gsd-pi 2.74.0 → 2.75.0-dev.fd2382b9f

package/dist/resources/extensions/gsd/auto.js

@@ -36,7 +36,7 @@ import { captureAvailableSkills, resetSkillTelemetry, } from "./skill-telemetry.
 import { getRtkSessionSavings } from "../shared/rtk-session-stats.js";
 import { deactivateGSD } from "../shared/gsd-phase-state.js";
 import { initMetrics, resetMetrics, getLedger, getProjectTotals, formatCost, formatTokenCount, } from "./metrics.js";
-import { logWarning } from "./workflow-logger.js";
+import { setLogBasePath, logWarning } from "./workflow-logger.js";
 import { homedir } from "node:os";
 import { join } from "node:path";
 import { pathToFileURL } from "node:url";
@@ -45,8 +45,9 @@ import { atomicWriteSync } from "./atomic-write.js";
 import { autoCommitCurrentBranch, captureIntegrationBranch, detectWorktreeName, getCurrentBranch, getMainBranch, setActiveMilestoneId, } from "./worktree.js";
 import { GitServiceImpl } from "./git-service.js";
 import { getPriorSliceCompletionBlocker } from "./dispatch-guard.js";
-import { createAutoWorktree, enterAutoWorktree, teardownAutoWorktree, isInAutoWorktree, getAutoWorktreePath, mergeMilestoneToMain, autoWorktreeBranch, syncWorktreeStateBack, syncProjectRootToWorktree, checkResourcesStale, escapeStaleWorktree, } from "./auto-worktree.js";
+import { createAutoWorktree, enterAutoWorktree, enterBranchModeForMilestone, teardownAutoWorktree, isInAutoWorktree, getAutoWorktreePath, mergeMilestoneToMain, autoWorktreeBranch, syncWorktreeStateBack, syncProjectRootToWorktree, checkResourcesStale, escapeStaleWorktree, } from "./auto-worktree.js";
 import { pruneQueueOrder } from "./queue-order.js";
+import { startCommandPolling as _startCommandPolling, isRemoteConfigured } from "../remote-questions/manager.js";
 import { debugLog, isDebugEnabled, writeDebugSummary } from "./debug-logger.js";
 import { reconcileMergeState, } from "./auto-recovery.js";
 import { resolveDispatch, DISPATCH_RULES } from "./auto-dispatch.js";
@@ -61,10 +62,12 @@ import { clearCmuxSidebar, logCmuxEvent, syncCmuxSidebar } from "../cmux/index.j
 // ── Extracted modules ──────────────────────────────────────────────────────
 import { startUnitSupervision } from "./auto-timers.js";
 import { runPostUnitVerification } from "./auto-verification.js";
-import { postUnitPreVerification, postUnitPostVerification, } from "./auto-post-unit.js";
+import { autoCommitUnit, postUnitPreVerification, postUnitPostVerification, } from "./auto-post-unit.js";
 import { bootstrapAutoSession, openProjectDbIfPresent } from "./auto-start.js";
 import { initHealthWidget } from "./health-widget.js";
-import { autoLoop, resolveAgentEnd, resolveAgentEndCancelled, _resetPendingResolve, isSessionSwitchInFlight } from "./auto-loop.js";
+import { runLegacyAutoLoop, runUokKernelLoop, resolveAgentEnd, resolveAgentEndCancelled, _resetPendingResolve, isSessionSwitchInFlight } from "./auto-loop.js";
+import { runAutoLoopWithUok } from "./uok/kernel.js";
+import { resolveUokFlags } from "./uok/flags.js";
 import { WorktreeResolver, } from "./worktree-resolver.js";
 import { reorderForCaching } from "./prompt-ordering.js";
 // ─── Session State ─────────────────────────────────────────────────────────
@@ -140,6 +143,7 @@ export function startAutoDetached(ctx, pi, base, verboseMode, options) {
         debugLog("auto-start-failed", { error: message });
     });
 }
+/** Returns true if the project is configured for `isolation:worktree` mode. */
 export function shouldUseWorktreeIsolation() {
     const prefs = loadEffectiveGSDPreferences()?.preferences?.git;
     if (prefs?.isolation === "worktree")
@@ -183,6 +187,26 @@ function deregisterSigtermHandler() {
     _deregisterSigtermHandler(s.sigtermHandler);
     s.sigtermHandler = null;
 }
+/**
+ * Wrapper: start background command polling for the configured remote channel
+ * (currently Telegram only). Stores the cleanup function on the session so
+ * every exit path can stop the interval via stopCommandPolling().
