pi-crew 0.9.53 → 0.9.55

package/src/runtime/agent-control.ts

@@ -0,0 +1,215 @@
+import { randomUUID } from "node:crypto";
+import type { PiTeamsConfig } from "../config/config.ts";
+import type { ControlReservation, TeamRunManifest } from "../state/types.ts";
+import { appendTaskAttentionEvent } from "./attention-events.ts";
+import { upsertCrewAgent } from "./crew-agent-records.ts";
+import type { CrewAgentRecord } from "./crew-agent-runtime.ts";
+
+export interface CrewControlConfig {
+	enabled: boolean;
+	needsAttentionAfterMs: number;
+	consecutiveFailureThreshold: number;
+	longRunningMinutes: number;
+}
+
+const DEFAULT_NEEDS_ATTENTION_MS = 60_000;
+const DEFAULT_CONSECUTIVE_FAILURE_THRESHOLD = 3;
+const DEFAULT_LONG_RUNNING_MINUTES = 10;
+
+function positiveInt(value: unknown): number | undefined {
+	return typeof value === "number" && Number.isInteger(value) && value > 0 ? value : undefined;
+}
+
+export function resolveCrewControlConfig(config: PiTeamsConfig | undefined): CrewControlConfig {
+	const raw = config as
+		| (PiTeamsConfig & {
+				control?: {
+					enabled?: unknown;
+					needsAttentionAfterMs?: unknown;
+					consecutiveFailureThreshold?: unknown;
+					longRunningMinutes?: unknown;
+				};
+		  })
+		| undefined;
+	return {
+		enabled: raw?.control?.enabled === false ? false : true,
+		needsAttentionAfterMs: positiveInt(raw?.control?.needsAttentionAfterMs) ?? DEFAULT_NEEDS_ATTENTION_MS,
+		consecutiveFailureThreshold: positiveInt(raw?.control?.consecutiveFailureThreshold) ?? DEFAULT_CONSECUTIVE_FAILURE_THRESHOLD,
+		longRunningMinutes: positiveInt(raw?.control?.longRunningMinutes) ?? DEFAULT_LONG_RUNNING_MINUTES,
+	};
+}
+
+export function activityAgeMs(agent: CrewAgentRecord, now = Date.now()): number | undefined {
+	const timestamp = agent.progress?.lastActivityAt ?? agent.startedAt;
+	if (!timestamp) return undefined;
+	const ms = now - new Date(timestamp).getTime();
+	return Number.isFinite(ms) ? Math.max(0, ms) : undefined;
+}
+
+export function formatActivityAge(agent: CrewAgentRecord, now = Date.now()): string | undefined {
+	const age = activityAgeMs(agent, now);
+	if (age === undefined) return undefined;
+	if (age < 1000) return "active now";
+	const seconds = Math.floor(age / 1000);
+	if (seconds < 60) return agent.progress?.activityState === "needs_attention" ? `no activity for ${seconds}s` : `active ${seconds}s ago`;
+	const minutes = Math.floor(seconds / 60);
+	return agent.progress?.activityState === "needs_attention" ? `no activity for ${minutes}m` : `active ${minutes}m ago`;
+}
+
+export function applyAttentionState(
+	manifest: TeamRunManifest,
+	agent: CrewAgentRecord,
+	config: CrewControlConfig,
+	now = Date.now(),
+): CrewAgentRecord {
+	if (!config.enabled || agent.status !== "running") return agent;
+	const age = activityAgeMs(agent, now);
+	if (age === undefined || age <= config.needsAttentionAfterMs) return agent;
+	if (agent.progress?.activityState === "needs_attention") return agent;
+	const updated: CrewAgentRecord = {
+		...agent,
+		progress: {
+			...(agent.progress ?? {
+				recentTools: [],
+				recentOutput: [],
+				toolCount: agent.toolUses ?? 0,
+			}),
+			activityState: "needs_attention",
+		},
+	};
+	upsertCrewAgent(manifest, updated);
+	appendTaskAttentionEvent({
+		manifest,
+		taskId: agent.taskId,
+		message: `${agent.agent} needs attention (no observed activity for ${Math.floor(age / 1000)}s).`,
+		data: {
+			activityState: "needs_attention",
