@cydm/pie 2.6.5 → 2.6.7

package/dist/builtin/skills/chrome-devtools-axi/chrome-devtools-axi-bridge.js

@@ -4,10 +4,11 @@ import {
   BRIDGE_PORT_IN_USE_EXIT_CODE,
   PAGE_IDENTITY_CHANGED_ERROR,
   clearSelectedPageId,
+  readSelfStartTime,
   resolveSessionName,
   resolveSessionPidFile,
   resolveSessionPort
-} from "./chunk-UPATXALW.js";
+} from "./chunk-U75F4DPF.js";
 import {
   __commonJS,
   __export,
@@ -19599,7 +19600,7 @@ async function isBridgeTargetReachable(client) {
 function writePidFile(port) {
   const pidFile = resolveSessionPidFile();
   mkdirSync(dirname(pidFile), { recursive: true });
-  writeFileSync(pidFile, JSON.stringify({ pid: process.pid, port }));
+  writeFileSync(pidFile, JSON.stringify({ pid: process.pid, port, startTime: readSelfStartTime() }));
 }
 function removePidFile(pidFile = resolveSessionPidFile(), ownerPid = process.pid) {
   try {

package/dist/builtin/skills/chrome-devtools-axi/chrome-devtools-axi-runtime.js

@@ -21,6 +21,6 @@ function tryFastPath(argv, options) {
 
 // vendor/chrome-devtools-axi/bin/chrome-devtools-axi.ts
 if (!tryFastPath(process.argv.slice(2), { version: VERSION })) {
-  const { main } = await import("./cli-5Z4COIKQ.js");
+  const { main } = await import("./cli-4ITIR57L.js");
   await main(process.argv.slice(2));
 }

package/dist/builtin/skills/chrome-devtools-axi/{chunk-UPATXALW.js → chunk-U75F4DPF.js}

@@ -200,10 +200,70 @@ function resolveBridgeScript(importMetaDir) {
 var BRIDGE_PORT_IN_USE_EXIT_CODE = 48;
 var PAGE_IDENTITY_CHANGED_ERROR = "The browser reconnected and every page id changed, so this call did not target the page you selected";
 
+// vendor/chrome-devtools-axi/src/process-identity.ts
+import { execFileSync } from "node:child_process";
+function inspectProcessIdentity(pid) {
+  if (!Number.isInteger(pid) || pid <= 0) return {};
+  if (process.platform === "win32") {
+    try {
+      const output = execFileSync(
+        "powershell",
+        [
+          "-NoProfile",
+          "-NonInteractive",
+          "-Command",
+          `$p = Get-CimInstance Win32_Process -Filter "ProcessId=<pid>"; if ($p) { $p.CommandLine; $p.CreationDate.ToString('o') }`.replace("<pid>", String(pid))
+        ],
+        { encoding: "utf-8", timeout: 4e3 }
+      );
+      const lines = output.split(/\r?\n/).filter((line) => line.length > 0);
+      if (lines.length === 0) return {};
+      const startTime = lines.length >= 2 ? lines[lines.length - 1] : void 0;
+      return { command: lines.slice(0, lines.length - 1).join("\n"), startTime };
+    } catch {
+      return {};
+    }
+  }
+  try {
+    const psEnv = { ...process.env, LC_ALL: "C", LANG: "C" };
+    const output = execFileSync(
+      "ps",
+      ["-p", String(pid), "-o", "lstart=", "-o", "command="],
+      { encoding: "utf-8", timeout: 1e3, env: psEnv }
+    );
+    const firstNewline = output.indexOf("\n");
+    const firstLine = firstNewline === -1 ? output : output.slice(0, firstNewline);
+    const lstart = /^(?:Mon|Tue|Wed|Thu|Fri|Sat|Sun)\s+(?:Jan|Feb|Mar|Apr|May|Jun|Jul|Aug|Sep|Oct|Nov|Dec)\s+\d{1,2}\s+\d{1,2}:\d{2}:\d{2}\s+\d{4}\s+/.exec(firstLine);
+    if (lstart) {
+      return { startTime: lstart[0].trimEnd(), command: output.slice(lstart[0].length) };
+    }
+    try {
+      const command = execFileSync("ps", ["-p", String(pid), "-o", "command="], {
+        encoding: "utf-8",
+        timeout: 1e3,
+        env: psEnv
+      });
+      return { command };
+    } catch {
+      return {};
+    }
+  } catch {
+    return {};
+  }
+}
+function readProcessStartTime(pid) {
+  return inspectProcessIdentity(pid).startTime;
+}
+function readSelfStartTime() {
+  return readProcessStartTime(process.pid);
+}
+
 export {
   resolveBridgeScript,
   BRIDGE_PORT_IN_USE_EXIT_CODE,
   PAGE_IDENTITY_CHANGED_ERROR,
+  inspectProcessIdentity,
+  readSelfStartTime,
   resolveSessionName,
   resolveSessionPort,
   resolveSessionStateDir,

