@arhen/pi-core-subagent 1.1.6 → 1.1.8

package/README.md

@@ -10,7 +10,7 @@ Built for one job: delegate work to isolated subagents **without bloating the pa
 
 ## Design principles
 
-- **No agent files required, no discovery overhead.** The leader defines every subagent inline per call — name, system prompt, toolset. Existing files (`.agents/`, `.pi/agents/`, `~/.pi/agent/agents/`) are *looked up by name only when no inline prompt is given*; nothing is ever created or scanned per request.
+- **No agent files, no discovery.** The leader defines every subagent inline per call — name, system prompt, toolset. Nothing is read from or written to disk.
 - **Two toolsets only.** Read-only (`read, grep, find, ls` — default) or write (`read, grep, find, ls, bash, edit, write` — `write: true`). No per-agent tool config surface.
 - **In-process** — children are `AgentSession`s in the same runtime. No process spawn, no context bleed.
 - **Zero parent-context injection.** No catalog, no context hook. 6 slim tools total.
@@ -26,7 +26,7 @@ pi install npm:@arhen/pi-core-subagent
 
 ## Usage — the leader invents the agents
 
-Define agents inline (never creates files). If a task has **no inline `prompt`**, the extension looks up an existing file by agent name — `.agents/<name>.md` or `.pi/agents/<name>.md` (nearest project dir), then `~/.pi/agent/agents/<name>.md` — and inherits its prompt + tools. Inline params always win over the file. Model resolution: explicit `provider/model-id` (or bare id) via the pi model registry → agent-file `model` → the parent's current model → settings default.
+Define agents inline per call — never creates or reads agent files. Model resolution: explicit `provider/model-id` (or bare id) via the pi model registry → agent-file `model` → the parent's current model → settings default.
 
 ```json
 {
@@ -84,7 +84,7 @@ Background + intercom:
 
 ### Per-task fields
 
-`agent` (name you invent, or an existing file's name — required), `task` (required), `prompt` (system prompt, optional — falls back to file lookup, then a minimal default), `write` (toolset, default read-only), plus optional `model` (`provider/model-id`), `thinking` (validated enum: `off|minimal|low|medium|high|xhigh|max`), `tools` (explicit allowlist, overrides everything), `maxRuntimeMs`, `id`. Top-level only: `background`, `notifyPerTask`, `allowIntercom`, `concurrency`.
+`agent` (name you invent — required), `task` (required), `prompt` (system prompt, optional — minimal default used), `write` (toolset, default read-only), plus optional `model` (`provider/model-id`), `thinking` (validated enum: `off|minimal|low|medium|high|xhigh|max`), `tools` (explicit allowlist), `cwd`, `maxRuntimeMs`, `id`. Top-level only: `background`, `notifyPerTask`, `allowIntercom`, `concurrency`.
 
 ### Child talk tools (when `allowIntercom: true`)
 
@@ -99,7 +99,7 @@ Background + intercom:
 
 - Parent tools: 6 schemas with short descriptions. **No catalog, no context hook** — nothing injected per request.
 - Background completion: 3-line notice. Full text only via `subagent_result`.
-- Children: isolated sessions; talk tools injected only when `allowIntercom`; each child's prompt states its own task id and its siblings' so mailbox addressing works.
+- Children: isolated sessions; talk tools injected only when `allowIntercom`; each child's prompt states its own task id and its siblings' so mailbox addressing works. Model resolution: explicit `provider/model-id` or bare id via the pi model registry → the parent's current model → settings default. Thinking levels validated against the resolved model's `thinkingLevelMap`.
