@opengsd/gsd-pi 1.5.0 → 1.6.0

package/src/resources/extensions/gsd/auto.ts

@@ -116,7 +116,8 @@ import {
 } from "./auto-tool-tracking.js";
 import { closeoutUnit } from "./auto-unit-closeout.js";
 import { recoverTimedOutUnit } from "./auto-timeout-recovery.js";
-import { selectAndApplyModel, resolveModelId, clearToolBaseline } from "./auto-model-selection.js";
+import { selectAndApplyModel, resolveModelId, clearToolBaseline, isModelUnavailable, applyThinkingLevelForModel } from "./auto-model-selection.js";
+import { resolveModelWithFallbacksForUnit } from "./preferences-models.js";
 import { resetRoutingHistory, recordOutcome } from "./routing-history.js";
 import {
   checkPostUnitHooks,
@@ -127,6 +128,7 @@ import {
   runPreDispatchHooks,
   persistHookState,
   restoreHookState,
+  reconcileRestoredHookDispatch,
   clearPersistedHookState,
 } from "./post-unit-hooks.js";
 import { runGSDDoctor, rebuildState } from "./doctor.js";
@@ -177,24 +179,18 @@ import {
 import { GitServiceImpl } from "./git-service.js";
 import { nativeCheckoutBranch } from "./native-git-bridge.js";
 import { getPriorSliceCompletionBlocker } from "./dispatch-guard.js";
-import {
-  createAutoWorktree,
-  enterAutoWorktree,
-  enterBranchModeForMilestone,
-  teardownAutoWorktree,
-  isInAutoWorktree,
-  getAutoWorktreePath,
-  getAutoWorktreeOriginalBase,
-  mergeMilestoneToMain,
-  autoWorktreeBranch,
-  syncWorktreeStateBack,
-  readResourceVersion,
-  checkResourcesStale,
-  escapeStaleWorktree,
-} from "./auto-worktree.js";
+import { autoWorktreeBranch, enterBranchModeForMilestone } from "./auto-worktree-branch-lifecycle.js";
+import { createAutoWorktree } from "./auto-worktree-creation.js";
+import { enterAutoWorktree, isInAutoWorktree } from "./auto-worktree-entry.js";
+import { getAutoWorktreePath } from "./auto-worktree-path-resolution.js";
+import { checkResourcesStale, readResourceVersion } from "./auto-worktree-resource-version.js";
+import { escapeStaleWorktree } from "./auto-worktree-runtime-cleanup.js";
+import { getAutoWorktreeOriginalBase } from "./auto-worktree-session-registry.js";
+import { syncWorktreeStateBack } from "./auto-worktree-sync.js";
+import { teardownAutoWorktree } from "./auto-worktree-teardown.js";
 import { pruneQueueOrder } from "./queue-order.js";
 import { startCommandPolling as _startCommandPolling, isRemoteConfigured } from "../remote-questions/manager.js";
-import { createMilestoneMergeTransaction } from "./milestone-merge-transaction.js";
+import { createDefaultMilestoneMergeTransaction } from "./milestone-merge-transaction.js";
 
 import { debugLog, isDebugEnabled, writeDebugSummary } from "./debug-logger.js";
 import {
@@ -621,11 +617,27 @@ export function startAutoDetached(
     milestoneLock?: string | null;
   },
 ): void {
-  void withDetachedAutoKeepalive(startAuto(ctx, pi, base, verboseMode, options)).catch((err) => {
+  void withDetachedAutoKeepalive(startAuto(ctx, pi, base, verboseMode, options)).catch(async (err) => {
     const message = getErrorMessage(err);
     ctx.ui.notify(`Auto-start failed: ${message}`, "error");
     logWarning("engine", `auto start error: ${message}`, { file: "auto.ts" });
     debugLog("auto-start-failed", { error: message });
+    // Backstop cleanup (#1235): if startAuto threw after auto-mode was activated
+    // (e.g. an exception during bootstrap, before the loop's try/finally could
+    // run cleanupAfterLoopExit), s.active and the on-disk auto lock leak, so the
+    // re-entry guard silently no-ops every later /gsd auto until restart. Clear
+    // the leaked activation here so a failed start is always recoverable.
+    if (s.active) {
+      try {
+        await cleanupAfterLoopExit(ctx);
+      } catch (cleanupErr) {
+        logWarning(
+          "engine",
+          `auto start cleanup failed: ${cleanupErr instanceof Error ? cleanupErr.message : String(cleanupErr)}`,
+          { file: "auto.ts" },
+        );
+      }
+    }
   });
 }
 