+ * No-op when no remote channel is configured.
+ */
+function startAutoCommandPolling(basePath) {
+    if (!isRemoteConfigured())
+        return;
+    // Clear any existing interval before starting a new one (e.g. resume path).
+    stopAutoCommandPolling();
+    s.commandPollingCleanup = _startCommandPolling(basePath);
+}
+/** Wrapper: stop background command polling and clear the stored cleanup. */
+function stopAutoCommandPolling() {
+    if (s.commandPollingCleanup) {
+        s.commandPollingCleanup();
+        s.commandPollingCleanup = null;
+    }
+}
 export function getAutoDashboardData() {
     const ledger = getLedger();
     const totals = ledger ? getProjectTotals(ledger.units) : null;
@@ -246,6 +270,14 @@ export function getActiveRunDir() {
 export function getAutoModeStartModel() {
     return s.autoModeStartModel;
 }
+/**
+ * Update the dashboard-facing dispatched model label.
+ * Used when runtime recovery switches models mid-unit (e.g. provider fallback)
+ * so the AUTO box reflects the active model immediately.
+ */
+export function setCurrentDispatchedModelId(model) {
+    s.currentDispatchedModelId = model ? `${model.provider}/${model.id}` : null;
+}
 // Tool tracking — delegates to auto-tool-tracking.ts
 export function markToolStart(toolCallId, toolName) {
     _markToolStart(toolCallId, s.active, toolName);
@@ -360,10 +392,22 @@ function clearUnitTimeout() {
 }
 /** Build snapshot metric opts. */
 function buildSnapshotOpts(_unitType, _unitId) {
+    const prefs = loadEffectiveGSDPreferences()?.preferences;
+    const uokFlags = resolveUokFlags(prefs);
     return {
         ...(s.autoStartTime > 0 ? { autoSessionKey: String(s.autoStartTime) } : {}),
         promptCharCount: s.lastPromptCharCount,
         baselineCharCount: s.lastBaselineCharCount,
+        traceId: s.currentTraceId ?? undefined,
+        turnId: s.currentTurnId ?? undefined,
+        ...(uokFlags.gitops
+            ? {
+                gitAction: uokFlags.gitopsTurnAction,
+                gitPush: uokFlags.gitopsTurnPush,
+                gitStatus: s.lastGitActionStatus ?? undefined,
+                gitError: s.lastGitActionFailure ?? undefined,
+            }
+            : {}),
         ...(s.currentUnitRouting ?? {}),
     };
 }
@@ -378,6 +422,7 @@ function handleLostSessionLock(ctx, lockStatus) {
     s.paused = false;
     deactivateGSD();
     clearUnitTimeout();
+    stopAutoCommandPolling();
     restoreProjectRootEnv();
     restoreMilestoneLockEnv();
     deregisterSigtermHandler();
@@ -413,6 +458,7 @@ function cleanupAfterLoopExit(ctx) {
     s.active = false;
     deactivateGSD();
     clearUnitTimeout();
+    stopAutoCommandPolling();
     restoreProjectRootEnv();
     restoreMilestoneLockEnv();
     // Clear crash lock and release session lock so the next `/gsd next` does
@@ -457,6 +503,7 @@ export async function stopAuto(ctx, pi, reason) {
         // ── Step 1: Timers and locks ──
         try {
             clearUnitTimeout();
+            stopAutoCommandPolling();
             if (lockBase())
                 clearLock(lockBase());
             if (lockBase())
@@ -710,6 +757,7 @@ export async function pauseAuto(ctx, _pi, _errorContext) {
     if (!s.active)
         return;
     clearUnitTimeout();
+    stopAutoCommandPolling();
     // Flush queued follow-up messages (#3512).
     // Late async notifications (async_job_result, gsd-auto-wrapup) can trigger
     // extra LLM turns after pause. Flush them the same way run-unit.ts does.
@@ -800,6 +848,7 @@ function buildResolverDeps() {
         teardownAutoWorktree,
         createAutoWorktree,
         enterAutoWorktree,
+        enterBranchModeForMilestone,
         getAutoWorktreePath,
         autoCommitCurrentBranch,
         getCurrentBranch,
@@ -884,6 +933,7 @@ function buildLoopDeps() {
         getMainBranch,
         // Unit closeout + runtime records
         closeoutUnit,
+        autoCommitUnit,
         recordOutcome,
         writeLock,
         captureAvailableSkills,
@@ -926,6 +976,11 @@ function buildLoopDeps() {
         emitJournalEvent: (entry) => _emitJournalEvent(s.basePath, entry),
     };
 }
+/**
+ * Start auto-mode. Handles both fresh-start and resume paths, sets up session
+ * state, enters the milestone worktree or branch, and dispatches the first unit.