 
 ## Development
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@arhen/pi-core-subagent",
-  "version": "1.1.6",
+  "version": "1.1.8",
   "type": "module",
   "description": "pi extension: fast in-process subagents with single/parallel/chain, background runs, intercom and agent-to-agent mailbox. Leader defines agents inline.",
   "license": "MIT",

package/src/index.ts

@@ -98,7 +98,7 @@ interface RunSnapshot {
 	concurrency: number;
 	/** True once the parent awaited this run — completion notices are redundant then. */
 	awaited?: boolean;
-	/** Wake the parent (queued follow-up turn) as each task completes. Default true. */
+	/** Wake the parent (queued follow-up turn) as each task completes. Default false. */
 	notifyPerTask: boolean;
 	tasks: TaskSnapshot[];
 	aggregateUsage: UsageStats;
@@ -135,7 +135,9 @@ function aggregateUsage(tasks: TaskSnapshot[]): UsageStats {
 }
 function truncateText(text: string, max = FINAL_OUTPUT_CAP): string {
 	if (Buffer.byteLength(text, "utf8") <= max) return text;
-	return `${text.slice(0, max)}\n\n[Output truncated. Full child session is available in the session file.]`;
+	let out = text.slice(0, max);
+	while (Buffer.byteLength(out, "utf8") > max) out = out.slice(0, -1); // multibyte-safe
+	return `${out}\n\n[Output truncated. Full child session is available in the session file.]`;
 }
 function getFirstText(message: AssistantMessage): string {
 	for (const part of message?.content ?? []) {
@@ -251,8 +253,6 @@ function compactLines(run: RunSnapshot): string[] {
  *   └─ ✓ reviewer · 6 tools · 44s
  * Static icons (no animation); latest activity + tool count + runtime per agent.
  */
-const WIDGET_MAX_RUNS = 3;
-const WIDGET_MAX_TASKS_PER_RUN = 4;
 const WIDGET_MAX_LINES = 10;
 
 class SubagentsWidget implements Component {
@@ -266,38 +266,39 @@ class SubagentsWidget implements Component {
 	}
 
 	render(width: number): string[] {
-		const runs = this.getRuns();
+		// ONE flat tree: every run's tasks concatenated under a single heading.
+		// Whether the model spawned N runs or one tasks[] call, the pane reads the same.
+		const runs = this.getRuns().filter((r) => r.tasks.length > 0);
 		if (runs.length === 0) return [];
-		const lines: string[] = [];
-		let renderedRuns = 0;
-		for (const run of runs.slice(0, WIDGET_MAX_RUNS)) {
-			if (run.tasks.length === 0) continue;
-			if (lines.length > 0) lines.push("");
-			const done = run.tasks.filter((t) => TERMINAL.includes(t.status)).length;
-			const active = !TERMINAL.includes(run.status);
-			const head = active ? "accent" : "dim";
-			lines.push(truncateToWidth(`${this.theme.fg(head, active ? "●" : "○")} ${this.theme.fg(head, `Subagents (${done}/${run.tasks.length})`)}`, width, "…"));
+		const total = runs.reduce((n, r) => n + r.tasks.length, 0);
+		const done = runs.reduce((n, r) => n + r.tasks.filter((t) => TERMINAL.includes(t.status)).length, 0);
+		const live = total - done;
+		const head = live > 0 ? "accent" : "dim";
+		const lines = [truncateToWidth(`${this.theme.fg(head, live > 0 ? "●" : "○")} ${this.theme.fg(head, `Subagents (${done}/${total})`)}`, width, "…")];
+		const budget = WIDGET_MAX_LINES - 1;
+		let shown = 0;
+		outer: for (const run of runs) {
 			const allDone = TERMINAL.includes(run.status);
-			const visible = run.tasks.slice(0, WIDGET_MAX_TASKS_PER_RUN);
-			visible.forEach((task, i) => {
-				const last = i === visible.length - 1 && run.tasks.length <= WIDGET_MAX_TASKS_PER_RUN;
-				const conn = this.theme.fg("dim", last ? "└─" : "├─");
+			for (const task of run.tasks) {
+				if (shown >= budget) break outer;
+				shown += 1;
 				const activity =
 					!TERMINAL.includes(task.status) && task.lastActivity
 						? `${this.theme.fg("dim", `→ ${task.lastActivity}`)} · `
 						: "";
-				// Finished run: dim everything except the agent name.