@@ -2326,7 +2338,7 @@ export function buildWorktreeLifecycleDeps(): WorktreeLifecycleDeps {
       return new GitServiceImpl(basePath, gitConfig);
     },
     worktreeProjection: new WorktreeStateProjection(),
-    mergeMilestone: createMilestoneMergeTransaction(mergeMilestoneToMain),
+    mergeMilestone: createDefaultMilestoneMergeTransaction(),
   };
 }
 
@@ -2833,6 +2845,10 @@ export async function startAuto(
       "info",
     );
     restoreHookState(s.basePath);
+    // A restored activeHook has no live dispatch (the sidecar queue is not
+    // persisted); re-enqueue it so the hook runs instead of blocking the next
+    // unrelated unit's close-out (#1246).
+    reconcileRestoredHookDispatch(s.basePath, s.sidecarQueue);
     // Re-sync managed resources on resume so long-lived auto sessions pick up
     // bundled extension updates before resume-time verification/state logic runs.
     // GSD_PKG_ROOT is set by loader.ts and points to the gsd-pi package root.
@@ -2901,15 +2917,18 @@ export async function startAuto(
       debugLog("resume-orchestration-resume", { error: err instanceof Error ? err.message : String(err) });
     }
     startAutoCommandPolling(s.basePath);
-    await runAutoLoopWithUok({
-      ctx,
-      pi,
-      s,
-      deps: loopDeps,
-      runKernelLoop: runUokKernelLoop,
-      runLegacyLoop: runLegacyAutoLoop,
-    });
-    await cleanupAfterLoopExit(ctx);
+    try {
+      await runAutoLoopWithUok({
+        ctx,
+        pi,
+        s,
+        deps: loopDeps,
+        runKernelLoop: runUokKernelLoop,
+        runLegacyLoop: runLegacyAutoLoop,
+      });
+    } finally {
+      await cleanupAfterLoopExit(ctx);
+    }
     return;
   }
 
@@ -2937,7 +2956,15 @@ export async function startAuto(
     bootstrapDeps,
     freshStartAssessment,
   );
-  if (!ready) return;
+  if (!ready) {
+    // bootstrapAutoSession sets s.active = true (auto-start.ts) before several
+    // post-activation bail-outs return false without a loop ever running (e.g.
+    // the SQLite-unavailable gate). Without this, s.active and the auto lock
+    // leak and every later /gsd auto silently no-ops via the re-entry guard
+    // (#1235). Run the canonical cleanup so a failed bootstrap stays recoverable.
+    if (s.active) await cleanupAfterLoopExit(ctx);
+    return;
+  }
 
   // Build scope after bootstrap has populated s.basePath / s.originalBasePath /
   // s.currentMilestoneId (including worktree setup inside bootstrapAutoSession).
@@ -2963,15 +2990,18 @@ export async function startAuto(
   startAutoCommandPolling(s.basePath);
 
   // Dispatch the first unit
-  await runAutoLoopWithUok({
-    ctx,
-    pi,
-    s,
-    deps: loopDeps,
-    runKernelLoop: runUokKernelLoop,
-    runLegacyLoop: runLegacyAutoLoop,
-  });
-  await cleanupAfterLoopExit(ctx);
+  try {
+    await runAutoLoopWithUok({
+      ctx,
+      pi,
+      s,
+      deps: loopDeps,
+      runKernelLoop: runUokKernelLoop,
+      runLegacyLoop: runLegacyAutoLoop,
+    });
+  } finally {
+    await cleanupAfterLoopExit(ctx);
+  }
 }
 
 // describeNextUnit is imported from auto-dashboard.ts and re-exported
@@ -3129,19 +3159,56 @@ export async function dispatchHookUnit(
     workspaceRoot: s.basePath,
   });
 