+ * No-ops if auto-mode is already active.
+ */
 export async function startAuto(ctx, pi, base, verboseMode, options) {
     if (s.active) {
         debugLog("startAuto", { phase: "already-active", skipping: true });
@@ -1081,6 +1136,11 @@ export async function startAuto(ctx, pi, base, verboseMode, options) {
         s.stepMode = requestedStepMode;
         s.cmdCtx = ctx;
         s.basePath = base;
+        // Ensure the workflow-logger audit log is pinned to the project root
+        // even when auto-mode is entered via a path that bypasses the
+        // bootstrap/dynamic-tools ensureDbOpen() → setLogBasePath() chain
+        // (e.g. /clear resume, hot-reload).
+        setLogBasePath(base);
         s.unitDispatchCount.clear();
         s.unitLifetimeDispatches.clear();
         if (!getLedger())
@@ -1092,9 +1152,9 @@ export async function startAuto(ctx, pi, base, verboseMode, options) {
             const level = to === "red" ? "error" : to === "yellow" ? "warning" : "info";
             ctx.ui.notify(summary, level);
         });
-        // ── Auto-worktree: re-enter worktree on resume ──
+        // ── Auto-worktree / branch-mode: re-enter on resume ──
         if (s.currentMilestoneId &&
-            shouldUseWorktreeIsolation() &&
+            getIsolationMode() !== "none" &&
             s.originalBasePath &&
             !isInAutoWorktree(s.basePath) &&
             !detectWorktreeName(s.basePath) &&
@@ -1159,7 +1219,15 @@ export async function startAuto(ctx, pi, base, verboseMode, options) {
         writeLock(lockBase(), "resuming", s.currentMilestoneId ?? "unknown");
         logCmuxEvent(loadEffectiveGSDPreferences()?.preferences, s.stepMode ? "Step-mode resumed." : "Auto-mode resumed.", "progress");
         captureProjectRootEnv(s.originalBasePath || s.basePath);
-        await autoLoop(ctx, pi, s, buildLoopDeps());
+        startAutoCommandPolling(s.basePath);
+        await runAutoLoopWithUok({
+            ctx,
+            pi,
+            s,
+            deps: buildLoopDeps(),
+            runKernelLoop: runUokKernelLoop,
+            runLegacyLoop: runLegacyAutoLoop,
+        });
         cleanupAfterLoopExit(ctx);
         return;
     }
@@ -1182,29 +1250,18 @@ export async function startAuto(ctx, pi, base, verboseMode, options) {
         logWarning("engine", `cmux sync failed: ${err instanceof Error ? err.message : String(err)}`, { file: "auto.ts" });
     }
     logCmuxEvent(loadEffectiveGSDPreferences()?.preferences, requestedStepMode ? "Step-mode started." : "Auto-mode started.", "progress");
+    startAutoCommandPolling(s.basePath);
     // Dispatch the first unit
-    await autoLoop(ctx, pi, s, buildLoopDeps());
+    await runAutoLoopWithUok({
+        ctx,
+        pi,
+        s,
+        deps: buildLoopDeps(),
+        runKernelLoop: runUokKernelLoop,
+        runLegacyLoop: runLegacyAutoLoop,
+    });
     cleanupAfterLoopExit(ctx);
 }
-// ─── Agent End Handler ────────────────────────────────────────────────────────
-/**
- * Deprecated thin wrapper — kept as export for backward compatibility.
- * The actual agent_end processing now happens via resolveAgentEnd() in auto-loop.ts,
- * which is called directly from index.ts. The autoLoop() while loop handles all
- * post-unit processing (verification, hooks, dispatch) that this function used to do.
- *
- * If called by straggler code, it simply resolves the pending promise so the loop
- * can continue.
- */
-export async function handleAgentEnd(ctx, pi) {
-    if (!s.active || !s.cmdCtx) {
-        // Even when inactive, resolve any pending promise so the loop is unblocked.