package/dist/builtin/skills/chrome-devtools-axi/{cli-5Z4COIKQ.js → cli-4ITIR57L.js}

@@ -7,6 +7,7 @@ import {
   PAGE_IDENTITY_CHANGED_ERROR,
   clearSelectedPageId,
   getSelectedPageId,
+  inspectProcessIdentity,
   overlaySessionSelected,
   rememberToolRouting,
   resolveBridgeScript,
@@ -14,7 +15,7 @@ import {
   resolveSessionPidFile,
   resolveSessionPort,
   resolveSessionStateDir
-} from "./chunk-UPATXALW.js";
+} from "./chunk-U75F4DPF.js";
 import "./chunk-OJ6N23UZ.js";
 
 // ../../node_modules/chrome-devtools-axi/node_modules/@toon-format/toon/dist/index.mjs
@@ -1964,42 +1965,79 @@ function commandLineLooksLikeBridge(command) {
   const tokens = splitCommandLine(command);
   if (tokens.length === 0) return false;
   if (!/(?:^|[\\/])(node|npx|tsx)(?:\.(?:exe|cmd))?$/i.test(tokens[0])) return false;
+  const operandOptions = /* @__PURE__ */ new Set([
+    "--require",
+    "-r",
+    "--import",
+    "--loader",
+    "--experimental-loader",
+    "--conditions",
+    "-C",
+    "--inspect-port",
+    "--title",
+    "--watch-path",
+    "--max-old-space-size",
+    "--max-semi-space-size"
+  ]);
+  let optionsEnded = false;
+  const npxRunner = /(?:^|[\\/])npx(?:\.(?:exe|cmd))?$/i.test(tokens[0]);
+  let loaderSeen = false;
   for (let index = 1; index < tokens.length; index++) {
     const token = tokens[index];
-    if (token.startsWith("-")) continue;
-    if (/(?:^|[\\/])tsx(?:\.(?:exe|cmd))?$/i.test(token)) continue;
+    if (!optionsEnded && token === "--") {
+      optionsEnded = true;
+      continue;
+    }
+    if (!optionsEnded && token.startsWith("-")) {
+      const option = token.split("=", 1)[0];
+      if (["-e", "--eval", "-p", "--print"].includes(option)) return false;
+      if (!token.includes("=") && operandOptions.has(option)) index++;
+      continue;
+    }
+    if (npxRunner && !loaderSeen && /(?:^|[\\/])tsx(?:\.(?:exe|cmd))?$/i.test(token)) {
+      loaderSeen = true;
+      continue;
+    }
     const base = token.split(/[\\/]/).pop() ?? token;
     return /^chrome-devtools-axi-bridge\.(?:js|mjs|cjs|ts)$/i.test(base);
   }
   return false;
 }
-function isBridgeProcess(pid) {
-  if (process.platform === "win32") {
-    try {
-      const command = execFileSync(
-        "powershell",
-        [
-          "-NoProfile",
-          "-NonInteractive",
-          "-Command",
-          `(Get-CimInstance Win32_Process -Filter "ProcessId=${pid}").CommandLine`
-        ],
-        { encoding: "utf-8", timeout: 4e3 }
-      );
-      return commandLineLooksLikeBridge(command);
-    } catch {
-      return false;
+function isBridgeProcess(pid, expectedStartTime) {
+  const identity = inspectProcessIdentity(pid);
+  if (identity.command !== void 0) {
+    if (!commandLineLooksLikeBridge(identity.command)) return false;
+  } else {
+    if (process.platform === "win32") {
+      try {
+        const command = execFileSync(
+          "powershell",
+          [
+            "-NoProfile",
+            "-NonInteractive",
+            "-Command",
+            `(Get-CimInstance Win32_Process -Filter "ProcessId=${pid}").CommandLine`
+          ],
+          { encoding: "utf-8", timeout: 4e3 }
+        );
+        if (!commandLineLooksLikeBridge(command)) return false;
+      } catch {
+        return false;
+      }
+    } else {
+      try {
+        const command = execFileSync("ps", ["-p", String(pid), "-o", "command="], {
+          encoding: "utf-8",
+          timeout: 1e3
+        });
+        if (!commandLineLooksLikeBridge(command)) return false;
+      } catch {
+        return false;
+      }
     }
   }
-  try {
-    const command = execFileSync("ps", ["-p", String(pid), "-o", "command="], {
-      encoding: "utf-8",
-      timeout: 1e3
-    });
-    return commandLineLooksLikeBridge(command);
-  } catch {
-    return false;
-  }