-  if (hookModel) {
-    const availableModels = ctx.modelRegistry.getAvailable();
-    const match = resolveModelId(hookModel, availableModels, ctx.model?.provider);
-    if (match) {
+  // Resolve the hook's model honoring the configured `fallbacks[]` chain
+  // (#1229). The manual trigger path bypasses selectAndApplyModel, so resolve
+  // the primary→fallbacks order here and apply the first candidate available in
+  // the registry — a blocked or unconfigured primary then transparently drops
+  // to a fallback instead of silently reverting to the session model.
+  const availableModels = ctx.modelRegistry.getAvailable();
+  const availableModelIds = availableModels.map(
+    (m: { provider: string; id: string }) => `${m.provider}/${m.id}`,
+  );
+  const hookModelConfig = resolveModelWithFallbacksForUnit(
+    hookUnitType,
+    targetBasePath,
+    availableModelIds,
+  );
+  const modelCandidates = hookModelConfig
+    ? [hookModelConfig.primary, ...hookModelConfig.fallbacks]
+    : hookModel
+      ? [hookModel]
+      : [];
+  if (modelCandidates.length > 0) {
+    let applied = false;
+    for (const candidate of modelCandidates) {
+      const match = resolveModelId(candidate, availableModels, ctx.model?.provider);
+      if (!match) continue;
+      // Skip models the runtime has marked blocked or temporarily unavailable
+      // (e.g. a primary that just tripped a provider limit) so the configured
+      // fallbacks[] chain actually engages — parity with auto-mode's
+      // selectAndApplyModel, which consults the same blocked-models store (#1229).
+      if (isModelUnavailable(targetBasePath, match.provider, match.id)) continue;
       try {
-        await pi.setModel(match);
+        if (await pi.setModel(match)) {
+          // The manual trigger path bypasses selectAndApplyModel, so apply the
+          // hook's per-field `thinking` (from `post_unit_hooks[].model`'s object
+          // form) here against the just-set model rather than leaving the hook at
+          // the session level (#1269). Absent → session level, unchanged.
+          if (hookModelConfig?.thinking) {
+            applyThinkingLevelForModel(pi, hookModelConfig.thinking, match, ctx);
+          }
+          applied = true;
+          break;
+        }
       } catch (err) {
-        /* non-fatal */
-        logWarning("dispatch", `hook model set failed: ${err instanceof Error ? err.message : String(err)}`, { file: "auto.ts" });
+        /* non-fatal — try the next fallback */
+        logWarning("dispatch", `hook model set failed for ${candidate}: ${err instanceof Error ? err.message : String(err)}`, { file: "auto.ts" });
       }
-    } else {
+    }
+    if (!applied) {
       ctx.ui.notify(
-        `Hook model "${hookModel}" not found in available models. Falling back to current session model. ` +
+        `Hook model${modelCandidates.length > 1 ? "s" : ""} "${modelCandidates.join(", ")}" not available. ` +
+        `Falling back to current session model. ` +
         `Ensure the model is defined in models.json and has auth configured.`,
         "warning",
       );

package/src/resources/extensions/gsd/bootstrap/dynamic-tools.ts

@@ -14,7 +14,7 @@ import {
   type WorkflowDatabaseOpenResult,
   type WorkflowDatabaseStatus,
 } from "../db-workspace.js";
-import { getAutoWorktreePath } from "../auto-worktree.js";
+import { getAutoWorktreePath } from "../auto-worktree-path-resolution.js";
 import { resolveWorktreeProjectRoot } from "../worktree-root.js";
 import { worktreesDirs } from "../worktree-placement.js";
 

package/src/resources/extensions/gsd/bootstrap/register-hooks.ts

@@ -40,7 +40,7 @@ import {
 } from "../auto-tool-tracking.js";
 import { applyProviderPayloadPolicy } from "../provider-payload-policy.js";
 
-import { checkToolCallLoop, recordToolCallLoopMutation, resetToolCallLoopGuard } from "./tool-call-loop-guard.js";
+import { checkToolCallLoop, configureToolCallLoopGuard, recordToolCallLoopMutation, resetToolCallLoopGuard } from "./tool-call-loop-guard.js";
 import { MINIMAL_AUTO_BASE_TOOL_NAMES } from "./core-session-tools.js";
 import { maybePauseAutoForApprovalGate, resetPendingGatePauseGuard } from "./pending-gate-pause.js";
 import { saveActivityLog } from "../activity-log.js";
@@ -594,6 +594,22 @@ async function applyCompactionThresholdOverride(ctx: ExtensionContext): Promise<
   }
 }
 
