@yusukeshib/pi-babysit 0.3.15 → 0.3.17

package/README.md

@@ -157,10 +157,12 @@ grace window (`PI_BABYSIT_REAP_AFTER`, default 120s) using the same parked-turn
 rule, so a subagent waiting on a long build is never false-killed. Give bounded
 recon/review tasks at least one cost, turn, tool-call, or token budget; omit
 budgets only for intentionally open-ended work. Optional task budgets are
-checked by the parent poller. On the first exceeded limit the worker
-is steered to stop using tools and return its best answer; if it remains active
-after `PI_BABYSIT_BUDGET_GRACE`, termination is verified before the task is
-marked budget-killed. Usage shown by check/wait is cumulative for the task.
+observed by the parent poller. They are wrap-up thresholds rather than exact
+hard caps: an already in-flight model call or parallel tool batch can overshoot
+before the next poll. On the first observed threshold the worker is steered to
+stop using tools and return its best answer; if it remains active after
+`PI_BABYSIT_BUDGET_GRACE`, termination is verified before the task is marked
+budget-killed. Usage shown by check/wait is cumulative for the task.
 
 ## Environment overrides
 
@@ -172,7 +174,7 @@ marked budget-killed. Usage shown by check/wait is cumulative for the task.
 | `PI_BABYSIT_VIEW_CMD` | bundled `format-stream.mjs` | live-attach pretty printer for subagent JSONL (`""` disables) |
 | `PI_BABYSIT_QUICK_GRACE` | `2s` | interactive process grace before a still-running command is returned as background work; use `foreground: true` to wait explicitly |
 | `PI_BABYSIT_REAP_AFTER` | `120s` | idle grace before a finished subagent self-exits (`off`/`none`/`0` disables) |
-| `PI_BABYSIT_BUDGET_GRACE` | `30s` | grace after a subagent budget is exceeded before verified termination |
+| `PI_BABYSIT_BUDGET_GRACE` | `90s` | grace after a subagent budget is exceeded before verified termination |
 | `PI_BABYSIT_RPC_LOG_MODE` | `compact` | `compact` removes duplicate RPC lifecycle payloads; `standard` opts into legacy payloads |
 | `PI_BABYSIT_RETENTION_DAYS` | unset | when set to a positive number, remove safe terminal roots older than this at session startup |
 | `PI_BABYSIT_TAIL_MAX_BYTES` | `8000` | cap for explicit log tails/screens returned by `babysit_check` |

package/index.ts

@@ -791,6 +791,33 @@ export function parseRpcResponseBytes(
 // it. Distinguishes: success, explicit failure (success:false → the error
 // message, e.g. "No API key found for …"), subagent death, and timeout — this
 // is what makes bad-model/config failures LOUD instead of silent.
+export function rpcResponsePattern(command: string): string {
+	const escaped = command.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
+	// Match only a complete response record. A loose `"command":"…"` can occur
+	// in assistant/tool text, and matching before the terminating newline can
+	// race the writer while the JSON record is still partial.
+	return `(?m)^\\{(?:"id":"[^"\\n]*",)?"type":"response","command":"${escaped}"[^\\n]*\\}\\r?\\n`;
+}
+
+export function readLogBytesFrom(file: string, since: number): Buffer {
+	const size = fs.statSync(file).size;
+	const offset = Math.min(Math.max(0, since), size);
+	const length = size - offset;
+	const bytes = Buffer.allocUnsafe(length);
+	const fd = fs.openSync(file, "r");
+	let read = 0;
+	try {
+		while (read < length) {
+			const count = fs.readSync(fd, bytes, read, length - read, offset + read);
+			if (count === 0) break;
+			read += count;
+		}
+	} finally {
+		fs.closeSync(fd);
+	}
+	return bytes.subarray(0, read);
+}
+
 async function rpcResponse(
 	id: string,
 	since: number,
@@ -802,7 +829,16 @@ async function rpcResponse(
 	| { ok: false; error: string }
 > {
 	const e = await bs(
-		["expect", "-s", id, "--since", String(since), "--timeout", timeout, `"command":"${command}"`],
+		[
+			"expect",
+			"-s",
+			id,
+			"--since",
+			String(since),
+			"--timeout",
+			timeout,
+			rpcResponsePattern(command),
+		],
 		{ signal },
 	);
 	if (e.code !== 0) {
@@ -810,8 +846,9 @@ async function rpcResponse(
 		if (st && st.state !== "running") {
 			let structuredError = "";
 			try {
-				const bytes = fs.readFileSync(logPath(id));
-				structuredError = parseEvents(bytes.subarray(Math.min(since, bytes.length)).toString("utf8")).errorMsg ?? "";
+				// Long-lived follow-up workers can have very large historical logs. Read
+				// only the response window rather than synchronously loading all history.
+				structuredError = parseEvents(readLogBytesFrom(logPath(id), since).toString("utf8")).errorMsg ?? "";
 			} catch {
 				/* full log path below remains the diagnostic source */
 			}
@@ -832,22 +869,7 @@ async function rpcResponse(
 		};
 	}
 	try {
-		const file = logPath(id);
-		const size = fs.statSync(file).size;
-		const length = Math.max(0, size - since);
-		const bytes = Buffer.allocUnsafe(length);
-		const fd = fs.openSync(file, "r");
-		let read = 0;
-		try {
-			while (read < length) {
-				const count = fs.readSync(fd, bytes, read, length - read, since + read);
-				if (count === 0) break;
-				read += count;
-			}
-		} finally {
-			fs.closeSync(fd);
-		}
-		return parseRpcResponseBytes(bytes.subarray(0, read), since, command);
+		return parseRpcResponseBytes(readLogBytesFrom(logPath(id), since), since, command);
 	} catch (error) {
 		return { ok: false, error: `could not read ${command} response: ${String(error)}` };
 	}
@@ -887,7 +909,7 @@ const NOTIFY_BATCH_MAX_BYTES = byteLimitFromEnv("PI_BABYSIT_NOTIFY_BATCH_MAX_BYT
 const ANSWER_MAX_BYTES = 24_000; // single subagent answers / structured error messages
 const MAX_MULTI_WAIT_SESSIONS = 32;
 const SUBAGENT_BUDGET_GRACE_MS =
-	parseDurMs(process.env.PI_BABYSIT_BUDGET_GRACE ?? "30s") ?? 30_000;
+	parseDurMs(process.env.PI_BABYSIT_BUDGET_GRACE ?? "90s") ?? 90_000;
 