+			reason: "idle",
+			elapsedMs: age,
+			taskId: agent.taskId,
+			agentName: agent.agent,
+			suggestedAction: "Check worker status, wait, steer, or cancel if needed.",
+		},
+	});
+	return updated;
+}
+
+/** @internal */
+function applyLongRunningCheck(
+	manifest: TeamRunManifest,
+	agent: CrewAgentRecord,
+	config: CrewControlConfig,
+	now = Date.now(),
+): CrewAgentRecord {
+	if (!config.enabled || agent.status !== "running") return agent;
+	if (agent.progress?.activityState === "needs_attention") return agent;
+
+	const startedAt = agent.startedAt ? new Date(agent.startedAt).getTime() : undefined;
+	if (!startedAt) return agent;
+
+	const runtimeMs = now - startedAt;
+	const thresholdMs = config.longRunningMinutes * 60 * 1000;
+	if (runtimeMs <= thresholdMs) return agent;
+
+	// Already flagged as long_running
+	if (agent.progress?.activityState === "active_long_running") return agent;
+
+	const updated: CrewAgentRecord = {
+		...agent,
+		progress: {
+			...(agent.progress ?? {
+				recentTools: [],
+				recentOutput: [],
+				toolCount: agent.toolUses ?? 0,
+			}),
+			activityState: "active_long_running",
+		},
+	};
+	upsertCrewAgent(manifest, updated);
+	appendTaskAttentionEvent({
+		manifest,
+		taskId: agent.taskId,
+		message: `${agent.agent} has been running for ${Math.floor(runtimeMs / 60000)}m (threshold: ${config.longRunningMinutes}m).`,
+		data: {
+			activityState: "active_long_running",
+			reason: "idle",
+			elapsedMs: runtimeMs,
+			taskId: agent.taskId,
+			agentName: agent.agent,
+			suggestedAction: "Check worker progress, steer, or cancel if needed.",
+		},
+	});
+	return updated;
+}
+
+/** @internal */
+function trackConsecutiveToolFailure(
+	manifest: TeamRunManifest,
+	agent: CrewAgentRecord,
+	toolName: string,
+	error: string | undefined,
+	config: CrewControlConfig,
+): CrewAgentRecord {
+	if (!config.enabled || agent.status !== "running") return agent;
+
+	const failures = agent.progress?.consecutiveFailures ?? 0;
+	const newFailures = failures + 1;
+
+	const updated: CrewAgentRecord = {
+		...agent,
+		progress: {
+			...(agent.progress ?? {
+				recentTools: [],
+				recentOutput: [],
+				toolCount: agent.toolUses ?? 0,
+			}),
+			consecutiveFailures: newFailures,
+		},
+	};
+
+	if (newFailures >= config.consecutiveFailureThreshold) {
+		upsertCrewAgent(manifest, updated);
+		appendTaskAttentionEvent({
+			manifest,
+			taskId: agent.taskId,
+			message: `${agent.agent} has ${newFailures} consecutive tool failures (threshold: ${config.consecutiveFailureThreshold}). Last: ${toolName}${error ? ` - ${error.slice(0, 100)}` : ""}`,
+			data: {
+				activityState: "needs_attention",
+				reason: "tool_failures",
+				taskId: agent.taskId,
+				agentName: agent.agent,
+				suggestedAction: "Investigate tool failures, steer, or cancel if needed.",
+			},
+		});
+	} else {
+		upsertCrewAgent(manifest, updated);
+	}
+
+	return updated;
+}
+
+/** @internal */
+function resetConsecutiveToolFailures(manifest: TeamRunManifest, agent: CrewAgentRecord): void {
+	if (!agent.progress?.consecutiveFailures) return;
+	const updated: CrewAgentRecord = {
+		...agent,
+		progress: {
+			...agent.progress,
+			consecutiveFailures: 0,
+		},
+	};
+	upsertCrewAgent(manifest, updated);
+}
+
+/**
+ * Reserve a control channel for a task before spawning its worker.
+ * This ensures cancel/steer requests can be queued immediately
+ * while the worker is still starting up.