+/**
+ * Apply user-tunable tool-call loop guard thresholds (#1198) from GSD
+ * preferences plus GSD_TOOL_LOOP_* env overrides. Runs at session boundaries
+ * so both interactive sessions and `/gsd auto` pick up the configuration.
+ * Non-fatal: falls back to built-in defaults when preferences cannot be loaded.
+ */
+async function applyToolCallLoopGuardConfig(basePath: string): Promise<void> {
+  try {
+    const { loadEffectiveGSDPreferences } = await import("../preferences.js");
+    const prefs = loadEffectiveGSDPreferences(basePath);
+    configureToolCallLoopGuard(prefs?.preferences.tool_call_loop_guard);
+  } catch {
+    configureToolCallLoopGuard(null);
+  }
+}
+
 function clearDeferredApprovalGate(basePath?: string): void {
   if (!basePath) {
     deferredApprovalGates.clear();
@@ -996,6 +1012,7 @@ export function registerHooks(
     }
     resetWriteGateState(basePath);
     resetToolCallLoopGuard();
+    await applyToolCallLoopGuardConfig(basePath);
     approvalQuestionAbortInFlight = false;
     clearDeferredApprovalGate();
     await resetAskUserQuestionsTurnCache();
@@ -1087,6 +1104,7 @@ export function registerHooks(
     initSessionNotifications(ctx);
     resetWriteGateState(basePath);
     resetToolCallLoopGuard();
+    await applyToolCallLoopGuardConfig(basePath);
     clearDeferredApprovalGate();
     await resetAskUserQuestionsTurnCache();
     clearDiscussionFlowState(basePath);
@@ -1707,7 +1725,7 @@ export function registerHooks(
       }
     }
     if (!event.isError) {
-      recordToolCallLoopMutation(toolName);
+      recordToolCallLoopMutation(toolName, event.details);
       refreshSourceObservationAfterMutation(toolName, event.input, ctx);
       clearSourceObservationsAfterShell(toolName);
     }

package/src/resources/extensions/gsd/bootstrap/system-context.ts

@@ -14,7 +14,7 @@ import { resolveModelWithFallbacksForUnit } from "../preferences-models.js";
 import { resolveGsdRootFile, resolveSliceFile, resolveSlicePath, resolveTaskFile, resolveTaskFiles, resolveTasksDir, relSliceFile, relSlicePath, relTaskFile } from "../paths.js";
 import { extractIntroAndRules } from "../knowledge-parser.js";
 import { ensureCodebaseMapFresh, readCodebaseMap } from "../codebase-generator.js";
-import { getActiveAutoWorktreeContext } from "../auto-worktree.js";
+import { getActiveAutoWorktreeContext } from "../auto-worktree-session-registry.js";
 import { getActiveWorktreeName, getWorktreeOriginalCwd } from "../worktree-session-state.js";
 import { deriveState } from "../state.js";
 import { formatOverridesSection, formatShortcut, loadActiveOverrides, loadFile, parseContinue, parseSummary } from "../files.js";

package/src/resources/extensions/gsd/bootstrap/tool-call-loop-guard.ts

@@ -14,13 +14,21 @@
  * to stop tooling for the rest of that turn and answer in text.
  *
  * The per-tool-name check (#783 Brief C) tracks call counts within a
- * turn regardless of args, decaying after successful state-mutating tools.
+ * turn regardless of args, decaying after successful state-progressing tools.
  * This catches improvisation
  * loops where the model attempts the same missing workflow tool through
  * varied surfaces (bash → `node -e` → CLI), each with a different
  * signature, so the identical-args streak never trips, while allowing
- * tool-heavy turns that are making file-mutation progress. Whichever guard
- * trips first blocks.
+ * tool-heavy turns that are making progress. Progress means a successful file
+ * mutation (edit/write) or a successful shell/exec call (bg_shell,
+ * gsd_uat_exec, …) — so a local model running a legitimate iterative
+ * debugging loop (restart server, npm install, curl an endpoint) is not
+ * blocked as a false loop (#1206). Whichever guard trips first blocks.
+ *
+ * Thresholds, exempt tools, and enable flags are user-tunable (#1198) via the
+ * `tool_call_loop_guard` key in `.gsd/PREFERENCES.md` and `GSD_TOOL_LOOP_*`
+ * environment variables, applied through {@link configureToolCallLoopGuard}.
+ * Defaults preserve the original hardcoded behavior.
  */
 