 export function clip(s: string, maxBytes = TAIL_MAX_BYTES): string {
 	if (maxBytes <= 0) return "";
@@ -1285,6 +1307,8 @@ export interface Progress {
 	turns: number;
 	toolCalls: ToolCall[];
 	finalText: string;
+	/** Best-effort text from the currently streaming assistant message. */
+	streamingText: string;
 	/** Context size reported by the most recent assistant response. */
 	tokens?: number;
 	/** Cumulative model usage for the current subagent task. */
@@ -1339,6 +1363,7 @@ function emptyProgress(): Progress {
 		turns: 0,
 		toolCalls: [],
 		finalText: "",
+		streamingText: "",
 		modelCalls: 0,
 		usageTokens: 0,
 		inputTokens: 0,
@@ -1417,7 +1442,17 @@ function parseEventLine(progress: Progress, raw: string): void {
 	switch (event.type) {
 		case "turn_start":
 			progress.turns++;
+			progress.streamingText = "";
+			break;
+		case "message_update": {
+			const update = event.assistantMessageEvent as
+				| { type?: string; delta?: string }
+				| undefined;
+			if (update?.type === "text_delta" && typeof update.delta === "string") {
+				progress.streamingText += update.delta;
+			}
 			break;
+		}
 		case "tool_execution_start": {
 			const name = String(event.toolName ?? "tool");
 			progress.toolCalls.push({
@@ -1448,6 +1483,7 @@ function parseEventLine(progress: Progress, raw: string): void {
 					.map((content) => content.text)
 					.join("");
 				if (text.trim()) progress.finalText = text;
+				progress.streamingText = "";
 				if (message.usage) {
 					const finite = (value: number | undefined) =>
 						typeof value === "number" && Number.isFinite(value) ? value : 0;
@@ -1531,6 +1567,7 @@ export function buildSubagentExitDiagnostic(
 ): string {
 	const body = clip(
 		progress.errorMsg ||
+			progress.streamingText.trim() ||
 			progress.finalText.trim() ||
 			"(no structured error was emitted; inspect the full log)",
 		ANSWER_MAX_BYTES,
@@ -1546,6 +1583,22 @@ interface TaskProgressCache {
 }
 const taskProgressCache = new Map<string, TaskProgressCache>();
 
