throughline 0.4.5 → 0.4.6

package/CHANGELOG.md

@@ -10,6 +10,15 @@ shipped to npm but were not individually tagged on GitHub.
 
 ## [Unreleased]
 
+## [0.4.6] — 2026-05-09
+
+### Changed
+
+- Codex monitor usage now overlays transient `output_tokens` while a Codex turn
+  is open. During an in-flight turn the row displays `input_tokens +
+  output_tokens` and marks the model with `live+<tokens>`; after `task_complete`
+  the row drops back to verified `input_tokens` only.
+
 ## [0.4.5] — 2026-05-09
 
 ### Fixed

package/README.ja.md

@@ -216,6 +216,8 @@ throughline monitor --session <id-prefix>
 監視中は Claude transcript / Codex rollout をライブに読み、Stop hook の state
 snapshot はライブ usage が取れない場合の控えとして使います。これにより表示更新は
 Stop 完了待ちではなくなります。
+Codex は open turn 中だけ `input_tokens + output_tokens` を表示し、モデル欄に
+`live+<tokens>` を付けます。`task_complete` 後は verified `input_tokens` のみに戻ります。
 
 詳細仕様 (resize 追従、1M context 検出、ステイル隠し、Stop hook の非同期化など) は
 [英語版 README](README.md#multi-session-token-monitor) を参照してください。

package/README.md

@@ -519,10 +519,12 @@ Example output:
 - **Codex token counts use the rollout `token_count` event when present.** The
   Codex Stop hook writes `codex:<thread_id>` monitor state with the rollout
   path. While the monitor is running it reads the live rollout every tick and
-  prefers the latest verified `token_count` sample. If a Codex rollout has no
-  token-count event, Throughline can show an explicit estimate with
-  `estimated: true` and the monitor marks it with `est`; it is not presented as
-  exact usage.
+  prefers the latest verified `token_count` sample. During an open Codex turn,
+  the monitor overlays transient `output_tokens` on top of `input_tokens` and
+  marks the model with `live+<tokens>`; when `task_complete` arrives it drops
+  back to verified `input_tokens` only. If a Codex rollout has no token-count
+  event, Throughline can show an explicit estimate with `estimated: true` and
+  the monitor marks it with `est`; it is not presented as exact usage.
 - **Codex auto-refresh mutates at the verified 90% threshold.** The Codex Stop
   hook captures DB memory, writes monitor state, and when verified usage reaches
   the threshold it attempts rollback + Throughline DB memory injection for the

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "throughline",
-  "version": "0.4.5",
+  "version": "0.4.6",
   "type": "module",
   "description": "Claude Code hooks plugin for structured context compression (/clear-safe persistent memory)",
   "keywords": [

package/src/codex-usage.mjs

@@ -18,6 +18,7 @@ export function readLatestCodexUsage(rolloutPath) {
   let latest = null;
   let model = 'codex';
   let provider = null;
+  let openTaskCount = 0;
   for (const line of raw.split('\n')) {
     if (!line.trim()) continue;
     let row;
@@ -39,24 +40,43 @@ export function readLatestCodexUsage(rolloutPath) {
       continue;
     }
 
-    if (row?.type !== 'event_msg' || payload?.type !== 'token_count') continue;
+    if (row?.type !== 'event_msg') continue;
+
+    if (payload?.type === 'task_started') {
+      openTaskCount++;
+      continue;
+    }
+
+    if (payload?.type === 'task_complete') {
+      openTaskCount = Math.max(0, openTaskCount - 1);
+      continue;
+    }
+
+    if (payload?.type !== 'token_count') continue;
     const info = payload.info ?? {};
     const last = info.last_token_usage ?? {};
-    const tokens = Number(last.input_tokens);
-    if (!Number.isFinite(tokens) || tokens < 0) continue;
+    const inputTokens = Number(last.input_tokens);
+    if (!Number.isFinite(inputTokens) || inputTokens < 0) continue;
+    const outputTokens = Number.isFinite(Number(last.output_tokens)) ? Number(last.output_tokens) : 0;
+    const transientOutputTokens = openTaskCount > 0 ? outputTokens : 0;
 
     const windowSize = Number(info.model_context_window);
     latest = {
-      tokens,
+      tokens: inputTokens + transientOutputTokens,
+      inputTokens,
       model: model === 'codex' && provider ? provider : model,
       contextWindowSize:
         Number.isFinite(windowSize) && windowSize > 0
           ? windowSize
           : DEFAULT_CODEX_CONTEXT_WINDOW_SIZE,
       contextWindowEstimated: !(Number.isFinite(windowSize) && windowSize > 0),
-      outputTokens: Number.isFinite(Number(last.output_tokens)) ? Number(last.output_tokens) : 0,
+      outputTokens,
+      transientOutputTokens,
+      liveTurn: openTaskCount > 0,
       estimated: false,
-      source: 'codex-rollout-token-count',
+      source: openTaskCount > 0
+        ? 'codex-rollout-token-count-live-turn'
+        : 'codex-rollout-token-count',
     };
   }
 

package/src/codex-usage.test.mjs

@@ -44,10 +44,13 @@ test('readLatestCodexUsage: reads verified Codex token_count event shape', () =>
 
     assert.deepEqual(readLatestCodexUsage(rollout), {
       tokens: 151914,
+      inputTokens: 151914,
       model: 'gpt-5.5',
       contextWindowSize: 258400,
       contextWindowEstimated: false,
       outputTokens: 60,
+      transientOutputTokens: 0,
+      liveTurn: false,
       estimated: false,
       source: 'codex-rollout-token-count',
     });
@@ -56,6 +59,86 @@ test('readLatestCodexUsage: reads verified Codex token_count event shape', () =>
   }
 });
 
