pi-ptc-meter 1.0.25 → 1.0.27

package/README.md

@@ -56,7 +56,7 @@ pi install npm:pi-ptc-meter
 
 ## 作用（源码事实）
 
-- `tool_call`：计数 +1（不拦截、不改调用）。
+- `tool_call`：顶层调用计数 +1（codemode 内层调用不另计；不拦截、不改调用）。
 - `turn_end`：记录本轮次数与耗时（不落盘）。
 - `agent_end`：每 run 落盘一行日志。
 - `before_agent_start`：结算上一轮并以 `nextTurn` 尾部消息回灌——计数、耗时、偏差、基线中位、**缓存命中率**（仅当该轮有 usage 数据时出现；< 90% 标「低」）——（`display: false`）。
@@ -72,27 +72,38 @@ PTC 计量：上一轮 9 次工具调用 · 实际 34s · 命中 92% · 预计 8
 
 ## 原理（专业）
 
-```mermaid
-flowchart TD
-    A[本轮开工前<br/>声明 预计 N 次 / 预计 Ns] --> B[tool_call<br/>计数 +1]
-    B --> C[turn_end<br/>记录实际次数与耗时 + 解析预计值]
-    C --> D[agent_end<br/>落盘一行日志（含预计）]
-    D --> E[before_agent_start<br/>结算上一轮]
-    E --> F[对比：实际 vs 预计 → 偏差 %]
-    F -->|nextTurn 尾部消息| A
-    F -.->|system 只留静态骨架| S[ptc_meter section]
+```text
+本轮开工前 ─ 声明「预计 N 次 / 预计 Ns」
+        ↓
+tool_call ─ 仅顶层调用计数 +1
+        ↓
+turn_end ─ 记录实际次数与耗时 + 解析预计值
+        ↓
+agent_end ─ 落盘一行日志（含预计）
+        ↓
+before_agent_start ─ 结算上一轮
+        ↓
+对比：实际 vs 预计 → 偏差 %
+        ↓ nextTurn 尾部消息
+（回到本轮开工前）
+
+system 只留静态骨架（ptc_meter section）
 ```
 
 ## 原理（小白）
 
 **像记账对账**：开工前先说「这趟大概 5 步」，收工后对账「实际 7 步，比预计多 40%」，下次开工前把账报给你。
 
-```mermaid
-flowchart TD
-    A[开工前先估：预计几步] --> B[干活：实际跑几步]
-    B --> C[收工对比：实际 vs 预计]
-    C --> D[下次开工时报出偏差]
-    D --> A
+```text
+开工前先估：预计几步
+        ↓
+干活：实际跑几步
+        ↓
+收工对比：实际 vs 预计
+        ↓
+下次开工时报出偏差
+        ↓
+（回到开工前先估）
 ```
 
 ## License
@@ -152,7 +163,7 @@ Also: log `~/.pi/ptc-meter.jsonl` (1 MiB truncation, tail-preserving); no enviro
 
 ## What it does (source facts)
 
-- `tool_call`: increments a counter (never intercepts or rewrites).
+- `tool_call`: increments the counter for top-level calls only (nested codemode calls are not counted separately; never intercepts or rewrites).
 - `turn_end`: records calls and duration (no disk write).
 - `agent_end`: writes one log line per run.
 - `before_agent_start`: settles the previous turn and feeds back counts, duration, deviation, the baseline median and the **cache hit rate** (only when that turn reported usage; flagged low below 90%) as a `nextTurn` trailing message (`display: false`).
@@ -168,27 +179,38 @@ PTC metering: last turn 9 tool calls · 34s actual · 92% cache hit · estimated
 
 ## How it works (technical)
 
-```mermaid
-flowchart TD
-    A[before the turn<br/>declare: about N calls / Ns] --> B[tool_call<br/>counter +1]
-    B --> C[turn_end<br/>record actual calls and duration + parse estimate]
-    C --> D[agent_end<br/>write one log line (includes estimate)]
-    D --> E[before_agent_start<br/>settle the previous turn]
-    E --> F[compare: actual vs estimate -> deviation %]
-    F -->|nextTurn message| A
-    F -.->|system keeps a static skeleton| S[ptc_meter section]
+```text
+before the turn ─ declare: about N calls / Ns
+        ↓
+tool_call ─ top-level call: counter +1
+        ↓
+turn_end ─ record actual calls and duration + parse estimate
+        ↓
+agent_end ─ write one log line (includes estimate)
+        ↓
+before_agent_start ─ settle the previous turn
+        ↓
+compare: actual vs estimate -> deviation %
+        ↓ nextTurn message
+(back to before the turn)
+
+the system prompt keeps only a static skeleton (ptc_meter section)
 ```
 
 ## How it works (plain words)
 
 **Like a ledger**: you estimate "about 5 steps" before starting, the ledger compares it with the actual 7 (40% over), and reports the gap before the next job.
 
-```mermaid
-flowchart TD
-    A[before starting: guess N steps] --> B[work: actually N steps]
-    B --> C[at the end: actual vs guess]
-    C --> D[report the gap next time]
-    D --> A
+```text
+before starting: guess N steps
+        ↓
+work: actually N steps
+        ↓
+at the end: actual vs guess
+        ↓
+report the gap next time
+        ↓
+(back to before starting)
 ```
 ## License
 

package/index.ts

@@ -268,9 +268,13 @@ export default function ptcMeter(pi: ExtensionAPI) {
     turns = 0;
     turnStartedAt = Date.now();
   });
-  // 计数：不读 event 字段、不返回任何值 —— 这是最安全的形态（该层无 try/catch）。
-  pi.on("tool_call", async () => {
-    try { callsThisTurn++; } catch { /* 静默：计量不该影响调用 */ }
+  // 计数：只计**顶层**调用 —— codemode 的内层 tools.X 也走本 hook（pi 的 _executeNestedToolCall
+  // 显式带 parentToolCallId），计进去会把一次往返记成 N 次（实测 21 次顶层被记成 53）。
+  pi.on("tool_call", async (event) => {
+    try {
+      if ((event as { parentToolCallId?: string } | undefined)?.parentToolCallId) return undefined;
+      callsThisTurn++;
+    } catch { /* 静默：计量不该影响调用 */ }
     return undefined;
   });
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-ptc-meter",
-  "version": "1.0.25",
+  "version": "1.0.27",
   "description": "Measures tool-call batching per turn and feeds the numbers back to the model.",
   "type": "module",
   "license": "MIT",