 import { createHash } from "node:crypto";
@@ -28,26 +36,189 @@ import { INHERENTLY_REPEATABLE_TOOL_SET } from "./core-session-tools.js";
 import { hasBrowserContractPrefix } from "../../shared/browser-contract.js";
 import { canonicalToolName } from "../engine-hook-contract.js";
 
-const MAX_CONSECUTIVE_IDENTICAL_CALLS = 4;
+/** Built-in defaults. Preserved when preferences/env do not override them. */
+const DEFAULT_MAX_CONSECUTIVE_IDENTICAL_CALLS = 4;
+const DEFAULT_PER_TOOL_DEFAULT_CAP = 6;
+const DEFAULT_PER_TOOL_REPEATABLE_CAP = 15;
+const DEFAULT_PER_TOOL_CAP_EXEMPT_TOOLS = ["find", "glob", "grep", "ls", "read", "search_and_read"] as const;
 
 /** Interactive/user-facing tools where even 1 duplicate is confusing. */
 const STRICT_LOOP_TOOLS = new Set(["ask_user_questions"]);
 const MAX_CONSECUTIVE_STRICT = 1;
 
+const STATE_MUTATING_TOOL_SET = new Set(["edit", "write", "multi_edit", "notebook_edit"]);
+
 /**
- * Per-turn cap on calls to the SAME tool name, regardless of args (#783).
+ * Successful shell/exec calls are state progression too (#1206), not just file
+ * mutations. A local model debugging a UAT scenario legitimately restarts
+ * servers, runs `npm install`, and curls endpoints via bg_shell / gsd_uat_exec
+ * with varied args — productive work that never touches a file, so Guard 2's
+ * per-tool cap would otherwise misfire as a false loop. Decaying on these
+ * successful calls keeps the guard from aborting a progressing turn.
  *
- * General-purpose execution tools are routinely called many times per turn
- * (touching multiple files, running several commands), so they get a higher
- * ceiling. Everything else — workflow one-shot tools (e.g. gsd_complete_milestone)
- * and any non-allowlisted tool — gets the default cap. The default is generous
- * enough to absorb legitimate retries but catches the reported improvisation
- * loop (~51 calls) well before a cost spike.
+ * Only successful calls reach {@link recordToolCallLoopMutation}: the
+ * tool_result hook skips `isError` results, and the bash tool throws on
+ * non-zero exit, so a genuinely stuck improvisation loop of failing commands
+ * (#783) never decays and still trips the cap. `async_bash` is excluded
+ * because it only registers a background job; command success is reported
+ * later via `await_job`. `bg_shell` may return `isError: false` for failed
+ * `run`/`start` outcomes, so the hook passes `details` for validation.
  */
-const PER_TOOL_DEFAULT_CAP = 6;
-const PER_TOOL_REPEATABLE_CAP = 15;
-const PER_TOOL_CAP_EXEMPT_TOOLS = new Set(["find", "glob", "grep", "ls", "read", "search_and_read"]);
-const STATE_MUTATING_TOOL_SET = new Set(["edit", "write", "multi_edit", "notebook_edit"]);
+const STATE_PROGRESSING_EXEC_TOOL_SET = new Set([
+  "bash",
+  "bg_shell",
+  "shell",
+  "powershell",
+  "gsd_exec",
+  "gsd_uat_exec",
+]);
+
+function isSuccessfulBgShellLoopProgress(details: unknown): boolean {
+  if (!details || typeof details !== "object") return false;
+  const record = details as Record<string, unknown>;
+  switch (record.action) {
+    case "run":
+      return record.exitCode === 0 && record.timedOut !== true;
+    case "start":
+    case "restart": {
+      const process = record.process;
+      if (!process || typeof process !== "object") return false;
+      return (process as Record<string, unknown>).alive === true;
+    }
+    case "wait_for_ready":
+      return record.ready === true;
+    case "send_and_wait":
+      return record.matched === true;
+    default:
+      return true;
+  }
+}
+
+/**
+ * User-tunable configuration shape for the loop guard (#1198).
+ *
+ * Mirrors the `tool_call_loop_guard` key in `.gsd/PREFERENCES.md`. All fields
+ * are optional; anything omitted falls back to the built-in defaults so that
+ * existing installs keep their current behavior.
+ */
+export interface ToolCallLoopGuardConfig {
+  enabled?: boolean;
+  identical_args?: {
+    enabled?: boolean;
+    max_consecutive_calls?: number;
+  };
+  repeated_tool?: {
+    enabled?: boolean;
+    default_cap?: number;
+    repeatable_cap?: number;
+    exempt_tools?: string[];
+  };
+}
+
+/**
+ * Active, resolved configuration. Set by {@link configureToolCallLoopGuard}
+ * from preferences + environment overrides, and deliberately NOT touched by
+ * {@link resetToolCallLoopGuard} so a session's tuning survives turn
+ * boundaries.
+ */