+export function pruneTerminalSessionCache<T>(
+	cache: Map<string, T>,
+	sessions: Array<{ id: string; state: string }>,
+): number {
+	const running = new Set(
+		sessions.filter((session) => session.state === "running").map((session) => session.id),
+	);
+	let removed = 0;
+	for (const id of cache.keys()) {
+		if (running.has(id)) continue;
+		cache.delete(id);
+		removed++;
+	}
+	return removed;
+}
+
 /** Parse only bytes appended since the previous observation of this task. */
 function taskProgressOf(id: string): { progress: Progress; offset: number } {
 	const base = readMeta(id)?.promptOffset ?? 0;
@@ -2409,16 +2462,8 @@ export default function (pi: ExtensionAPI) {
 		for (const session of sessions) {
 			if (session.state !== "running") continue;
 			await withSessionRpcLock(session.id, async () => {
-				const latestStatus = await statusOf(session.id);
 				const meta = readMeta(session.id);
-				if (
-					latestStatus?.state !== "running" ||
-					meta?.kind !== "subagent" ||
-					!meta.budget ||
-					meta.budgetKilled
-				) {
-					return;
-				}
+				if (meta?.kind !== "subagent" || !meta.budget || meta.budgetKilled) return;
 
 				let progress: Progress;
 				try {
@@ -2438,6 +2483,12 @@ export default function (pi: ExtensionAPI) {
 				if (decision.action === "none") return;
 				const reason = decision.reason as string;
 
+				// The poll already supplied a running-session snapshot. Revalidate only
+				// when a limit requires action; checking every healthy worker made each
+				// interval launch N extra `babysit list` subprocesses.
+				const latestStatus = await statusOf(session.id);
+				if (latestStatus?.state !== "running") return;
+
 				if (decision.action === "steer") {
 					// Start the grace period only after Pi accepts the steering command.
 					const sent = await sendRpc(session.id, {
@@ -2525,7 +2576,7 @@ export default function (pi: ExtensionAPI) {
 					: "failed";
 			const output = await inlineOutput(session.id, session, NOTIFY_OUTPUT_MAX_BYTES);
 			const runtime = meta.startedAt
-				? `${Math.round((Date.now() - meta.startedAt) / 1000)}s`
+				? `${Math.round(((meta.completionObservedAt ?? Date.now()) - meta.startedAt) / 1000)}s`
 				: "?";
 			const summary = ok
 				? `Process "${session.id}" completed successfully after ${runtime}.`
@@ -2545,6 +2596,8 @@ export default function (pi: ExtensionAPI) {
 			});
 		}
 
+		if (prepared.length === 0) return;
+
 		// Output loading above is asynchronous. Refresh sessions and metadata
 		// immediately before the single send so concurrent wait/kill calls and newly
 		// started notification-group members are honored.
@@ -2763,6 +2816,7 @@ export default function (pi: ExtensionAPI) {
 					notifyEndedProcesses(ctx, snapshot),
 					refreshWidget(ctx, snapshot),
 				]);
+				pruneTerminalSessionCache(taskProgressCache, snapshot);
 				pollNeeded = shouldKeepPolling(snapshot, readMeta);
 			})()
 				.catch(() => {
@@ -2896,15 +2950,24 @@ export default function (pi: ExtensionAPI) {
 				}),
 			),
 			maxTurns: Type.Optional(
-				Type.Integer({ minimum: 1, description: "Subagent only: maximum turns per task." }),
+				Type.Integer({
+					minimum: 1,
+					description:
+						"Subagent only: observed turn threshold for steering it to wrap up. In-flight work can overshoot before the poller intervenes.",
+				}),
 			),
 			maxToolCalls: Type.Optional(
-				Type.Integer({ minimum: 1, description: "Subagent only: maximum tool calls per task." }),
+				Type.Integer({
+					minimum: 1,
+					description:
+						"Subagent only: observed tool-call threshold for steering it to wrap up. A parallel in-flight tool batch can overshoot.",
+				}),
 			),
 			maxUsageTokens: Type.Optional(
 				Type.Integer({
 					minimum: 1,
-					description: "Subagent only: maximum cumulative reported totalTokens across model calls.",
+					description:
+						"Subagent only: observed cumulative reported totalTokens threshold for steering it to wrap up.",
 				}),
 			),
 			agentScope: Type.Optional(
@@ -3436,7 +3499,7 @@ export default function (pi: ExtensionAPI) {
 					if (el) header += ` elapsed=${el}`;
 				}
 				if (st.exit_code != null) header += ` exit_code=${st.exit_code}`;
-				if (meta?.command) header += `\ncommand: ${meta.command}`;
+				if (meta?.command) header += `\ncommand: ${summarizeNotificationCommand(meta.command)}`;
 				header += `\nlog: ${logPath(params.id)}`;
 				if (st.note) header += ` ⚑ ${st.note}`;
 				parts.push(header);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@yusukeshib/pi-babysit",
-  "version": "0.3.15",
+  "version": "0.3.17",
   "description": "Run any shell command and pi subagents under babysit, with context-safe captured output.",
   "keywords": [
     "pi-package",