gsd-pi 2.36.0 → 2.37.0-dev.68605cd

package/src/resources/extensions/gsd/tests/auto-loop.test.ts

@@ -14,6 +14,7 @@ import {
   type AgentEndEvent,
   type LoopDeps,
 } from "../auto-loop.js";
+import type { SessionLockStatus } from "../session-lock.js";
 
 // ─── Helpers ─────────────────────────────────────────────────────────────────
 
@@ -317,6 +318,8 @@ function makeMockDeps(
     },
     clearUnitTimeout: () => {},
     updateProgressWidget: () => {},
+    syncCmuxSidebar: () => {},
+    logCmuxEvent: () => {},
     invalidateAllCaches: () => {
       callLog.push("invalidateAllCaches");
     },
@@ -339,7 +342,7 @@ function makeMockDeps(
     preDispatchHealthGate: async () => ({ proceed: true, fixesApplied: [] }),
     syncProjectRootToWorktree: () => {},
     checkResourcesStale: () => null,
-    validateSessionLock: () => true,
+    validateSessionLock: () => ({ valid: true } as SessionLockStatus),
     updateSessionLock: () => {
       callLog.push("updateSessionLock");
     },
@@ -530,6 +533,41 @@ test("autoLoop exits on terminal complete state", async (t) => {
   );
 });
 
+test("autoLoop passes structured session-lock failure details to the handler", async () => {
+  _resetPendingResolve();
+
+  const ctx = makeMockCtx();
+  ctx.ui.setStatus = () => {};
+  const pi = makeMockPi();
+  const s = makeLoopSession();
+  let observedLockStatus: SessionLockStatus | undefined;
+
+  const deps = makeMockDeps({
+    validateSessionLock: () =>
+      ({
+        valid: false,
+        failureReason: "compromised",
+        expectedPid: process.pid,
+      }) as SessionLockStatus,
+    handleLostSessionLock: (_ctx, lockStatus) => {
+      observedLockStatus = lockStatus;
+      deps.callLog.push("handleLostSessionLock");
+    },
+  });
+
+  await autoLoop(ctx, pi, s, deps);
+
+  assert.deepEqual(observedLockStatus, {
+    valid: false,
+    failureReason: "compromised",
+    expectedPid: process.pid,
+  });
+  assert.ok(
+    !deps.callLog.includes("resolveDispatch"),
+    "should stop before dispatch after lock validation fails",
+  );
+});
+
 test("autoLoop exits on terminal blocked state", async (t) => {
   _resetPendingResolve();
 

package/src/resources/extensions/gsd/tests/auto-worktree.test.ts

@@ -153,6 +153,25 @@ async function main(): Promise<void> {
     // After teardown, originalBase should be null
     assertEq(getAutoWorktreeOriginalBase(), null, "no split-brain: originalBase cleared");
 
+    // ─── #1526: getMainBranch returns milestone branch in auto-worktree ──
+    console.log("\n=== #1526: getMainBranch() returns milestone/<MID> in auto-worktree ===");
+    {
+      const { GitServiceImpl } = await import("../git-service.ts");
+
+      // Create worktree
+      const wtPath = createAutoWorktree(tempDir, "M005");
+      // Don't set main_branch pref so getMainBranch falls through to worktree detection
+      const gitService = new GitServiceImpl(wtPath);
+      gitService.setMilestoneId("M005");
+
+      // Verify getMainBranch returns the milestone branch
+      const mainBranch = gitService.getMainBranch();
+      assertEq(mainBranch, "milestone/M005", "getMainBranch returns milestone/<MID> in auto-worktree");
+
+      // Cleanup
+      teardownAutoWorktree(tempDir, "M005");
+    }
+
     // ─── #778: reconcile plan checkboxes on re-attach ─────────────────
     console.log("\n=== #778: reconcile plan checkboxes on re-attach ===");
     {