-        resolveAgentEndCancelled();
-        return;
-    }
-    clearUnitTimeout();
-    resolveAgentEnd({ messages: [] });
-}
 // describeNextUnit is imported from auto-dashboard.ts and re-exported
 export { describeNextUnit } from "./auto-dashboard.js";
 /** Thin wrapper: delegates to auto-dashboard.ts, passing state accessors. */

package/dist/resources/extensions/gsd/bootstrap/agent-end-recovery.js

@@ -1,6 +1,6 @@
 import { logWarning } from "../workflow-logger.js";
 import { checkAutoStartAfterDiscuss } from "../guided-flow.js";
-import { getAutoDashboardData, getAutoModeStartModel, isAutoActive, pauseAuto } from "../auto.js";
+import { getAutoDashboardData, getAutoModeStartModel, isAutoActive, pauseAuto, setCurrentDispatchedModelId } from "../auto.js";
 import { getNextFallbackModel, resolveModelWithFallbacksForUnit } from "../preferences.js";
 import { pauseAutoForProviderError } from "../provider-error-pause.js";
 import { isSessionSwitchInFlight, resolveAgentEnd } from "../auto-loop.js";
@@ -106,26 +106,11 @@ export async function handleAgentEnd(pi, event, ctx) {
         const explicitRetryAfterMs = ("retryAfterMs" in lastMsg && typeof lastMsg.retryAfterMs === "number") ? lastMsg.retryAfterMs : undefined;
         // ── 1. Classify using rawErrorMsg to avoid prose false-positives ────
         const cls = classifyError(rawErrorMsg, explicitRetryAfterMs);
-        // ── 1b. Defer to Core RetryHandler for transient errors ─────────────
-        // The Core RetryHandler (agent-session.ts) processes retryable errors
-        // AFTER this extension handler, in the same _processAgentEvent() call.
-        // For transient errors (overloaded, rate limit, server), the Core will
-        // retry in-context — same session, same conversation — which is strictly
-        // better than our Layer 2 pause+resume (which creates a new session).
-        //
-        // If we react here AND the Core also retries, we race: pauseAuto tears
-        // down the session while agent.continue() starts a new turn.
-        //
-        // Solution: Do nothing for transient errors. The Core RetryHandler
-        // runs next in _processAgentEvent and will either:
-        //   a) Retry successfully → new agent_end (success) → we see it next time
-        //   b) Exhaust retries → the agent stays idle, autoLoop's unit timeout
-        //      or stuck detection handles it
-        //
-        // We do NOT call resolveAgentEnd here — that would unblock autoLoop
-        // prematurely while the Core is still retrying in the same session.
-        // We do NOT call pauseAuto — that would tear down the session.
-        if (isTransient(cls)) {
+        // ── 1b. Defer to Core RetryHandler for most transient errors ────────
+        // Core retries transient failures in-session after this handler.
+        // Keep that behavior for non-rate-limit classes to avoid pause/retry races,
+        // but let rate-limit continue into model fallback logic below (#4373).
+        if (isTransient(cls) && cls.kind !== "rate-limit") {
             return;
         }
         // Cap rate-limit backoff for CLI-style providers (openai-codex, google-gemini-cli)
@@ -177,6 +162,7 @@ export async function handleAgentEnd(pi, event, ctx) {
                         if (modelToSet) {
                             const ok = await pi.setModel(modelToSet, { persist: false });
                             if (ok) {
+                                setCurrentDispatchedModelId({ provider: modelToSet.provider, id: modelToSet.id });
                                 ctx.ui.notify(`Model error${errorDetail}. Switched to fallback: ${nextModelId} and resuming.`, "warning");
                                 pi.sendMessage({ customType: "gsd-auto-timeout-recovery", content: "Continue execution.", display: false }, { triggerTurn: true });
                                 return;
@@ -193,6 +179,7 @@ export async function handleAgentEnd(pi, event, ctx) {
                     if (startModel) {
                         const ok = await pi.setModel(startModel, { persist: false });
                         if (ok) {
+                            setCurrentDispatchedModelId({ provider: startModel.provider, id: startModel.id });
                             retryState.networkRetryCount = 0;
                             retryState.currentRetryModelId = undefined;
                             ctx.ui.notify(`Model error${errorDetail}. Restored session model: ${sessionModel.provider}/${sessionModel.id} and resuming.`, "warning");

package/dist/resources/extensions/gsd/bootstrap/db-tools.js

@@ -421,10 +421,14 @@ export function registerDbTools(pi) {
                 depends: Type.Array(Type.String(), { description: "Slice dependency IDs" }),
                 demo: Type.String({ description: "Roadmap demo text / After this" }),
                 goal: Type.String({ description: "Slice goal" }),
-                successCriteria: Type.String({ description: "Slice success criteria block" }),
-                proofLevel: Type.String({ description: "Slice proof level" }),
-                integrationClosure: Type.String({ description: "Slice integration closure" }),
-                observabilityImpact: Type.String({ description: "Slice observability impact" }),
+                // ADR-011: heavy planning fields are optional for sketch slices; required for full slices.