+test('readLatestCodexUsage: during an open Codex turn overlays transient output tokens', () => {
+  const dir = mkdtempSync(join(tmpdir(), 'tl-codex-usage-'));
+  try {
+    const rollout = join(dir, 'rollout.jsonl');
+    writeFileSync(
+      rollout,
+      [
+        row('session_meta', { id: '019dfaba-thread', cwd: '/repo' }),
+        row('turn_context', { turn_id: '019dfaba-turn', model: 'gpt-5.5' }),
+        event('task_started'),
+        event('token_count', {
+          info: {
+            last_token_usage: {
+              input_tokens: 151914,
+              output_tokens: 1200,
+            },
+            model_context_window: 258400,
+          },
+        }),
+      ]
+        .map((r) => JSON.stringify(r))
+        .join('\n') + '\n',
+    );
+
+    const usage = readLatestCodexUsage(rollout);
+    assert.equal(usage.tokens, 153114);
+    assert.equal(usage.inputTokens, 151914);
+    assert.equal(usage.outputTokens, 1200);
+    assert.equal(usage.transientOutputTokens, 1200);
+    assert.equal(usage.liveTurn, true);
+    assert.equal(usage.source, 'codex-rollout-token-count-live-turn');
+  } finally {
+    rmSync(dir, { recursive: true, force: true });
+  }
+});
+
+test('readLatestCodexUsage: task_complete drops transient output overlay', () => {
+  const dir = mkdtempSync(join(tmpdir(), 'tl-codex-usage-'));
+  try {
+    const rollout = join(dir, 'rollout.jsonl');
+    writeFileSync(
+      rollout,
+      [
+        row('session_meta', { id: '019dfaba-thread', cwd: '/repo' }),
+        row('turn_context', { turn_id: '019dfaba-turn', model: 'gpt-5.5' }),
+        event('task_started'),
+        event('token_count', {
+          info: {
+            last_token_usage: {
+              input_tokens: 151914,
+              output_tokens: 1200,
+            },
+            model_context_window: 258400,
+          },
+        }),
+        event('task_complete'),
+        event('token_count', {
+          info: {
+            last_token_usage: {
+              input_tokens: 151914,
+              output_tokens: 1200,
+            },
+            model_context_window: 258400,
+          },
+        }),
+      ]
+        .map((r) => JSON.stringify(r))
+        .join('\n') + '\n',
+    );
+
+    const usage = readLatestCodexUsage(rollout);
+    assert.equal(usage.tokens, 151914);
+    assert.equal(usage.transientOutputTokens, 0);
+    assert.equal(usage.liveTurn, false);
+    assert.equal(usage.source, 'codex-rollout-token-count');
+  } finally {
+    rmSync(dir, { recursive: true, force: true });
+  }
+});
+
 test('readLatestCodexUsage: falls back to model_provider when no turn_context model exists', () => {
   const dir = mkdtempSync(join(tmpdir(), 'tl-codex-usage-'));
   try {

package/src/token-monitor.mjs

@@ -318,8 +318,11 @@ function formatLine({ state, usage, isActive, now = Date.now() }) {
   const tokCol = `${formatNumber(tokens).padStart(6)} / ${formatNumber(max).padStart(6)}`;
   const estimateMark = usage?.estimated ? ' est' : '';
   const windowMark = usage?.contextWindowEstimated ? ' win?' : '';
+  const liveMark = usage?.liveTurn && usage?.transientOutputTokens
+    ? ` live+${formatNumber(usage.transientOutputTokens)}`
+    : '';
   const modelCol = usage?.model
-    ? color(ANSI.dim, `${usage.model}${estimateMark}${windowMark}`)
+    ? color(ANSI.dim, `${usage.model}${estimateMark}${windowMark}${liveMark}`)
     : color(ANSI.dim, '(未取得)');
   // 最終更新からの経過: 表示が「止まって見える」とき、それが idle なのか障害なのかを
   // 即座に判別できるようにする。updatedAt は state.writeSessionState 時の Date.now()。

package/src/token-monitor.test.mjs

@@ -451,6 +451,27 @@ test('formatLine: Codex estimated usage は host と est marker を表示する'
   assert.ok(out.includes('codex est win?'));
 });
 
+test('formatLine: Codex live turn usage は transient output marker を表示する', () => {
+  const args = makeLineArgs(0.5);
+  args.state.host = 'codex';
+  args.usage = {
+    tokens: 101_200,
+    inputTokens: 100_000,
+    model: 'gpt-5.5',
+    contextWindowSize: 200_000,
+    contextWindowEstimated: false,
+    outputTokens: 1200,
+    transientOutputTokens: 1200,
+    liveTurn: true,
+    estimated: false,
+    source: 'codex-rollout-token-count-live-turn',
+  };
+  const out = stripColors(formatLine(args));
+  assert.ok(out.includes('Codex'));
+  assert.ok(out.includes('101.2k / 200.0k'));
+  assert.ok(out.includes('gpt-5.5 live+1.2k'));
+});
+
 test('formatLine: 70% 以上で "!" マーカーと弱めの文言', () => {
   const out = stripColors(formatLine(makeLineArgs(0.75)));
   assert.ok(out.includes('!'), 'should include ! marker');