@henryqw/pi-subagent 2.3.4 → 2.4.0

package/CONTEXT.md

@@ -18,6 +18,7 @@ Provide validated user Roles, shared task-model Pi launch policy, generic manage
 ## Invariants
 
 - One Delegated Task creates one ephemeral child process and no saved session. Its soft deadline is 10 minutes; active model/tool execution or activity within the last minute grants one 5-minute grace period before a hard stop.
+- Up to four active ephemeral `delegate_task` children run per Main; excess calls wait FIFO. Queued calls do not start a child or consume child timeout. Managed Herdr workers are unaffected.
 - Ambient child extensions and Skills stay disabled; Role explicitly selects extension sources and named Skills. Pi loads Skills supplied by those extension packages or their resource discovery. Omitted Role tools use Pi's effective `defaultTools`; an explicit list sets base tools while loaded extension tools activate automatically.
 - Role Skill names resolve through Main's effective Pi Skill registry; unavailable names warn and skip without blocking delegation.
 - Main selects Role and may override Model Class per task; omitted class uses shared `pi-subagent/delegateTask` assignment, initially `balanced`. Library callers select Role plus their own shared task ID.

package/README.md

@@ -21,11 +21,13 @@ pi install npm:@henryqw/pi-subagent
 
 | Surface | Type | Purpose |
 | --- | --- | --- |
-| `delegate_task` | tool | Start one isolated child for `role`, `task`, and optional `modelClass`. |
+| `delegate_task` | tool | Start one isolated child for `role`, `task`, and optional `model` or `modelClass`. |
+
+An explicit `model` (`provider/modelId`) overrides `modelClass` and resolves against the currently available text models; an unknown reference rejects with the list of available models. Thinking level defaults to `medium` when supported, otherwise the highest supported level.
 
 `modelClass` is `fast`, `balanced`, or `frontier`. Omitted class uses the shared `pi-subagent/delegateTask` assignment, which defaults to `balanced`. Primary route is resolved against current scoped text models; fallback is tried only before launch. If no route is usable, delegation rejects with `Run /task-models`. A started child is never retried.
 
-Main splits broad work into independent bounded tasks and keeps integration and cross-cutting decisions. Each `task` states its objective, exact scope and exclusions, relevant context and constraints, expected deliverable, and validation. Each call uses the least capable `modelClass` that can reliably complete its task. Independent sibling calls can run concurrently; concurrent edit tasks must own non-overlapping files.
+Main splits broad work into independent bounded tasks and keeps integration and cross-cutting decisions. Each `task` states its objective, exact scope and exclusions, relevant context and constraints, expected deliverable, and validation. Each call uses the least capable `modelClass` that can reliably complete its task. Independent sibling calls can run concurrently; concurrent edit tasks must own non-overlapping files. Up to four active ephemeral `delegate_task` children run per Main; excess calls wait FIFO. Queued calls do not start a child or consume child timeout. Managed Herdr workers are unaffected.
 
 Each call starts one isolated child (`pi --mode json -p --no-session`). Ambient extensions and Skills are off. Role/caller extensions load; those packages' tools and Skills auto-load, plus any extra `skills` names. Child uses the delegated working directory and Main's project approval. Abort kills the child process group. An inactive child times out after 10 minutes; current model/tool execution or activity in the last minute grants one 5-minute grace period, then the child stops. Streaming output is capped at 50 KiB. Unused JSON event types are discarded before payload buffering; consumed or unclassifiable events above 1 MiB fail delegation.
 

package/extensions/subagent.ts

@@ -4,7 +4,15 @@ import { basename } from "node:path";
 import { StringEnum } from "@earendil-works/pi-ai";
 import { type ExtensionAPI, type ExtensionContext, type Theme } from "@earendil-works/pi-coding-agent";
 import { type Component, truncateToWidth, type TUI, visibleWidth } from "@earendil-works/pi-tui";
-import { modelReference, PROFILE_NAMES, type ProfileName } from "@henryqw/pi-task-models";
+import {
+	availableTaskModels,
+	modelReference,
+	PROFILE_NAMES,
+	type ProfileName,
+	resolveAvailableModel,
+	type ResolvedTaskRoute,
+	taskThinkingLevels,
+} from "@henryqw/pi-task-models";
 import { Type } from "typebox";
 import { createRoleLaunch, isProfileName, loadRoles, resolveRoleLaunch, resolveTaskRoute } from "@henryqw/pi-subagent";
 
@@ -12,6 +20,7 @@ const MODEL_CLASSES = PROFILE_NAMES;
 const SUBAGENT_TASK = "pi-subagent/delegateTask";
 const MAX_OUTPUT_BYTES = 50 * 1024;
 const MAX_JSON_EVENT_BYTES = 1024 * 1024;
+const MAX_ACTIVE_CHILDREN = 4;
 const CONSUMED_JSON_EVENTS = new Set(["message_start", "message_update", "message_end"]);
 const JSON_EVENT_TYPE = /^\s*\{\s*"type"\s*:\s*"([^"\\]+)"/;
 const WIDGET_KEY = "subagent-status";
@@ -414,11 +423,24 @@ const Parameters = Type.Object({
 	task: Type.String({
 		description: "Bounded task packet: objective; exact scope and exclusions; relevant context and constraints; expected deliverable; validation. Never the whole parent request.",
 	}),
+	model: Type.Optional(Type.String({
+		description: "Designated model as provider/modelId; overrides modelClass. Unknown references reject with the list of available models.",
+	})),
 	modelClass: Type.Optional(StringEnum(MODEL_CLASSES, {
 		description: "Classify task complexity: fast for narrow lookups or mechanical edits; balanced for normal bounded work; frontier for ambiguous, cross-cutting, or high-risk reasoning. Defaults to the shared pi-subagent/delegateTask assignment.",
 	})),
 });
 