+  if (expectedStartTime === void 0) return true;
+  return identity.startTime !== void 0 && identity.startTime === expectedStartTime;
 }
 async function probeBridgeIdentity(port, sessionName) {
   try {
@@ -2022,11 +2060,12 @@ async function judgeRecordedBridge(pidInfo, sessionName) {
   if (probe.identity === "own") {
     if (probe.reportedPid !== void 0) {
       if (probe.reportedPid === pidInfo.pid) {
-        const verified = isBridgeProcess(pidInfo.pid);
+        const verified = isBridgeProcess(pidInfo.pid, pidInfo.startTime);
         return {
           terminate: verified,
           killProcessGroup: verified,
-          targetPid: pidInfo.pid
+          targetPid: pidInfo.pid,
+          expectedStartTime: pidInfo.startTime
         };
       }
       const reportedIsBridge = isBridgeProcess(probe.reportedPid);
@@ -2036,14 +2075,14 @@ async function judgeRecordedBridge(pidInfo, sessionName) {
         targetPid: probe.reportedPid
       };
     }
-    const bridge2 = isBridgeProcess(pidInfo.pid);
-    return { terminate: bridge2, killProcessGroup: bridge2, targetPid: pidInfo.pid };
+    const bridge2 = isBridgeProcess(pidInfo.pid, pidInfo.startTime);
+    return { terminate: bridge2, killProcessGroup: bridge2, targetPid: pidInfo.pid, expectedStartTime: pidInfo.startTime };
   }
   if (probe.identity === "other") {
     return { terminate: false, killProcessGroup: false };
   }
-  const bridge = isBridgeProcess(pidInfo.pid);
-  return { terminate: bridge, killProcessGroup: bridge, targetPid: pidInfo.pid };
+  const bridge = isBridgeProcess(pidInfo.pid, pidInfo.startTime);
+  return { terminate: bridge, killProcessGroup: bridge, targetPid: pidInfo.pid, expectedStartTime: pidInfo.startTime };
 }
 function removePidFile(pidFile) {
   try {
@@ -2065,9 +2104,13 @@ function taskkillTree(pid) {
 async function terminateBridgeProcess(pid, opts = {}) {
   if (!isProcessAlive(pid)) return true;
   const killProcessGroup = opts.killProcessGroup === true;
+  const identityHolds = () => !opts.requireBridgeIdentity || isBridgeProcess(pid, opts.expectedStartTime);
+  const refuseIfUnverified = async () => !await isProcessAlive(pid);
   if (process.platform === "win32") {
+    if (!identityHolds()) return refuseIfUnverified();
     return taskkillTree(pid) && !await isProcessAlive(pid);
   }
+  if (!identityHolds()) return refuseIfUnverified();
   try {
     process.kill(pid, "SIGTERM");
   } catch (error) {
@@ -2082,10 +2125,12 @@ async function terminateBridgeProcess(pid, opts = {}) {
     }
     return true;
   }
+  if (!identityHolds()) return refuseIfUnverified();
   if (killProcessGroup) {
     try {
       process.kill(-pid, "SIGKILL");
     } catch {
+      if (!identityHolds()) return refuseIfUnverified();
       try {
         process.kill(pid, "SIGKILL");
       } catch {
@@ -2188,7 +2233,9 @@ async function ensureBridge(spawnBridge = spawnBridgeProcess, notice) {
     const verdict = await judgeRecordedBridge(pidInfo, sessionName);
     if (verdict.terminate) {
       await terminateBridgeProcess(verdict.targetPid ?? pidInfo.pid, {
-        killProcessGroup: verdict.killProcessGroup
+        killProcessGroup: verdict.killProcessGroup,
+        requireBridgeIdentity: true,
+        expectedStartTime: verdict.expectedStartTime
       });
     } else {
       removePidFile(pidFile);
@@ -2466,7 +2513,9 @@ async function stopBridge() {
   const verdict = await judgeRecordedBridge(pidInfo, resolveSessionName());
   if (!verdict.terminate) return "noop";
   const terminated = await terminateBridgeProcess(verdict.targetPid ?? pidInfo.pid, {
-    killProcessGroup: verdict.killProcessGroup
+    killProcessGroup: verdict.killProcessGroup,
+    requireBridgeIdentity: true,
+    expectedStartTime: verdict.expectedStartTime
   });
   clearSelectedPageId();
   return terminated ? "stopped" : "failed";