+ */
+export function reserveControlChannel(taskId: string, runId: string): ControlReservation {
+	return {
+		reservedAt: new Date().toISOString(),
+		controllerId: `ctrl:${taskId}:${randomUUID()}`,
+		acceptsControlEvents: true,
+	};
+}

package/src/runtime/agent-memory.ts

@@ -0,0 +1,81 @@
+import * as fs from "node:fs";
+import * as os from "node:os";
+import * as path from "node:path";
+
+export type AgentMemoryScope = "user" | "project" | "local";
+const MAX_MEMORY_LINES = 200;
+
+export function isUnsafeMemoryName(name: string): boolean {
+	return !name || name.length > 128 || !/^[a-zA-Z0-9][a-zA-Z0-9._-]*$/.test(name);
+}
+
+export function isSymlink(filePath: string): boolean {
+	try {
+		return fs.lstatSync(filePath).isSymbolicLink();
+	} catch {
+		return false;
+	}
+}
+
+export function safeReadMemoryFile(filePath: string): string | undefined {
+	if (!fs.existsSync(filePath) || isSymlink(filePath)) return undefined;
+	try {
+		return fs.readFileSync(filePath, "utf-8");
+	} catch {
+		return undefined;
+	}
+}
+
+export function resolveMemoryDir(agentName: string, scope: AgentMemoryScope, cwd: string): string {
+	if (isUnsafeMemoryName(agentName)) throw new Error(`Unsafe agent name for memory directory: ${agentName}`);
+	if (scope === "user") return path.join(os.homedir(), ".pi", "agent-memory", agentName);
+	if (scope === "project") return path.join(cwd, ".pi", "agent-memory", agentName);
+	return path.join(cwd, ".pi", "agent-memory-local", agentName);
+}
+
+export function ensureMemoryDir(memoryDir: string): void {
+	if (fs.existsSync(memoryDir)) {
+		if (isSymlink(memoryDir)) throw new Error(`Refusing to use symlinked memory directory: ${memoryDir}`);
+		return;
+	}
+	fs.mkdirSync(memoryDir, { recursive: true });
+}
+
+export function readMemoryIndex(memoryDir: string): string | undefined {
+	if (isSymlink(memoryDir)) return undefined;
+	const memPath = path.join(memoryDir, "MEMORY.md");
+	const content = safeReadMemoryFile(memPath);
+	if (content === undefined) return undefined;
+	const lines = content.split(/\r?\n/);
+	return lines.length > MAX_MEMORY_LINES
+		? `${lines.slice(0, MAX_MEMORY_LINES).join("\n")}\n... (truncated at 200 lines). Full file: ${memPath} — use the \`read\` tool if you need entries beyond the head.`
+		: content;
+}
+
+export function buildMemoryBlock(agentName: string, scope: AgentMemoryScope, cwd: string, writable: boolean): string {
+	const memoryDir = resolveMemoryDir(agentName, scope, cwd);
+	if (writable) ensureMemoryDir(memoryDir);
+	const existing = readMemoryIndex(memoryDir);
+	const mode = writable ? "read-write" : "read-only";
+	return [
+		`# Agent Memory (${mode})`,
+		`Memory scope: ${scope}`,
+		`Memory directory: ${memoryDir}`,
+		writable
+			? "Use this persistent directory to maintain useful long-term notes for this agent."
+			: "You may reference existing memory, but do not create or modify memory files.",
+		"",
+		existing ? `## Current MEMORY.md\n${existing}` : "No MEMORY.md exists yet.",
+		writable
+			? [
+					"",
+					"## Memory Instructions",
+					"- Keep MEMORY.md concise (under 200 lines); store details in separate linked files.",
+					"- Reject stale memories; update or remove outdated notes.",
+					"- Use safe relative filenames inside the memory directory only.",
+				].join("\n")
+			: "",
+	]
+		.filter(Boolean)
+		.join("\n");
+}

package/src/runtime/agent-observability.ts

@@ -0,0 +1,122 @@
+import * as fs from "node:fs";
+import type { TeamRunManifest } from "../state/types.ts";
+import { agentOutputPath, readCrewAgents } from "./crew-agent-records.ts";
+import type { CrewAgentRecord } from "./crew-agent-runtime.ts";
+
+const TOOL_LABELS: Record<string, string> = {
+	read: "reading",
+	bash: "running command",
+	edit: "editing",
+	write: "writing",
+	grep: "searching",
+	find: "finding files",
+	ls: "listing",
+};
+
+export interface TextTailResult {
+	path: string;
+	text: string;
+	bytes: number;
+	truncated: boolean;
+}
+
+export function readTextTail(filePath: string, maxBytes = 64_000): TextTailResult {
+	if (!fs.existsSync(filePath)) return { path: filePath, text: "", bytes: 0, truncated: false };
+	const stat = fs.statSync(filePath);
+	const bytesToRead = Math.min(stat.size, Math.max(0, maxBytes));