+				// Tasks of a finished run: dim everything except the agent name.
 				const line = allDone
 					? `${this.theme.fg("dim", `${statusIcon(task.status)} `)}${task.agent} ${this.theme.fg("dim", `· ${taskStatsWithUsage(task)} · ${taskTimer(task)}`)}`
 					: `${statusIcon(task.status)} ${task.agent} · ${activity}${taskStatsWithUsage(task)} · ${taskTimer(task)}`;
-				lines.push(truncateToWidth(`${conn} ${line}`, width, "…"));
-			});
-			if (run.tasks.length > WIDGET_MAX_TASKS_PER_RUN) lines.push(`${this.theme.fg("dim", "└─")} ${this.theme.fg("dim", `+${run.tasks.length - WIDGET_MAX_TASKS_PER_RUN} more tasks`)}`);
-			renderedRuns += 1;
-			if (lines.length >= WIDGET_MAX_LINES) break; // keep the editor visible
+				lines.push(truncateToWidth(`${this.theme.fg("dim", "├─")} ${line}`, width, "…"));
+			}
+		}
+		const hidden = total - shown;
+		if (hidden > 0) {
+			lines.push(`${this.theme.fg("dim", "└─")} ${this.theme.fg("dim", `+${hidden} more`)}`);
+		} else if (lines.length > 1) {
+			lines[lines.length - 1] = lines[lines.length - 1]!.replace("├─", "└─");
 		}
-		const hiddenRuns = runs.length - renderedRuns;
-		if (hiddenRuns > 0) lines.push(`${this.theme.fg("dim", "└─")} ${this.theme.fg("dim", `+${hiddenRuns} more run${hiddenRuns > 1 ? "s" : ""}`)}`);
 		return lines;
 	}
 }
@@ -460,9 +461,17 @@ class SubagentManager {
 			if (!Array.isArray(raw)) return;
 			runs = (raw as RunSnapshot[]).map((run) => {
 				const interrupted = run.tasks.some((t) => !TERMINAL.includes(t.status));
+				// A persisted "running" run whose tasks are all terminal (crash between
+				// task end and run end) must not stay "running" forever.
+				let status = interrupted ? ("aborted" as RunStatus) : run.status;
+				if (!TERMINAL.includes(status)) {
+					const anyFailed = run.tasks.some((t) => t.status === "failed");
+					const anyAborted = run.tasks.some((t) => t.status === "aborted");
+					status = anyFailed ? "failed" : anyAborted ? "aborted" : "completed";
+				}
 				return {
 					...run,
-					status: interrupted ? ("aborted" as RunStatus) : run.status,
+					status,
 					endedAt: interrupted ? Date.now() : run.endedAt,
 					tasks: run.tasks.map((t) => (TERMINAL.includes(t.status) ? t : { ...t, status: "aborted" as TaskStatus, error: t.error || "Interrupted by session reload" })),
 				};
@@ -486,7 +495,7 @@ class SubagentManager {
 			const parentFile = getParentSessionFile(ctx);
 			if (!parentFile) return;
 			const sidecar = parentFile.replace(/\.jsonl$/, ".subagents.json");
-			import("fs").then(({ writeFileSync }) => writeFileSync(sidecar, JSON.stringify(this.listRuns().map(cloneRun), null, 2)));
+			import("fs").then(({ writeFileSync }) => writeFileSync(sidecar, JSON.stringify(this.listRuns().slice(0, 50).map(cloneRun), null, 2)));
 		} catch {
 			/* ignore */
 		}
@@ -850,6 +859,13 @@ class SubagentManager {
 				? params.tasks!