+                successCriteria: Type.Optional(Type.String({ description: "Slice success criteria block (required for full slices; omit for sketches)" })),
+                proofLevel: Type.Optional(Type.String({ description: "Slice proof level (required for full slices; omit for sketches)" })),
+                integrationClosure: Type.Optional(Type.String({ description: "Slice integration closure (required for full slices; omit for sketches)" })),
+                observabilityImpact: Type.Optional(Type.String({ description: "Slice observability impact (required for full slices; omit for sketches)" })),
+                // ADR-011 sketch-then-refine fields.
+                isSketch: Type.Optional(Type.Boolean({ description: "ADR-011: true marks this slice as a sketch awaiting refine-slice expansion" })),
+                sketchScope: Type.Optional(Type.String({ description: "ADR-011: 2–3 sentence scope boundary, required when isSketch=true" })),
             }), { description: "Planned slices for the milestone" }),
             // ── Enrichment metadata (optional — defaults to empty) ────────────
             status: Type.Optional(Type.String({ description: "Milestone status (defaults to active)" })),
@@ -589,6 +593,20 @@ export function registerDbTools(pi) {
             keyFiles: Type.Optional(Type.Array(Type.String(), { description: "List of key files created or modified" })),
             keyDecisions: Type.Optional(Type.Array(Type.String(), { description: "List of key decisions made during this task" })),
             blockerDiscovered: Type.Optional(Type.Boolean({ description: "Whether a plan-invalidating blocker was discovered" })),
+            // ADR-011 Phase 2: mid-execution escalation — agent asks the user to resolve an ambiguity.
+            escalation: Type.Optional(Type.Object({
+                question: Type.String({ description: "The question the user needs to answer — one clear sentence." }),
+                options: Type.Array(Type.Object({
+                    id: Type.String({ description: "Short id (e.g. 'A', 'B') used by /gsd escalate resolve." }),
+                    label: Type.String({ description: "One-line label." }),
+                    tradeoffs: Type.String({ description: "1-2 sentences on the tradeoffs of this option." }),
+                }), { minItems: 2, maxItems: 4, description: "2–4 options the user can choose between." }),
+                recommendation: Type.String({ description: "Option id the executor recommends." }),
+                recommendationRationale: Type.String({ description: "Why the recommendation — 1–2 sentences." }),
+                continueWithDefault: Type.Boolean({
+                    description: "When true, loop continues (artifact logged for later review). When false, auto-mode pauses until the user resolves via /gsd escalate resolve.",
+                }),
+            }, { description: "ADR-011 Phase 2: optional escalation payload. Only honored when phases.mid_execution_escalation is true." })),
             verificationEvidence: Type.Optional(Type.Array(Type.Union([
                 Type.Object({
                     command: Type.String({ description: "Verification command that was run" }),

package/dist/resources/extensions/gsd/bootstrap/provider-error-resume.js

@@ -1,5 +1,6 @@
 import { getAutoDashboardData, startAuto } from "../auto.js";
 import { resetTransientRetryState } from "./agent-end-recovery.js";
+import { resetSessionTimeoutState } from "../auto/phases.js";
 const defaultDeps = {
     getSnapshot: () => getAutoDashboardData(),
     startAuto,
@@ -14,10 +15,11 @@ export async function resumeAutoAfterProviderDelay(pi, ctx, deps = defaultDeps)
         ctx.ui.notify("Provider error recovery delay elapsed, but no paused auto-mode base path was available. Leaving auto-mode paused.", "warning");
         return "missing-base";
     }
-    // Reset the transient retry counter before restarting — without this,
-    // consecutiveTransientCount accumulates across pause/resume cycles and
-    // permanently locks out auto-resume after MAX_TRANSIENT_AUTO_RESUMES errors.