+	const fd = fs.openSync(filePath, "r");
+	try {
+		const buffer = Buffer.alloc(bytesToRead);
+		fs.readSync(fd, buffer, 0, bytesToRead, stat.size - bytesToRead);
+		return {
+			path: filePath,
+			text: buffer.toString("utf-8"),
+			bytes: stat.size,
+			truncated: stat.size > bytesToRead,
+		};
+	} finally {
+		fs.closeSync(fd);
+	}
+}
+
+function compactDuration(ms: number | undefined): string | undefined {
+	if (ms === undefined || !Number.isFinite(ms)) return undefined;
+	if (ms < 1000) return `${Math.round(ms)}ms`;
+	if (ms < 60_000) return `${(ms / 1000).toFixed(1)}s`;
+	return `${Math.floor(ms / 60_000)}m${Math.floor((ms % 60_000) / 1000)}s`;
+}
+
+function ageBetween(start: string | undefined, end: string | undefined): string | undefined {
+	if (!start) return undefined;
+	const stop = end ? new Date(end).getTime() : Date.now();
+	const ms = Math.max(0, stop - new Date(start).getTime());
+	return compactDuration(ms);
+}
+
+function activityText(agent: CrewAgentRecord): string {
+	const parts: string[] = [];
+	if (agent.progress?.activityState) parts.push(agent.progress.activityState);
+	if (agent.progress?.currentTool) parts.push(TOOL_LABELS[agent.progress.currentTool] ?? `tool=${agent.progress.currentTool}`);
+	if (agent.toolUses !== undefined) parts.push(`tools=${agent.toolUses}`);
+	if (agent.progress?.tokens !== undefined) parts.push(`tokens=${agent.progress.tokens}`);
+	if (agent.progress?.turns !== undefined) parts.push(`turns=${agent.progress.turns}`);
+	const duration = compactDuration(agent.progress?.durationMs) ?? ageBetween(agent.startedAt, agent.completedAt);
+	if (duration) parts.push(duration);
+	if (agent.progress?.failedTool) parts.push(`failedTool=${agent.progress.failedTool}`);
+	if (agent.progress?.recentOutput?.length) parts.push(`last=${agent.progress.recentOutput.at(-1)}`);
+	return parts.join(" ") || "idle";
+}
+
+function statusGlyph(status: CrewAgentRecord["status"]): string {
+	if (status === "completed") return "✓";
+	if (status === "failed") return "✗";
+	if (status === "running") return "▶";
+	if (status === "cancelled" || status === "stopped") return "■";
+	return "·";
+}
+
+function outputWarning(manifest: TeamRunManifest, agent: CrewAgentRecord): string {
+	if (agent.status !== "completed") return "";
+	try {
+		const outputPath = agentOutputPath(manifest, agent.taskId);
+		if (!fs.existsSync(outputPath)) return " no-output";
+		return fs.statSync(outputPath).size === 0 ? " no-output" : "";
+	} catch {
+		return " no-output";
+	}
+}
+
+function agentLine(manifest: TeamRunManifest, agent: CrewAgentRecord): string {
+	return `- ${statusGlyph(agent.status)} ${agent.taskId} ${agent.role} → ${agent.agent} · ${agent.status} · ${agent.runtime} · ${activityText(agent)}${outputWarning(manifest, agent)}${agent.error ? ` · error=${agent.error}` : ""}`;
+}
+
+export function buildAgentDashboard(manifest: TeamRunManifest): {
+	text: string;
+	groups: Record<string, CrewAgentRecord[]>;
+} {
+	const agents = readCrewAgents(manifest);
+	const groups: Record<string, CrewAgentRecord[]> = {
+		running: agents.filter((agent) => agent.status === "running"),
+		queued: agents.filter((agent) => agent.status === "queued"),
+		recent: agents.filter((agent) => agent.status !== "running" && agent.status !== "queued"),
+	};
+	const lines = [
+		`Crew agents for ${manifest.runId}`,
+		`Run: ${manifest.status} · ${manifest.team}/${manifest.workflow ?? "none"} · agents=${agents.length}`,
+		`Counts: running=${groups.running.length}, queued=${groups.queued.length}, recent=${groups.recent.length}`,
+		"",
+		"## Running",
+		...(groups.running.length ? groups.running.map((agent) => agentLine(manifest, agent)) : ["- (none)"]),
+		"",
+		"## Queued",
+		...(groups.queued.length ? groups.queued.map((agent) => agentLine(manifest, agent)) : ["- (none)"]),
+		"",
+		"## Recent",
+		...(groups.recent.length ? groups.recent.slice(-10).map((agent) => agentLine(manifest, agent)) : ["- (none)"]),
+	];
+	return { text: lines.join("\n"), groups };
+}
+
+export function readAgentOutput(manifest: TeamRunManifest, taskId: string, maxBytes?: number): TextTailResult {
+	return readTextTail(agentOutputPath(manifest, taskId), maxBytes);
+}