package/src/resources/extensions/gsd/tests/cmux.test.ts

@@ -0,0 +1,122 @@
+import test, { describe } from "node:test";
+import assert from "node:assert/strict";
+import * as fs from "node:fs";
+import * as path from "node:path";
+import { fileURLToPath } from "node:url";
+import {
+  buildCmuxProgress,
+  buildCmuxStatusLabel,
+  detectCmuxEnvironment,
+  markCmuxPromptShown,
+  resetCmuxPromptState,
+  resolveCmuxConfig,
+  shouldPromptToEnableCmux,
+} from "../../cmux/index.ts";
+import type { GSDState } from "../types.ts";
+
+test("detectCmuxEnvironment requires workspace, surface, and socket", () => {
+  const detected = detectCmuxEnvironment(
+    {
+      CMUX_WORKSPACE_ID: "workspace:1",
+      CMUX_SURFACE_ID: "surface:2",
+      CMUX_SOCKET_PATH: "/tmp/cmux.sock",
+    },
+    (path) => path === "/tmp/cmux.sock",
+    () => true,
+  );
+  assert.equal(detected.available, true);
+  assert.equal(detected.cliAvailable, true);
+});
+
+test("resolveCmuxConfig enables only when preference and environment are both active", () => {
+  const config = resolveCmuxConfig(
+    { cmux: { enabled: true, notifications: true, sidebar: true, splits: true } },
+    {
+      CMUX_WORKSPACE_ID: "workspace:1",
+      CMUX_SURFACE_ID: "surface:2",
+      CMUX_SOCKET_PATH: "/tmp/cmux.sock",
+    },
+    () => true,
+    () => true,
+  );
+  assert.equal(config.enabled, true);
+  assert.equal(config.notifications, true);
+  assert.equal(config.sidebar, true);
+  assert.equal(config.splits, true);
+});
+
+test("shouldPromptToEnableCmux only prompts once per session", () => {
+  resetCmuxPromptState();
+  assert.equal(shouldPromptToEnableCmux({}, {}, () => false, () => true), false);
+
+  assert.equal(
+    shouldPromptToEnableCmux(
+      {},
+      {
+        CMUX_WORKSPACE_ID: "workspace:1",
+        CMUX_SURFACE_ID: "surface:2",
+        CMUX_SOCKET_PATH: "/tmp/cmux.sock",
+      },
+      () => true,
+      () => true,
+    ),
+    true,
+  );
+  markCmuxPromptShown();
+  assert.equal(
+    shouldPromptToEnableCmux(
+      {},
+      {
+        CMUX_WORKSPACE_ID: "workspace:1",
+        CMUX_SURFACE_ID: "surface:2",
+        CMUX_SOCKET_PATH: "/tmp/cmux.sock",
+      },
+      () => true,
+      () => true,
+    ),
+    false,
+  );
+  resetCmuxPromptState();
+});
+
+test("buildCmuxStatusLabel and progress prefer deepest active unit", () => {
+  const state: GSDState = {
+    activeMilestone: { id: "M001", title: "Milestone" },
+    activeSlice: { id: "S02", title: "Slice" },
+    activeTask: { id: "T03", title: "Task" },
+    phase: "executing",
+    recentDecisions: [],
+    blockers: [],
+    nextAction: "Keep going",
+    registry: [],
+    progress: {
+      milestones: { done: 0, total: 1 },
+      slices: { done: 1, total: 3 },
+      tasks: { done: 2, total: 5 },
+    },
+  };
+
+  assert.equal(buildCmuxStatusLabel(state), "M001 S02/T03 · executing");
+  assert.deepEqual(buildCmuxProgress(state), { value: 0.4, label: "2/5 tasks" });
+});
+
+describe("cmux extension discovery opt-out", () => {
+  test("cmux directory has package.json with pi manifest to prevent auto-discovery as extension", () => {
+    const cmuxDir = path.resolve(
+      path.dirname(fileURLToPath(import.meta.url)),
+      "../../cmux",
+    );
+    const pkgPath = path.join(cmuxDir, "package.json");
+    assert.ok(fs.existsSync(pkgPath), `${pkgPath} must exist`);
+
+    const pkg = JSON.parse(fs.readFileSync(pkgPath, "utf-8"));
+    assert.ok(
+      pkg.pi !== undefined && typeof pkg.pi === "object",
+      'package.json must have a "pi" field to opt out of extension auto-discovery',
+    );
+    assert.ok(
+      !pkg.pi.extensions?.length,
+      "pi.extensions must be empty or absent — cmux is a library, not an extension",
+    );
+  });
+});

package/src/resources/extensions/gsd/tests/preferences.test.ts

@@ -171,6 +171,29 @@ test("notification fields validate correctly", () => {
   assert.equal(preferences.notifications?.on_complete, false);
 });
 