 				: params.chain!;
 		if (inputs.length > MAX_TASKS) throw new Error(`Too many subagent tasks (${inputs.length}). Max is ${MAX_TASKS}.`);
+		const ids = new Set<string>();
+		for (const input of inputs) {
+			if (input.id !== undefined) {
+				if (ids.has(input.id)) throw new Error(`Duplicate task id: ${input.id}`);
+				ids.add(input.id);
+			}
+		}
 
 		const run: RunSnapshot = {
 			id: newId("run"),
@@ -909,7 +925,7 @@ class SubagentManager {
 				const input = { ...inputs[i]!, task: next };
 				task.task = input.task;
 				await this.runChild(run, task, input, ctx, signal, onUpdate);
-				if (run.notifyPerTask && TERMINAL.includes(task.status)) {
+				if (run.notifyPerTask && run.background && TERMINAL.includes(task.status)) {
 					this.notifyTask(run, task, task.status as "completed" | "failed" | "aborted");
 				}
 				if (task.status !== "completed") break;
@@ -919,7 +935,7 @@ class SubagentManager {
 			await mapWithConcurrency(run.tasks, run.mode === "single" ? 1 : run.concurrency, async (task) => {
 				const index = run.tasks.indexOf(task);
 				await this.runChild(run, task, inputs[index]!, ctx, signal, onUpdate);
-				if (run.notifyPerTask && TERMINAL.includes(task.status)) {
+				if (run.notifyPerTask && run.background && TERMINAL.includes(task.status)) {
 					this.notifyTask(run, task, task.status as "completed" | "failed" | "aborted");
 				}
 			});
@@ -969,7 +985,10 @@ class SubagentManager {
 				}
 				this.settleRun(run.id, run);
 				this.runControllers.delete(run.id);
+				for (const task of run.tasks) this.mailboxes.close(`${run.id}:${task.id}`);
+				this.emit("subagent:run-completed", { runId: run.id, status: "failed", run: cloneRun(run) });
 				this.notifyParent(run, "failed");
+				this.persist(ctx);
 			});
 		return { run: cloneRun(run), background: true };
 	}
@@ -988,7 +1007,7 @@ class SubagentManager {
 		for (const task of run.tasks) {
 			if (TERMINAL.includes(task.status)) continue;
 			task.status = "aborted";
-			task.error = task.error || "Canceled by subagent_cancel";
+			task.error = task.error || "Canceled by subagent_cancel"; // never overwrite a real error
 			task.endedAt = Date.now();
 			aborted += 1;
 		}
@@ -1026,9 +1045,10 @@ class SubagentManager {
 		if (!timeoutMs) return settled;
 		return Promise.race([
 			settled,
-			new Promise<RunSnapshot | undefined>((resolve) =>
-				setTimeout(() => resolve(this.runs.get(runId) ? cloneRun(this.runs.get(runId)!) : undefined), timeoutMs),
-			),
+			new Promise<RunSnapshot | undefined>((resolve) => {
+				const timer = setTimeout(() => resolve(this.runs.get(runId) ? cloneRun(this.runs.get(runId)!) : undefined), timeoutMs);
+				settled.then(() => clearTimeout(timer));
+			}),
 		]);
 	}
 }
@@ -1113,8 +1133,7 @@ export default function (pi: ExtensionAPI) {
 				ctx.ui.notify("No subagent runs in this session.", "info");
 				return;
 			}
-			ctx.ui.setWidget("subagents", runs.flatMap((run) => compactLines(run).concat("")), { placement: "aboveEditor" });
-			ctx.ui.notify(`Showing ${runs.length} subagent run(s).`, "info");
+			ctx.ui.notify(runs.flatMap((run) => compactLines(run).concat("")).join("\n") || "No subagent runs in this session.", "info");
 		},
 	});
 
@@ -1144,7 +1163,7 @@ export default function (pi: ExtensionAPI) {
 		promptSnippet: "Define and delegate work to specialized subagents.",
 		promptGuidelines: [
 			"Use subagent when independent review, testing, research, or parallel analysis improves quality.",
-			"Decompose parallelizable work: if the request has 2+ independent sub-tasks (separate files, separate concerns, independent research/review), delegate each to its own subagent — in ONE call with tasks[] (single shot), not multiple parallel subagent calls.",
+			"Decompose parallelizable work: if the request has 2+ independent sub-tasks (separate files, separate concerns, independent research/review), spawn N agents with a SINGLE call: subagent({ tasks: [{agent, task}, ...] }). NEVER make multiple parallel subagent calls for parallel work — one call, one run, N tasks.",
 			"If independent sub-tasks are sequential (each builds on the previous one's output), use chain mode with {previous}.",
 			"Define each subagent yourself: an invented name, a focused system prompt (prompt:), and a toolset — read-only (default) or write (write:true).",
 			"Prefer read-only subagents unless the task explicitly needs edits.",