package/dist/builtin/skills/skill-creator/SKILL.md

@@ -166,9 +166,9 @@ This section is one continuous sequence — don't stop partway through. Do NOT u
 
 Put results in `<skill-name>-workspace/` as a sibling to the skill directory. Within the workspace, organize results by iteration (`iteration-1/`, `iteration-2/`, etc.) and within that, each test case gets a directory (`eval-0/`, `eval-1/`, etc.). Don't create all of this upfront — just create directories as you go.
 
-### Step 1: Spawn all runs (with-skill AND baseline) in the same turn
+### Step 1: Schedule paired runs (with-skill AND baseline)
 
-For each test case, spawn two subagents in the same turn — one with the skill, one without. This is important: don't spawn the with-skill runs first and then come back for baselines later. Launch everything at once so it all finishes around the same time.
+For each test case, prepare paired runs — one with the skill, one baseline. Read-only runs may be batched; Pie uses bounded concurrency (default 4), not unlimited simultaneous execution. Default `mode: "read"` tasks cannot write source files: ask them to return outputs and grading evidence, then have the parent persist artifacts, metadata, timing and grades. When evaluation genuinely requires file edits, use explicit `mode: "edit"` with exactly one task per call, wait for it to finish before any parent edit or next edit task, and review its changes. The workspace is shared, not isolated. Use parent-owned persistence or serialized single edit tasks for all evaluation writes; a save-path instruction does not grant a read task write permission.
 
