@os-eco/overstory-cli 0.7.0 → 0.7.3

package/src/types.ts

@@ -22,6 +22,14 @@ export interface ResolvedModel {
 	env?: Record<string, string>;
 }
 
+/** Configuration for the Pi runtime's model alias expansion. */
+export interface PiRuntimeConfig {
+	/** Provider prefix for unqualified model aliases (e.g., "anthropic", "amazon-bedrock"). */
+	provider: string;
+	/** Maps short aliases (e.g., "opus") to provider-qualified model IDs. */
+	modelMap: Record<string, string>;
+}
+
 // === Task Tracker ===
 
 /** Backend for the task tracker. Defined here for use in OverstoryConfig. */
@@ -89,6 +97,14 @@ export interface OverstoryConfig {
 	runtime?: {
 		/** Default runtime adapter name (default: "claude"). */
 		default: string;
+		/**
+		 * Runtime adapter for headless one-shot AI calls (--print mode).
+		 * Used by merge/resolver.ts and watchdog/triage.ts.
+		 * Falls back to runtime.default when omitted.
+		 */
+		printCommand?: string;
+		/** Pi runtime configuration for model alias expansion. */
+		pi?: PiRuntimeConfig;
 	};
 }
 
@@ -145,6 +161,7 @@ export interface AgentSession {
 	lastActivity: string;
 	escalationLevel: number; // Progressive nudge stage: 0=warn, 1=nudge, 2=escalate, 3=terminate
 	stalledSince: string | null; // ISO timestamp when agent first entered stalled state
+	transcriptPath: string | null; // Runtime-provided transcript JSONL path (decoupled from ~/.claude/)
 }
 
 // === Agent Identity ===

package/src/watchdog/daemon.test.ts

@@ -15,11 +15,12 @@
  */
 
 import { afterEach, beforeEach, describe, expect, test } from "bun:test";
-import { mkdir, mkdtemp, rm } from "node:fs/promises";
+import { mkdir, mkdtemp } from "node:fs/promises";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
 import { createEventStore } from "../events/store.ts";
 import { createSessionStore } from "../sessions/store.ts";
+import { cleanupTempDir } from "../test-helpers.ts";
 import type { AgentSession, HealthCheck, StoredEvent } from "../types.ts";
 import { buildCompletionMessage, runDaemonTick } from "./daemon.ts";
 
@@ -75,6 +76,7 @@ function makeSession(overrides: Partial<AgentSession> = {}): AgentSession {
 		runId: null,
 		escalationLevel: 0,
 		stalledSince: null,
+		transcriptPath: null,
 		startedAt: new Date().toISOString(),
 		lastActivity: new Date().toISOString(),
 		...overrides,
@@ -162,7 +164,7 @@ beforeEach(async () => {
 });
 
 afterEach(async () => {
-	await rm(tempRoot, { recursive: true, force: true });
+	await cleanupTempDir(tempRoot);
 });
 
 describe("daemon tick", () => {
@@ -804,6 +806,7 @@ describe("daemon tick", () => {
 			pid: process.pid,
 			escalationLevel: 0,
 			stalledSince: null,
+			transcriptPath: null,
 		});
 
 		writeSessionsToStore(tempRoot, [session]);
@@ -1098,7 +1101,7 @@ describe("daemon mulch failure recording", () => {
 	});
 
 	afterEach(async () => {
-		await rm(tempRoot, { recursive: true, force: true });
+		await cleanupTempDir(tempRoot);
 	});
 
 	/** Track calls to the recordFailure mock. */
@@ -1389,7 +1392,7 @@ describe("run completion detection", () => {
 		expect(coordinatorNudges).toHaveLength(1);
 		// The test creates builders, so the message should be builder-specific
 		expect(coordinatorNudges[0]?.message).toContain("builder");
-		expect(coordinatorNudges[0]?.message).toContain("merge/cleanup");
+		expect(coordinatorNudges[0]?.message).toContain("Awaiting lead verification");
 	});
 
 	test("does not nudge when some workers still active", async () => {
@@ -1581,7 +1584,7 @@ describe("run completion detection", () => {
 		expect(coordinatorNudges).toHaveLength(1);
 		// The test creates builders, so the message should be builder-specific
 		expect(coordinatorNudges[0]?.message).toContain("builder");
-		expect(coordinatorNudges[0]?.message).toContain("merge/cleanup");
+		expect(coordinatorNudges[0]?.message).toContain("Awaiting lead verification");
 	});
 
 	test("does not nudge when no worker sessions in run", async () => {
@@ -1916,14 +1919,15 @@ describe("buildCompletionMessage", () => {
 		expect(msg).not.toContain("merge/cleanup");
 	});
 