+    // Reset retry counters before restarting — without this, counters
+    // accumulate across pause/resume cycles and permanently lock out
+    // auto-resume after their respective MAX thresholds.
     resetTransientRetryState();
+    resetSessionTimeoutState();
     await deps.startAuto(ctx, pi, snapshot.basePath, false, { step: snapshot.stepMode });
     return "resumed";
 }

package/dist/resources/extensions/gsd/bootstrap/query-tools.js

@@ -2,6 +2,7 @@
 import { Type } from "@sinclair/typebox";
 import { ensureDbOpen } from "./dynamic-tools.js";
 import { executeMilestoneStatus } from "../tools/workflow-tool-executors.js";
+import { checkpointDatabase } from "../gsd-db.js";
 export function registerQueryTools(pi) {
     pi.registerTool({
         name: "gsd_milestone_status",
@@ -27,4 +28,32 @@ export function registerQueryTools(pi) {
             return executeMilestoneStatus(params);
         },
     });
+    pi.registerTool({
+        name: "gsd_checkpoint_db",
+        label: "Checkpoint GSD Database",
+        description: "Flush the SQLite WAL (Write-Ahead Log) into the base gsd.db file. " +
+            "Call this before `git add .gsd/gsd.db` to ensure the committed database " +
+            "contains current milestone/slice/task state rather than stale pre-session content. " +
+            "Safe to call at any time while GSD is running.",
+        promptSnippet: "Flush WAL into gsd.db so git add stages current state",
+        promptGuidelines: [
+            "Call gsd_checkpoint_db immediately before staging .gsd/gsd.db with git add.",
+            "Do not use sqlite3 or shell commands to checkpoint — they are blocked. Use this tool instead.",
+        ],
+        parameters: Type.Object({}),
+        async execute(_toolCallId, _params, _signal, _onUpdate, _ctx) {
+            const dbAvailable = await ensureDbOpen();
+            if (!dbAvailable) {
+                return {
+                    content: [{ type: "text", text: "Error: GSD database is not available. Cannot checkpoint." }],
+                    details: { operation: "checkpoint_db", error: "db_unavailable" },
+                };
+            }
+            checkpointDatabase();
+            return {
+                content: [{ type: "text", text: "WAL checkpoint complete. gsd.db is now up to date and safe to stage with git add." }],
+                details: { operation: "checkpoint_db", status: "ok" },
+            };
+        },
+    });
 }

package/dist/resources/extensions/gsd/bootstrap/register-extension.js

@@ -1,7 +1,7 @@
 // GSD2 — Extension registration: wires all GSD tools, commands, and hooks into pi
-import { registerGSDCommand } from "../commands.js";
 import { registerExitCommand } from "../exit-command.js";
 import { registerWorktreeCommand } from "../worktree-command.js";
+import { loadEcosystemExtensions } from "../ecosystem/loader.js";
 import { registerDbTools } from "./db-tools.js";
 import { registerDynamicTools } from "./dynamic-tools.js";
 import { registerJournalTools } from "./journal-tools.js";
@@ -9,6 +9,7 @@ import { registerQueryTools } from "./query-tools.js";
 import { registerHooks } from "./register-hooks.js";
 import { registerShortcuts } from "./register-shortcuts.js";
 import { writeCrashLog } from "./crash-log.js";
+import { logWarning } from "../workflow-logger.js";
 export { writeCrashLog } from "./crash-log.js";
 export function handleRecoverableExtensionProcessError(err) {
     if (err.code === "EPIPE") {
@@ -52,20 +53,41 @@ function installEpipeGuard() {
     }
 }
 export function registerGsdExtension(pi) {
-    registerGSDCommand(pi);
+    // Note: registerGSDCommand is called by index.ts before this function,
+    // so we intentionally skip it here to avoid double-registration.
     registerWorktreeCommand(pi);
     registerExitCommand(pi);
     installEpipeGuard();
+    // Ecosystem handlers captured by the GSDExtensionAPI wrapper for the
+    // GSD-owned `before_agent_start` dispatch step (#3338).
+    const ecosystemHandlers = [];
     pi.registerCommand("kill", {
         description: "Exit GSD immediately (no cleanup)",
         handler: async (_args, _ctx) => {
             process.exit(0);
         },
     });
-    registerDynamicTools(pi);
-    registerDbTools(pi);
-    registerJournalTools(pi);
-    registerQueryTools(pi);
-    registerShortcuts(pi);
-    registerHooks(pi);
+    // Wrap non-critical registrations individually so one failure
+    // doesn't prevent the others from loading.