 **With-skill run:**
 
@@ -177,7 +177,7 @@ Execute this task:
 - Skill path: <path-to-skill>
 - Task: <eval prompt>
 - Input files: <eval files if any, or "none">
-- Save outputs to: <workspace>/iteration-<N>/eval-<ID>/with_skill/outputs/
+- Return outputs for parent persistence to: <workspace>/iteration-<N>/eval-<ID>/with_skill/outputs/ (or use a serialized single edit task)
 - Outputs to save: <what the user cares about — e.g., "the .docx file", "the final CSV">
 ```
 
@@ -206,29 +206,28 @@ Update the `eval_metadata.json` files and `evals/evals.json` with the assertions
 
 ### Step 3: As runs complete, capture timing data
 
-When each subagent task completes, you receive a notification containing `total_tokens` and `duration_ms`. Save this data immediately to `timing.json` in the run directory:
+Pie exposes per-task timing as `task.details.durationMs` when available and aggregate wall time as result `details.totalDuration` (milliseconds). It does not supply `total_tokens` or `duration_ms` notification fields. Have the parent persist available per-run timing to `timing.json`, converting units for the viewer:
 
 ```json
 {
-  "total_tokens": 84852,
   "duration_ms": 23332,
   "total_duration_seconds": 23.3
 }
 ```
 
-This is the only opportunity to capture this data — it comes through the task notification and isn't persisted elsewhere. Process each notification as it arrives rather than trying to batch them.
+Here `duration_ms` is an artifact field derived from `task.details.durationMs`, not a native notification field. Keep aggregate `totalDuration` separate from per-run duration; concurrency means it is not the sum of task times. If timing is absent, mark it unavailable rather than inventing a value. Do not report token counts unless separate instrumentation actually provides them; omit `total_tokens` otherwise. Viewer/aggregator default zeros are not measured token usage.
 
 ### Step 4: Grade, aggregate, and launch the viewer
 
 Once all runs are done:
 
-1. **Grade each run** — spawn a grader subagent (or grade inline) that reads `agents/grader.md` and evaluates each assertion against the outputs. Save results to `grading.json` in each run directory. The grading.json expectations array must use the fields `text`, `passed`, and `evidence` (not `name`/`met`/`details` or other variants) — the viewer depends on these exact field names. For assertions that can be checked programmatically, write and run a script rather than eyeballing it — scripts are faster, more reliable, and can be reused across iterations.
+1. **Grade each run** — spawn a grader subagent (or grade inline) that reads `agents/grader.md` and evaluates each assertion against the outputs. Have the parent save returned grades to `grading.json` in each run directory (or use a serialized single edit task); read graders must not write files. The grading.json expectations array must use the fields `text`, `passed`, and `evidence` (not `name`/`met`/`details` or other variants) — the viewer depends on these exact field names. For assertions that can be checked programmatically, write and run a script rather than eyeballing it — scripts are faster, more reliable, and can be reused across iterations.
 
 2. **Aggregate into benchmark** — run the aggregation script from the skill-creator directory:
    ```bash
    node <skill-creator-path>/scripts/aggregate_benchmark.mjs <workspace>/iteration-N --skill-name <name>
    ```
-   This produces `benchmark.json` and `benchmark.md` with pass_rate, time, and tokens for each configuration, with mean ± stddev and the delta. If generating benchmark.json manually, see `references/schemas.md` for the exact schema the viewer expects.
+   This produces `benchmark.json` and `benchmark.md` with pass_rate and available timing for each configuration, with mean ± stddev and the delta. Token fields are meaningful only with separate instrumentation; label them unavailable otherwise. If generating benchmark.json manually, see `references/schemas.md` for the exact schema the viewer expects.
 Put each with_skill version before its baseline counterpart.
 
 3. **Do an analyst pass** — read the benchmark data and surface patterns the aggregate stats might hide. See `agents/analyzer.md` (the "Analyzing Benchmark Results" section) for what to look for — things like assertions that always pass regardless of skill (non-discriminating), high-variance evals (possibly flaky), and time/token tradeoffs.