+interface ResolvedGuardConfig {
+  /** Master switch for both guards. */
+  guardEnabled: boolean;
+  identicalEnabled: boolean;
+  maxConsecutiveIdentical: number;
+  repeatedEnabled: boolean;
+  perToolDefaultCap: number;
+  perToolRepeatableCap: number;
+  /** Tool names exempt from Guard 2 (per-tool-name cap). */
+  perToolExempt: Set<string>;
+}
+
+function defaultGuardConfig(): ResolvedGuardConfig {
+  return {
+    guardEnabled: true,
+    identicalEnabled: true,
+    maxConsecutiveIdentical: DEFAULT_MAX_CONSECUTIVE_IDENTICAL_CALLS,
+    repeatedEnabled: true,
+    perToolDefaultCap: DEFAULT_PER_TOOL_DEFAULT_CAP,
+    perToolRepeatableCap: DEFAULT_PER_TOOL_REPEATABLE_CAP,
+    perToolExempt: new Set<string>(DEFAULT_PER_TOOL_CAP_EXEMPT_TOOLS),
+  };
+}
+
+let config: ResolvedGuardConfig = defaultGuardConfig();
+
+/** Parse a positive-integer env var, returning undefined when unset/invalid. */
+function envPositiveInt(name: string): number | undefined {
+  const raw = process.env[name];
+  if (raw === undefined || raw.trim() === "") return undefined;
+  const n = Number(raw);
+  return Number.isFinite(n) && n >= 1 ? Math.floor(n) : undefined;
+}
+
+/** Parse a boolean env var (`true`/`1` vs `false`/`0`), undefined when unset. */
+function envBool(name: string): boolean | undefined {
+  const raw = process.env[name];
+  if (raw === undefined || raw.trim() === "") return undefined;
+  const v = raw.trim().toLowerCase();
+  if (v === "true" || v === "1" || v === "yes" || v === "on") return true;
+  if (v === "false" || v === "0" || v === "no" || v === "off") return false;
+  return undefined;
+}
+
+/** Parse a comma-separated tool-name list env var. */
+function envToolList(name: string): string[] {
+  const raw = process.env[name];
+  if (!raw) return [];
+  return raw.split(",").map((s) => s.trim()).filter(Boolean);
+}
+
+/**
+ * Resolve preferences + environment overrides into the active configuration.
+ * Environment variables win over preferences; preferences win over built-in
+ * defaults. User-supplied exempt tools are additive to the built-in exempt
+ * set so defaults are always preserved.
+ *
+ * Applied to both interactive sessions and `/gsd auto` (called from the
+ * session_start / session_switch hooks).
+ */
+export function configureToolCallLoopGuard(prefs?: ToolCallLoopGuardConfig | null): void {
+  const next = defaultGuardConfig();
+
+  if (prefs) {
+    if (typeof prefs.enabled === "boolean") next.guardEnabled = prefs.enabled;
+    if (prefs.identical_args) {
+      if (typeof prefs.identical_args.enabled === "boolean") next.identicalEnabled = prefs.identical_args.enabled;
+      if (typeof prefs.identical_args.max_consecutive_calls === "number" && prefs.identical_args.max_consecutive_calls >= 1) {
+        next.maxConsecutiveIdentical = Math.floor(prefs.identical_args.max_consecutive_calls);
+      }
+    }
+    if (prefs.repeated_tool) {
+      if (typeof prefs.repeated_tool.enabled === "boolean") next.repeatedEnabled = prefs.repeated_tool.enabled;
+      if (typeof prefs.repeated_tool.default_cap === "number" && prefs.repeated_tool.default_cap >= 1) {
+        next.perToolDefaultCap = Math.floor(prefs.repeated_tool.default_cap);
+      }
+      if (typeof prefs.repeated_tool.repeatable_cap === "number" && prefs.repeated_tool.repeatable_cap >= 1) {
+        next.perToolRepeatableCap = Math.floor(prefs.repeated_tool.repeatable_cap);
+      }
+      for (const tool of prefs.repeated_tool.exempt_tools ?? []) {
+        if (typeof tool === "string" && tool.trim()) next.perToolExempt.add(tool.trim());
+      }
+    }
+  }
+
+  // Environment overrides (win over preferences).
+  const envEnabled = envBool("GSD_TOOL_LOOP_GUARD_ENABLED");
+  if (envEnabled !== undefined) next.guardEnabled = envEnabled;
+  const envIdenticalMax = envPositiveInt("GSD_TOOL_LOOP_IDENTICAL_MAX");
+  if (envIdenticalMax !== undefined) next.maxConsecutiveIdentical = envIdenticalMax;
+  const envDefaultCap = envPositiveInt("GSD_TOOL_LOOP_REPEATED_DEFAULT_CAP");
+  if (envDefaultCap !== undefined) next.perToolDefaultCap = envDefaultCap;
+  const envRepeatableCap = envPositiveInt("GSD_TOOL_LOOP_REPEATED_REPEATABLE_CAP");
+  if (envRepeatableCap !== undefined) next.perToolRepeatableCap = envRepeatableCap;
+  for (const tool of envToolList("GSD_TOOL_LOOP_EXEMPT_TOOLS")) next.perToolExempt.add(tool);
+
+  config = next;
+}
 