+    const nonCriticalRegistrations = [
+        ["dynamic-tools", () => registerDynamicTools(pi)],
+        ["db-tools", () => registerDbTools(pi)],
+        ["journal-tools", () => registerJournalTools(pi)],
+        ["query-tools", () => registerQueryTools(pi)],
+        ["shortcuts", () => registerShortcuts(pi)],
+        ["hooks", () => registerHooks(pi, ecosystemHandlers)],
+        ["ecosystem", () => {
+                void loadEcosystemExtensions(pi, ecosystemHandlers).catch((err) => {
+                    logWarning("ecosystem", `loader failed: ${err instanceof Error ? err.message : String(err)}`);
+                });
+            }],
+    ];
+    for (const [name, register] of nonCriticalRegistrations) {
+        try {
+            register();
+        }
+        catch (err) {
+            logWarning("bootstrap", `Failed to register ${name}: ${err instanceof Error ? err.message : String(err)}`);
+        }
+    }
 }

package/dist/resources/extensions/gsd/bootstrap/register-hooks.js

@@ -1,5 +1,7 @@
 import { join } from "node:path";
 import { isToolCallEventType } from "@gsd/pi-coding-agent";
+import { updateSnapshot } from "../ecosystem/gsd-extension-api.js";
+import { getEcosystemReadyPromise } from "../ecosystem/loader.js";
 import { buildMilestoneFileName, resolveMilestonePath, resolveSliceFile, resolveSlicePath } from "../paths.js";
 import { buildBeforeAgentStartResult } from "./system-context.js";
 import { handleAgentEnd } from "./agent-end-recovery.js";
@@ -11,6 +13,7 @@ import { loadToolApiKeys } from "../commands-config.js";
 import { loadFile, saveFile, formatContinue } from "../files.js";
 import { deriveState } from "../state.js";
 import { getAutoDashboardData, isAutoActive, isAutoPaused, markToolEnd, markToolStart, recordToolInvocationError } from "../auto.js";
+import { hideFooter } from "../auto-dashboard.js";
 import { isParallelActive, shutdownParallel } from "../parallel-orchestrator.js";
 import { checkToolCallLoop, resetToolCallLoopGuard } from "./tool-call-loop-guard.js";
 import { saveActivityLog } from "../activity-log.js";
@@ -29,7 +32,7 @@ async function syncServiceTierStatus(ctx) {
     const { getEffectiveServiceTier, formatServiceTierFooterStatus } = await import("../service-tier.js");
     ctx.ui.setStatus("gsd-fast", formatServiceTierFooterStatus(getEffectiveServiceTier(), ctx.model?.id));
 }
-export function registerHooks(pi) {
+export function registerHooks(pi, ecosystemHandlers) {
     pi.on("session_start", async (_event, ctx) => {
         initNotificationStore(process.cwd());
         installNotifyInterceptor(ctx);
@@ -39,8 +42,12 @@ export function registerHooks(pi) {
         resetToolCallLoopGuard();
         resetAskUserQuestionsCache();
         await syncServiceTierStatus(ctx);
-        const { prepareWorkflowMcpForProject } = await import("../workflow-mcp-auto-prep.js");
-        prepareWorkflowMcpForProject(ctx, process.cwd());
+        // Skip MCP auto-prep when running inside an auto-worktree (see session_switch below).
+        const { isInAutoWorktree } = await import("../auto-worktree.js");
+        if (!isInAutoWorktree(process.cwd())) {
+            const { prepareWorkflowMcpForProject } = await import("../workflow-mcp-auto-prep.js");
+            prepareWorkflowMcpForProject(ctx, process.cwd());
+        }
         // Apply show_token_cost preference (#1515)
         try {
             const { loadEffectiveGSDPreferences } = await import("../preferences.js");
@@ -71,6 +78,9 @@ export function registerHooks(pi) {
             catch { /* non-fatal */ }
         }
         loadToolApiKeys();
+        if (isAutoActive()) {
+            ctx.ui.setFooter(hideFooter);
+        }
     });
     pi.on("session_switch", async (_event, ctx) => {
         initNotificationStore(process.cwd());
@@ -80,12 +90,59 @@ export function registerHooks(pi) {
         resetAskUserQuestionsCache();
         clearDiscussionFlowState();
         await syncServiceTierStatus(ctx);
-        const { prepareWorkflowMcpForProject } = await import("../workflow-mcp-auto-prep.js");
-        prepareWorkflowMcpForProject(ctx, process.cwd());
+        // Skip MCP auto-prep when running inside an auto-worktree. The worktree
+        // already has .mcp.json from createAutoWorktree, and re-running the writer
+        // post-chdir rewrites the file mid-run (non-idempotent due to cwd-relative
+        // CLI path resolution), dirtying the tree and breaking the milestone merge.