+test("cmux fields validate correctly", () => {
+  const { preferences, errors } = validatePreferences({
+    cmux: {
+      enabled: true,
+      notifications: true,
+      sidebar: false,
+      splits: true,
+      browser: false,
+    },
+  });
+  assert.equal(errors.length, 0);
+  assert.equal(preferences.cmux?.enabled, true);
+  assert.equal(preferences.cmux?.sidebar, false);
+  assert.equal(preferences.cmux?.splits, true);
+});
+
+test("cmux unknown keys produce warnings", () => {
+  const { warnings } = validatePreferences({
+    cmux: { enabled: true, strange_mode: true } as any,
+  });
+  assert.ok(warnings.some((warning) => warning.includes('unknown cmux key "strange_mode"')));
+});
+
 test("git fields comprehensive validation", () => {
   const { preferences, errors } = validatePreferences({
     git: {

package/src/resources/extensions/gsd/tests/session-lock-regression.test.ts

@@ -17,6 +17,7 @@ import { tmpdir } from 'node:os';
 
 import {
   acquireSessionLock,
+  getSessionLockStatus,
   validateSessionLock,
   releaseSessionLock,
   readSessionLockData,
@@ -201,6 +202,50 @@ async function main(): Promise<void> {
     }
   }
 
+  // ─── 7b. getSessionLockStatus with missing metadata → reason surfaced ──
+  console.log('\n=== 7b. missing lock metadata → structured reason ===');
+  {
+    const base = mkdtempSync(join(tmpdir(), 'gsd-session-lock-'));
+    mkdirSync(join(base, '.gsd'), { recursive: true });
+
+    try {
+      const status = getSessionLockStatus(base);
+      assertEq(status.valid, false, 'missing lock metadata is invalid');
+      assertEq(status.failureReason, 'missing-metadata', 'missing metadata reason is surfaced');
+      assertEq(status.expectedPid, process.pid, 'expected PID is included');
+    } finally {
+      rmSync(base, { recursive: true, force: true });
+    }
+  }
+
+  // ─── 7c. getSessionLockStatus with foreign PID → reason surfaced ───────
+  console.log('\n=== 7c. foreign PID in lock file → structured reason ===');
+  {
+    const base = mkdtempSync(join(tmpdir(), 'gsd-session-lock-'));
+    mkdirSync(join(base, '.gsd'), { recursive: true });
+
+    try {
+      const foreignPid = process.pid + 1000;
+      const lockFile = join(gsdRoot(base), 'auto.lock');
+      writeFileSync(lockFile, JSON.stringify({
+        pid: foreignPid,
+        startedAt: new Date().toISOString(),
+        unitType: 'execute-task',
+        unitId: 'M001/S01/T01',
+        unitStartedAt: new Date().toISOString(),
+        completedUnits: 0,
+      }, null, 2));
+
+      const status = getSessionLockStatus(base);
+      assertEq(status.valid, false, 'foreign PID lock is invalid');
+      assertEq(status.failureReason, 'pid-mismatch', 'PID mismatch reason is surfaced');
+      assertEq(status.existingPid, foreignPid, 'existing PID is included');
+      assertEq(status.expectedPid, process.pid, 'expected PID is included');
+    } finally {
+      rmSync(base, { recursive: true, force: true });
+    }
+  }
+
   // ─── 8. Acquire after release is possible ─────────────────────────────
   console.log('\n=== 8. acquire after release → re-acquirable ===');
   {

package/src/resources/extensions/search-the-web/native-search.ts

@@ -16,6 +16,16 @@ export const CUSTOM_SEARCH_TOOL_NAMES = ["search-the-web", "search_and_read", "g
 /** Thinking block types that require signature validation by the API */
 const THINKING_TYPES = new Set(["thinking", "redacted_thinking"]);
 
+/**
+ * Maximum number of native web searches allowed per session (agent unit).
+ * The Anthropic API's `max_uses` is per-request — it resets on each API call.
+ * When `pause_turn` triggers a resubmit, the model gets a fresh budget.
+ * This session-level cap prevents unbounded search accumulation (#1309).
+ *
+ * 15 = 3 full turns of 5 searches each — generous for research, but bounded.
+ */
+export const MAX_NATIVE_SEARCHES_PER_SESSION = 15;
+
 /** When true, skip native web search injection and keep Brave/custom tools active on Anthropic. */
 export function preferBraveSearch(): boolean {
   // preferences.md takes priority over env var
@@ -74,6 +84,11 @@ export function registerNativeSearchHooks(pi: NativeSearchPI): { getIsAnthropic:
   let isAnthropicProvider = false;
   let modelSelectFired = false;
 
+  // Session-level native search counter (#1309).
+  // Tracks cumulative web_search_tool_result blocks across all turns in a session.
+  // Reset on session_start. Used to compute remaining budget for max_uses.
+  let sessionSearchCount = 0;
+
   // Track provider changes via model selection — also handles diagnostics
   // since model_select fires AFTER session_start and knows the provider.
   pi.on("model_select", async (event: any, ctx: any) => {
@@ -161,13 +176,41 @@ export function registerNativeSearchHooks(pi: NativeSearchPI): { getIsAnthropic:
     );
     payload.tools = tools;
 