+function resolveDesignatedRoute(ctx: ExtensionContext, reference: string): ResolvedTaskRoute {
+	const models = availableTaskModels(ctx);
+	const model = resolveAvailableModel(models, reference, ctx.model?.provider);
+	if (!model) {
+		throw new Error(`Unknown delegate_task model: ${reference}. Available models: ${models.map((candidate) => modelReference(candidate)).join(", ") || "none"}.`);
+	}
+	const levels = taskThinkingLevels(ctx, model);
+	return { model, thinkingLevel: levels.includes("medium") ? "medium" : levels.at(-1)! };
+}
+
 const roleSummary = (): string => {
 	try {
 		const roles = loadRoles();
@@ -433,6 +455,35 @@ export default function subagentExtension(
 	timeoutPolicy: TimeoutPolicy = DEFAULT_TIMEOUT_POLICY,
 ): void {
 	const widgetItems = new Map<string, WidgetItem>();
+	let activeChildren = 0;
+	const queuedChildren: Array<() => void> = [];
+	const acquireChildPermit = (signal: AbortSignal | undefined): Promise<void> => {
+		if (signal?.aborted) return Promise.reject(new Error("Subagent was aborted."));
+		if (activeChildren < MAX_ACTIVE_CHILDREN) {
+			activeChildren++;
+			return Promise.resolve();
+		}
+		return new Promise<void>((resolve, reject) => {
+			function abort() {
+				const index = queuedChildren.indexOf(grant);
+				if (index < 0) return;
+				queuedChildren.splice(index, 1);
+				signal?.removeEventListener("abort", abort);
+				reject(new Error("Subagent was aborted."));
+			}
+			const grant = () => {
+				signal?.removeEventListener("abort", abort);
+				resolve();
+			};
+			queuedChildren.push(grant);
+			signal?.addEventListener("abort", abort, { once: true });
+		});
+	};
+	const releaseChildPermit = () => {
+		const grant = queuedChildren.shift();
+		if (grant) grant();
+		else activeChildren--;
+	};
 	let widgetInstalled = false;
 	let widgetTimer: ReturnType<typeof setInterval> | undefined;
 	let spinnerIndex = 0;
@@ -523,7 +574,7 @@ export default function subagentExtension(
 	pi.registerTool({
 		name: "delegate_task",
 		label: "Subagent",
-		description: `Delegate one bounded, independently executable task to one isolated Pi Subagent. Roles: ${roleSummary()}. Choose fast for narrow work, balanced for normal work, or frontier for ambiguous and high-risk work; omit modelClass to use shared task-model settings.`,
+		description: `Delegate one bounded, independently executable task to one isolated Pi Subagent. Roles: ${roleSummary()}. Choose fast for narrow work, balanced for normal work, or frontier for ambiguous and high-risk work; omit modelClass to use shared task-model settings. When the user designates a specific model, pass it as provider/modelId in model.`,
 		promptSnippet: "Delegate one bounded, independently executable task to an isolated role",
 		promptGuidelines: [
 			"Before calling delegate_task, split broad work into the smallest independent bounded tasks; keep integration and cross-cutting decisions in Main.",
@@ -543,9 +594,11 @@ export default function subagentExtension(
 			if (params.modelClass !== undefined && !isModelClass(params.modelClass)) {
 				throw new Error("delegate_task modelClass must be fast, balanced, or frontier.");
 			}
-			const launch = params.modelClass === undefined
-				? resolveRoleLaunch(pi, ctx, { role, taskId: SUBAGENT_TASK })
-				: createRoleLaunch(pi, ctx, { role, route: resolveTaskRoute(ctx, params.modelClass) });
+			const launch = params.model !== undefined
+				? createRoleLaunch(pi, ctx, { role, route: resolveDesignatedRoute(ctx, cleanText(params.model, "model", "delegate_task")) })
+				: params.modelClass === undefined
+					? resolveRoleLaunch(pi, ctx, { role, taskId: SUBAGENT_TASK })
+					: createRoleLaunch(pi, ctx, { role, route: resolveTaskRoute(ctx, params.modelClass) });
 			const modelReferenceValue = modelReference(launch.model);
 			const thinkingLevel = launch.thinkingLevel;
 			if (launch.missingSkills.length) {
@@ -555,9 +608,10 @@ export default function subagentExtension(
 				);
 			}
 
+			const args = ["--mode", "json", "-p", ...launch.args, `Task: ${task}`];
+			await acquireChildPermit(signal);
 			let widgetStatus: Exclude<WidgetStatus, "working"> = "failure";
 			try {
-				const args = ["--mode", "json", "-p", ...launch.args, `Task: ${task}`];
 				startWidgetItem(toolCallId, role.name, launch.model.id, thinkingLevel, task, ctx);
 				const details = { role: role.name, model: modelReferenceValue, thinkingLevel };
 				const result = await runPi(
@@ -578,6 +632,7 @@ export default function subagentExtension(
 				if (signal?.aborted && !(error instanceof SubagentTimeoutError)) widgetStatus = "aborted";
 				throw error;
 			} finally {
+				releaseChildPermit();
 				finishWidgetItem(toolCallId, widgetStatus);
 			}
 		},

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@henryqw/pi-subagent",
-  "version": "2.3.4",
+  "version": "2.4.0",
   "description": "Delegate one task to an isolated Pi role with explicit extensions and skills.",
   "keywords": [
     "pi-package",