected entry.
-    fixture.emitEvent('neuralwatt:turn-energy', {
-      costUsd: 0.006,
-      energyJoules: 21.6,
-      turnIndex: 0,
-    });
-
-    const { appendEntrySpy, notifySpy } = fixture;
-    expect(appendEntrySpy.mock.calls.filter((c) => c[0] === 'tps')).toHaveLength(1);
-    expect(notifySpy).toHaveBeenCalledOnce();
-    const banner = notifySpy.mock.calls[0][0] as string;
-    expect(banner).toContain('$3.00/M');
-  });
-
-  it('ignores neuralwatt:turn-energy payloads lacking a numeric turnIndex', async () => {
-    // Defensive: malformed event must not pollute the cache.
-    fixture.emitEvent('neuralwatt:turn-energy', { costUsd: 0.006, energyJoules: 21.6 }); // no turnIndex
-    const message = makeMessageWithCost({ input: 1000, output: 1000, costTotal: 0.008 });
-
-    await runBurstTurn(fixture, message);
-
-    const { notifySpy } = fixture;
-    const banner = notifySpy.mock.calls[0][0] as string;
-    // Falls back to list-price since no valid turnIndex-keyed entry landed.
-    expect(banner).toContain('$4.00/M');
-  });
-
-  it('rehydrates a rate segment from structured telemetry on session resume', async () => {
-    const { handlers, notifySpy, mockEntries } = fixture;
-
-    mockEntries.push({
-      type: 'custom',
-      customType: 'tps',
-      data: {
-        model: { provider: 'openai', modelId: 'gpt-4' },
-        tokens: { input: 1000, output: 1000, cacheRead: 0, cacheWrite: 0, total: 2000 },
-        timing: {
-          ttftMs: 1000,
-          totalMs: 3000,
-          generationMs: 2000,
-          stallMs: 0,
-          stallCount: 0,
-          messageCount: 1,
-        },
-        tps: 10.0,
-        rateUsdPerMTokens: 4.0,
-        timestamp: Date.now(),
-      },
-    });
-
-    handlers['session_start']?.({ reason: 'resume' }, fixture.mockCtx);
-    await tick();
-
-    expect(notifySpy).toHaveBeenCalledOnce();
-    const banner = notifySpy.mock.calls[0][0] as string;
-    expect(banner).toContain('$4.00/M');
-  });
-
-  it('rehydrates older telemetry without a rate field without crashing (omits the segment)', async () => {
-    const { handlers, notifySpy, mockEntries } = fixture;
-
-    mockEntries.push({
-      type: 'custom',
-      customType: 'tps',
-      data: {
-        model: { provider: 'openai', modelId: 'gpt-4' },
-        tokens: { input: 50, output: 100, cacheRead: 0, cacheWrite: 0, total: 150 },
-        timing: {
-          ttftMs: 1000,
-          totalMs: 3000,
-          generationMs: 2000,
-          stallMs: 0,
-          stallCount: 0,
-          messageCount: 1,
-        },
-        tps: 10.0,
-        // rateUsdPerMTokens intentionally absent (pre-feature entry)
-        timestamp: Date.now(),
-      },
-    });
-
-    handlers['session_start']?.({ reason: 'resume' }, fixture.mockCtx);
-    await tick();
-
-    expect(notifySpy).toHaveBeenCalledOnce();
-    const banner = notifySpy.mock.calls[0][0] as string;
-    expect(banner).not.toMatch(/\$.*\/M/);
-  });
-
-  it('falls back to a persisted neuralwatt-energy entry when the live event is missed', async () => {
-    const now = Date.now();
-    const message = makeMessageWithCost({
-      input: 1000,
-      output: 1000,
-      costTotal: 0.008, // list price: $4.00/M
-      provider: 'neuralwatt',
-      model: 'moonshotai/Kimi-K2.5',
-    });
-
-    // Provider appended its energy entry before pi-tps handled turn_end.
-    fixture.mockEntries.push({
-      type: 'custom',
-      customType: 'neuralwatt-energy',
-      data: { energy_joules: 21.6, cost_usd: 0.006 }, // $3.00/M for 2000 tokens
-      timestamp: now,
-    });
-
-    const { handlers, notifySpy, appendEntrySpy } = fixture;
-    handlers['turn_start']?.({ type: 'turn_start', turnIndex: 0, timestamp: now - 100 });
-    await tick(50);
-    handlers['message_start']?.({ type: 'message_start', message });
-    await tick(50);
-    handlers['message_update']?.({
-      type: 'message_update',
-      message,
-      assistantMessageEvent: { type: 'text_delta', delta: 'H' },
-    });
-    handlers['message_end']?.({ type: 'message_end', message });
-    handlers['turn_end']?.(
-      { type: 'turn_end', turnIndex: 0, message, toolResults: [] },
-      fixture.mockCtx
-    );
-
-    expect(notifySpy).toHaveBeenCalledOnce();
-    const banner = notifySpy.mock.calls[0][0] as string;
-    expect(banner).toContain('$3.00/M');
-    expect(banner).not.toContain('$4.00/M');
-
-    const [, data] = appendEntrySpy.mock.calls[0];
-    expect(data.rateUsdPerMTokens).toBe(3.0);
-  });
-
-  it('ignores stale neuralwatt-energy entries from before this turn', async () => {
-    const now = Date.now();
-    const message = makeMessageWithCost({
-      input: 1000,
-      output: 1000,
-      costTotal: 0.008, // list price: $4.00/M
-      provider: 'neuralwatt',
-      model: 'moonshotai/Kimi-K2.5',
-    });
-
-    fixture.mockEntries.push({
-      type: 'custom',
-      customType: 'neuralwatt-energy',
-      data: { energy_joules: 1, cost_usd: 9999 }, // absurd cost, should be ignored
-      timestamp: now - 1000,
-    });
-
-    const { handlers, notifySpy } = fixture;
-    handlers['turn_start']?.({ type: 'turn_start', turnIndex: 0, timestamp: now });
-    await tick(50);
-    handlers['message_start']?.({ type: 'message_start', message });
-    await tick(50);
-    handlers['message_update']?.({
-      type: 'message_update',
-      message,
-      assistantMessageEvent: { type: 'text_delta', delta: 'H' },
-    });
-    handlers['message_end']?.({ type: 'message_end', message });
-    handlers['turn_end']?.(
-      { type: 'turn_end', turnIndex: 0, message, toolResults: [] },
-      fixture.mockCtx
-    );
-
-    expect(notifySpy).toHaveBeenCalledOnce();
-    const banner = notifySpy.mock.calls[0][0] as string;
-    expect(banner).toContain('$4.00/M');
-    expect(banner).not.toContain('$9999');
-  });
-});