+    // ── Session-level search budget (#1309) ──────────────────────────────
+    // Count web_search_tool_result blocks in the conversation history to
+    // determine how many native searches have already been used this session.
+    // The Anthropic API's max_uses resets per request, so without this guard,
+    // pause_turn → resubmit cycles allow unlimited total searches.
+    if (Array.isArray(messages)) {
+      let historySearchCount = 0;
+      for (const msg of messages) {
+        const content = msg.content;
+        if (!Array.isArray(content)) continue;
+        for (const block of content) {
+          if ((block as any)?.type === "web_search_tool_result") {
+            historySearchCount++;
+          }
+        }
+      }
+      // Sync counter from history (handles session restore / context replay)
+      sessionSearchCount = historySearchCount;
+    }
+
+    const remaining = Math.max(0, MAX_NATIVE_SEARCHES_PER_SESSION - sessionSearchCount);
+
+    if (remaining <= 0) {
+      // Budget exhausted — don't inject the search tool at all.
+      // The model will proceed without web search capability.
+      return payload;
+    }
+
     tools.push({
       type: "web_search_20250305",
       name: "web_search",
-      // Cap server-side searches per response to prevent the model from
-      // looping on web_search without synthesizing results (#817).
-      // 5 searches is generous — most queries need 1-2.
-      max_uses: 5,
+      // Cap per-request searches to the lesser of 5 (per-turn cap) or the
+      // remaining session budget (#1309). This prevents the model from
+      // consuming unlimited searches via pause_turn → resubmit cycles.
+      max_uses: Math.min(5, remaining),
     });
 
     return payload;
@@ -175,6 +218,9 @@ export function registerNativeSearchHooks(pi: NativeSearchPI): { getIsAnthropic:
 
   // Basic startup diagnostics — provider-specific info comes from model_select
   pi.on("session_start", async (_event: any, ctx: any) => {
+    // Reset session-level search budget (#1309)
+    sessionSearchCount = 0;
+
     const hasBrave = !!process.env.BRAVE_API_KEY;
     const hasJina = !!process.env.JINA_API_KEY;
     const hasAnswers = !!process.env.BRAVE_ANSWERS_KEY;

package/src/resources/extensions/shared/format-utils.ts

@@ -1,12 +1,12 @@
 /**
- * Shared formatting and layout utilities for TUI dashboard components.
+ * Shared pure formatting utilities — no @gsd/pi-tui dependency.
  *
- * Consolidates helpers that were previously duplicated across
- * auto-dashboard.ts, dashboard-overlay.ts, and visualizer-views.ts.
+ * ANSI-aware layout helpers (padRight, joinColumns, centerLine, fitColumns)
+ * live in layout-utils.ts to avoid pulling @gsd/pi-tui into modules that
+ * run outside jiti's alias resolution (e.g. HTML report generation via
+ * dynamic import in auto-loop).
  */
 
-import { truncateToWidth, visibleWidth } from "@gsd/pi-tui";
-
 // ─── Duration Formatting ──────────────────────────────────────────────────────
 
 /** Format a millisecond duration as a compact human-readable string. */
@@ -31,45 +31,6 @@ export function formatTokenCount(count: number): string {
   return `${(count / 1_000_000).toFixed(2)}M`;
 }
 
-// ─── Layout Helpers ───────────────────────────────────────────────────────────
-
-/** Pad a string with trailing spaces to fill `width` (ANSI-aware). */
-export function padRight(content: string, width: number): string {
-  const vis = visibleWidth(content);
-  return content + " ".repeat(Math.max(0, width - vis));
-}
-
-/** Build a line with left-aligned and right-aligned content. */
-export function joinColumns(left: string, right: string, width: number): string {
-  const leftW = visibleWidth(left);
-  const rightW = visibleWidth(right);
-  if (leftW + rightW + 2 > width) {
-    return truncateToWidth(`${left}  ${right}`, width);
-  }
-  return left + " ".repeat(width - leftW - rightW) + right;
-}
-
-/** Center content within `width` (ANSI-aware). */
-export function centerLine(content: string, width: number): string {
-  const vis = visibleWidth(content);
-  if (vis >= width) return truncateToWidth(content, width);
-  const leftPad = Math.floor((width - vis) / 2);
-  return " ".repeat(leftPad) + content;
-}
-
-/** Join as many parts as fit within `width`, separated by `separator`. */
-export function fitColumns(parts: string[], width: number, separator = "  "): string {
-  const filtered = parts.filter(Boolean);
-  if (filtered.length === 0) return "";
-  let result = filtered[0];
-  for (let i = 1; i < filtered.length; i++) {
-    const candidate = `${result}${separator}${filtered[i]}`;
-    if (visibleWidth(candidate) > width) break;
-    result = candidate;
-  }
-  return truncateToWidth(result, width);
-}
-
 // ─── Text Truncation ─────────────────────────────────────────────────────────
 