 let consecutiveCount = 0;
 let lastSignature = "";
@@ -82,17 +253,20 @@ function hashToolCall(toolName: string, args: Record<string, unknown>): string {
  * Returns `{ block: true, reason }` when the loop threshold is exceeded.
  *
  * Two independent guards run; whichever trips first blocks:
- *  1. Identical-signature streak (MAX_CONSECUTIVE_IDENTICAL_CALLS, strict for
+ *  1. Identical-signature streak (config.maxConsecutiveIdentical, strict for
  *     ask_user_questions).
- *  2. Per-tool-name cap (PER_TOOL_DEFAULT_CAP / PER_TOOL_REPEATABLE_CAP),
+ *  2. Per-tool-name cap (config.perToolDefaultCap / config.perToolRepeatableCap),
  *     independent of args, reset after file-mutation progress — catches
  *     improvisation loops (#783).
+ *
+ * Both guards, their thresholds, and the per-tool exempt set are user-tunable
+ * via {@link configureToolCallLoopGuard} (#1198).
  */
 export function checkToolCallLoop(
   toolName: string,
   args: Record<string, unknown>,
 ): { block: boolean; reason?: string; count?: number } {
-  if (!enabled) return { block: false, count: 0 };
+  if (!enabled || !config.guardEnabled) return { block: false, count: 0 };
 
   const sig = hashToolCall(toolName, args);
 
@@ -107,14 +281,16 @@ export function checkToolCallLoop(
   // ── Guard 1: identical-signature streak ──
   const threshold = STRICT_LOOP_TOOLS.has(toolName)
     ? MAX_CONSECUTIVE_STRICT
-    : MAX_CONSECUTIVE_IDENTICAL_CALLS;
+    : config.maxConsecutiveIdentical;
 