-	test("all builders → contains 'builder' and 'Ready for merge/cleanup'", () => {
+	test("all builders → contains 'builder' and 'Awaiting lead verification' (not merge authorization)", () => {
 		const sessions = [
 			makeSession({ capability: "builder", agentName: "builder-1" }),
 			makeSession({ capability: "builder", agentName: "builder-2" }),
 		];
 		const msg = buildCompletionMessage(sessions, testRunId);
 		expect(msg).toContain("builder");
-		expect(msg).toContain("Ready for merge/cleanup");
+		expect(msg).toContain("Awaiting lead verification");
+		expect(msg).not.toContain("merge/cleanup");
 	});
 
 	test("all reviewers → contains 'reviewer' and 'Reviews done'", () => {

package/src/watchdog/daemon.ts

@@ -148,7 +148,7 @@ export function buildCompletionMessage(
 			return `[WATCHDOG] All ${count} scout(s) in run ${runId} have completed. Ready for next phase.`;
 		}
 		if (capabilities.has("builder")) {
-			return `[WATCHDOG] All ${count} builder(s) in run ${runId} have completed. Ready for merge/cleanup.`;
+			return `[WATCHDOG] All ${count} builder(s) in run ${runId} have completed. Awaiting lead verification.`;
 		}
 		if (capabilities.has("reviewer")) {
 			return `[WATCHDOG] All ${count} reviewer(s) in run ${runId} have completed. Reviews done.`;

package/src/watchdog/health.test.ts

@@ -47,6 +47,7 @@ function makeSession(overrides: Partial<AgentSession> = {}): AgentSession {
 		lastActivity: new Date().toISOString(),
 		escalationLevel: 0,
 		stalledSince: null,
+		transcriptPath: null,
 		...overrides,
 	};
 }

package/src/watchdog/triage.test.ts

@@ -7,9 +7,10 @@
  */
 
 import { afterEach, beforeEach, describe, expect, test } from "bun:test";
-import { mkdir, mkdtemp, rm } from "node:fs/promises";
+import { mkdir, mkdtemp } from "node:fs/promises";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
+import { cleanupTempDir } from "../test-helpers.ts";
 import { buildTriagePrompt, classifyResponse, triageAgent } from "./triage.ts";
 
 describe("classifyResponse", () => {
@@ -97,7 +98,7 @@ describe("triageAgent", () => {
 	});
 
 	afterEach(async () => {
-		await rm(tempRoot, { recursive: true, force: true });
+		await cleanupTempDir(tempRoot);
 	});
 
 	test("returns 'extend' when no logs directory exists", async () => {

package/src/watchdog/triage.ts

@@ -9,6 +9,8 @@
 import { readdir } from "node:fs/promises";
 import { join } from "node:path";
 import { AgentError } from "../errors.ts";
+import { getRuntime } from "../runtimes/registry.ts";
+import type { OverstoryConfig } from "../types.ts";
 
 /**
  * Triage a stalled agent by analyzing its recent log output with Claude.
@@ -30,8 +32,10 @@ export async function triageAgent(options: {
 	lastActivity: string;
 	/** Timeout in ms for the Claude subprocess. Defaults to 30_000 (30s). */
 	timeoutMs?: number;
+	/** Overstory config for runtime resolution. */
+	config?: OverstoryConfig;
 }): Promise<"retry" | "terminate" | "extend"> {
-	const { agentName, root, lastActivity, timeoutMs } = options;
+	const { agentName, root, lastActivity, timeoutMs, config } = options;
 	const logsDir = join(root, ".overstory", "logs", agentName);
 
 	let logContent: string;
@@ -45,7 +49,7 @@ export async function triageAgent(options: {
 	const prompt = buildTriagePrompt(agentName, lastActivity, logContent);
 
 	try {
-		const response = await spawnClaude(prompt, timeoutMs);
+		const response = await spawnClaude(prompt, timeoutMs, config);
 		return classifyResponse(response);
 	} catch {
 		// Claude not available — default to extend (safe fallback)
@@ -130,10 +134,16 @@ const DEFAULT_TRIAGE_TIMEOUT_MS = 30_000;
  * @returns Claude's response text
  * @throws Error if claude is not installed, the process fails, or the timeout is reached
  */
-async function spawnClaude(prompt: string, timeoutMs?: number): Promise<string> {
+async function spawnClaude(
+	prompt: string,
+	timeoutMs?: number,
+	config?: OverstoryConfig,
+): Promise<string> {
 	const timeout = timeoutMs ?? DEFAULT_TRIAGE_TIMEOUT_MS;
 
-	const proc = Bun.spawn(["claude", "--print", "-p", prompt], {
+	const runtime = getRuntime(config?.runtime?.printCommand ?? config?.runtime?.default, config);
+	const argv = runtime.buildPrintCommand(prompt);
+	const proc = Bun.spawn(argv, {
 		stdout: "pipe",
 		stderr: "pipe",
 	});