+        const { isInAutoWorktree } = await import("../auto-worktree.js");
+        if (!isInAutoWorktree(process.cwd())) {
+            const { prepareWorkflowMcpForProject } = await import("../workflow-mcp-auto-prep.js");
+            prepareWorkflowMcpForProject(ctx, process.cwd());
+        }
         loadToolApiKeys();
+        if (isAutoActive()) {
+            ctx.ui.setFooter(hideFooter);
+        }
     });
     pi.on("before_agent_start", async (event, ctx) => {
-        return buildBeforeAgentStartResult(event, ctx);
+        // Wait for ecosystem loader to finish (no-op after first turn).
+        await getEcosystemReadyPromise();
+        // GSD's own context injection (existing behavior — unchanged).
+        const gsdResult = await buildBeforeAgentStartResult(event, ctx);
+        // Refresh the snapshot used by ecosystem getPhase()/getActiveUnit().
+        // deriveState has its own ~100ms cache so this is cheap on repeat calls.
+        try {
+            const state = await deriveState(process.cwd());
+            updateSnapshot(state);
+        }
+        catch {
+            updateSnapshot(null);
+        }
+        // Chain ecosystem handlers using pi's runner.ts chaining protocol:
+        // each handler sees the systemPrompt mutated by prior handlers.
+        let currentSystemPrompt = gsdResult?.systemPrompt ?? event.systemPrompt;
+        // `any` because pi's BeforeAgentStartEventResult.message uses an internal
+        // CustomMessage type that's not re-exported (see ecosystem/gsd-extension-api.ts).
+        let lastMessage = gsdResult?.message;
+        for (const handler of ecosystemHandlers) {
+            try {
+                const r = await handler({ ...event, systemPrompt: currentSystemPrompt }, ctx);
+                if (r?.systemPrompt !== undefined)
+                    currentSystemPrompt = r.systemPrompt;
+                if (r?.message)
+                    lastMessage = r.message;
+            }
+            catch (err) {
+                safetyLogWarning("ecosystem", `before_agent_start handler failed: ${err instanceof Error ? err.message : String(err)}`);
+            }
+        }
+        // Compose result. Return undefined if nothing changed (preserves runner contract).
+        if (currentSystemPrompt === event.systemPrompt && !lastMessage)
+            return undefined;
+        return {
+            systemPrompt: currentSystemPrompt !== event.systemPrompt ? currentSystemPrompt : undefined,
+            message: lastMessage,
+        };
     });
     pi.on("agent_end", async (event, ctx) => {
         resetToolCallLoopGuard();
@@ -116,8 +173,10 @@ export function registerHooks(pi) {
         const state = await deriveState(basePath);
         if (!state.activeMilestone || !state.activeSlice || !state.activeTask)
             return;
-        if (state.phase !== "executing")
-            return;
+        // Write checkpoint for ALL phases, not just "executing" — discuss, research,
+        // and planning also carry in-memory state (user answers, gate verification)
+        // that would be lost on compaction (#4258).
+        // if (state.phase !== "executing") return;
         const sliceDir = resolveSlicePath(basePath, state.activeMilestone.id, state.activeSlice.id);
         if (!sliceDir)
             return;
@@ -240,7 +299,8 @@ export function registerHooks(pi) {
     pi.on("tool_call", async (event, ctx) => {
         if (!isAutoActive())
             return;
-        safetyRecordToolCall(event.toolName, event.input);
+        markToolStart(event.toolCallId, event.toolName);
+        safetyRecordToolCall(event.toolCallId, event.toolName, event.input);
         // Destructive command classification (warn only, never block)
         if (isToolCallEventType("bash", event)) {
             const classification = classifyCommand(event.input.command);
@@ -253,6 +313,20 @@ export function registerHooks(pi) {
         }
     });
     pi.on("tool_result", async (event) => {
+        if (isAutoActive() && typeof event.toolCallId === "string") {
+            markToolEnd(event.toolCallId);
+        }
+        if (isAutoActive() && event.isError && event.toolName.startsWith("gsd_")) {
+            const resultPayload = ("result" in event ? event.result : undefined);
+            const errorText = typeof resultPayload === "string"
+                ? resultPayload
+                : (typeof resultPayload?.content?.[0]?.text === "string"
+                    ? resultPayload.content[0].text
+                    : (typeof event.content === "string"
+                        ? event.content
+                        : String(resultPayload ?? "")));
+            recordToolInvocationError(event.toolName, errorText);
+        }
         if (event.toolName !== "ask_user_questions")
             return;
         const milestoneId = getDiscussionMilestoneId(process.cwd());