-  if (consecutiveCount > threshold) {
+  if (config.identicalEnabled && consecutiveCount > threshold) {
     return {
       block: true,
       reason:
         `Tool loop detected (identical args): ${toolName} called ${consecutiveCount} times ` +
-        `with identical arguments. Blocking to prevent infinite loop. ` +
+        `with identical arguments (max ${threshold}). Blocking to prevent infinite loop. ` +
+        `Raise tool_call_loop_guard.identical_args.max_consecutive_calls in .gsd/PREFERENCES.md ` +
+        `(or GSD_TOOL_LOOP_IDENTICAL_MAX) if this is expected. ` +
         `Do not retry this tool or call other tools this turn — stop and respond to the user in text.`,
       count: consecutiveCount,
     };
@@ -139,15 +315,16 @@ export function checkToolCallLoop(
   // Guard 1's identical-signature streak still catches a genuinely stuck
   // browser loop.
   if (
-    PER_TOOL_CAP_EXEMPT_TOOLS.has(toolName) ||
+    !config.repeatedEnabled ||
+    config.perToolExempt.has(toolName) ||
     hasBrowserContractPrefix(canonicalToolName(toolName))
   ) {
     return { block: false, count: consecutiveCount };
   }
 
   const perToolCap = INHERENTLY_REPEATABLE_TOOL_SET.has(toolName)
-    ? PER_TOOL_REPEATABLE_CAP
-    : PER_TOOL_DEFAULT_CAP;
+    ? config.perToolRepeatableCap
+    : config.perToolDefaultCap;
 
   if (perToolCount > perToolCap) {
     return {
@@ -156,6 +333,9 @@ export function checkToolCallLoop(
         `Tool loop detected (repeated tool): ${toolName} called ${perToolCount} times ` +
         `this turn (cap ${perToolCap}). Blocking to prevent infinite loop. ` +
         `The tool may be unavailable or failing repeatedly. ` +
+        `Raise tool_call_loop_guard.repeated_tool.default_cap/repeatable_cap or add "${toolName}" ` +
+        `to tool_call_loop_guard.repeated_tool.exempt_tools in .gsd/PREFERENCES.md ` +
+        `(or GSD_TOOL_LOOP_REPEATED_* / GSD_TOOL_LOOP_EXEMPT_TOOLS) if this is expected. ` +
         `Do not retry this tool or pivot to other tools this turn — stop and respond to the user in text.`,
       count: perToolCount,
     };
@@ -164,10 +344,21 @@ export function checkToolCallLoop(
   return { block: false, count: consecutiveCount };
 }
 
-/** Record successful mutation progress so Guard 2 can decay on the next count. */
-export function recordToolCallLoopMutation(toolName: string): void {
+/**
+ * Record successful state progress so Guard 2 can decay on the next count.
+ * File mutations (edit/write/…) and successful shell/exec calls (bash,
+ * bg_shell, gsd_uat_exec, …) both count as progress (#1092, #1206). The caller
+ * only invokes this for non-error results, so failing improvisation loops
+ * (#783) never decay and still trip the per-tool cap.
+ */
+export function recordToolCallLoopMutation(toolName: string, details?: unknown): void {
   if (!enabled) return;
-  if (!STATE_MUTATING_TOOL_SET.has(toolName)) return;
+  if (STATE_MUTATING_TOOL_SET.has(toolName)) {
+    mutationEpoch++;
+    return;
+  }
+  if (!STATE_PROGRESSING_EXEC_TOOL_SET.has(toolName)) return;
+  if (toolName === "bg_shell" && details !== undefined && !isSuccessfulBgShellLoopProgress(details)) return;
   mutationEpoch++;
 }
 

package/src/resources/extensions/gsd/codebase-generator.ts

@@ -675,10 +675,16 @@ export function ensureCodebaseMapFresh(
 
   const metadata = parseCodebaseMapMetadata(existing);
   const existingDescriptions = parseCodebaseMap(existing);
+  // Compare against the truncated count: when the repository has more tracked
+  // files than maxFiles, the map only lists the first maxFiles entries, so a
+  // stored fileCount above the cap (e.g. from an older map that recorded the
+  // full count) still matches the current truncated listing. Comparing raw
+  // values would flag "file-count-changed" on every turn, rewriting the
+  // timestamp and invalidating the provider KV cache prefix.
   const staleReason =
     !metadata ? "missing-metadata"
     : metadata.fingerprint !== fingerprint ? "files-changed"
-    : metadata.fileCount !== listed.files.length ? "file-count-changed"
+    : Math.min(metadata.fileCount, resolved.maxFiles) !== listed.files.length ? "file-count-changed"
     : metadata.truncated !== listed.truncated ? "truncation-changed"
     : undefined;