 /** Truncate a string to `maxLength` characters, replacing the last character with an ellipsis if needed. */

package/src/resources/extensions/shared/layout-utils.ts

@@ -0,0 +1,49 @@
+/**
+ * ANSI-aware TUI layout utilities that depend on @gsd/pi-tui.
+ *
+ * Separated from format-utils.ts so that modules needing only pure
+ * formatting (e.g. HTML report generation) can import format-utils
+ * without pulling in the @gsd/pi-tui dependency — which fails when
+ * loaded outside jiti's alias resolution context.
+ */
+
+import { truncateToWidth, visibleWidth } from "@gsd/pi-tui";
+
+// ─── Layout Helpers ───────────────────────────────────────────────────────────
+
+/** Pad a string with trailing spaces to fill `width` (ANSI-aware). */
+export function padRight(content: string, width: number): string {
+  const vis = visibleWidth(content);
+  return content + " ".repeat(Math.max(0, width - vis));
+}
+
+/** Build a line with left-aligned and right-aligned content. */
+export function joinColumns(left: string, right: string, width: number): string {
+  const leftW = visibleWidth(left);
+  const rightW = visibleWidth(right);
+  if (leftW + rightW + 2 > width) {
+    return truncateToWidth(`${left}  ${right}`, width);
+  }
+  return left + " ".repeat(width - leftW - rightW) + right;
+}
+
+/** Center content within `width` (ANSI-aware). */
+export function centerLine(content: string, width: number): string {
+  const vis = visibleWidth(content);
+  if (vis >= width) return truncateToWidth(content, width);
+  const leftPad = Math.floor((width - vis) / 2);
+  return " ".repeat(leftPad) + content;
+}
+
+/** Join as many parts as fit within `width`, separated by `separator`. */
+export function fitColumns(parts: string[], width: number, separator = "  "): string {
+  const filtered = parts.filter(Boolean);
+  if (filtered.length === 0) return "";
+  let result = filtered[0];
+  for (let i = 1; i < filtered.length; i++) {
+    const candidate = `${result}${separator}${filtered[i]}`;
+    if (visibleWidth(candidate) > width) break;
+    result = candidate;
+  }
+  return truncateToWidth(result, width);
+}

package/src/resources/extensions/shared/mod.ts

@@ -13,15 +13,18 @@ export {
 	stripAnsi,
 	formatTokenCount,
 	formatDuration,
-	padRight,
-	joinColumns,
-	centerLine,
-	fitColumns,
 	sparkline,
 	normalizeStringArray,
 	fileLink,
 } from "./format-utils.js";
 
+export {
+	padRight,
+	joinColumns,
+	centerLine,
+	fitColumns,
+} from "./layout-utils.js";
+
 export { shortcutDesc } from "./terminal.js";
 export { toPosixPath } from "./path-display.js";
 export { showInterviewRound } from "./interview-ui.js";

package/src/resources/extensions/shared/terminal.ts

@@ -7,9 +7,14 @@
 
 const UNSUPPORTED_TERMS = ["apple_terminal", "warpterm"];
 
+export function isCmuxTerminal(env: NodeJS.ProcessEnv = process.env): boolean {
+  return Boolean(env.CMUX_WORKSPACE_ID && env.CMUX_SURFACE_ID);
+}
+
 export function supportsCtrlAltShortcuts(): boolean {
   const term = (process.env.TERM_PROGRAM || "").toLowerCase();
   const jetbrains = (process.env.TERMINAL_EMULATOR || "").toLowerCase().includes("jetbrains");
+  if (isCmuxTerminal()) return true;
   return !UNSUPPORTED_TERMS.some((t) => term.includes(t)) && !jetbrains;
 }
 

package/src/resources/extensions/shared/tests/format-utils.test.ts

@@ -2,13 +2,15 @@ import { describe, it } from "node:test";
 import assert from "node:assert/strict";
 import {
   formatDuration,
+  sparkline,
+  stripAnsi,
+} from "../format-utils.js";
+import {
   padRight,
   joinColumns,
   centerLine,
   fitColumns,
-  sparkline,
-  stripAnsi,
-} from "../format-utils.js";
+} from "../layout-utils.js";
 
 describe("formatDuration", () => {
   it("formats seconds", () => {