gsd-pi 2.76.0-dev.4100bd590 → 2.76.0-dev.479ad0e78

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

@@ -17,7 +17,7 @@
  */
 
 import type { Api, Model } from "@gsd/pi-ai";
-import type { ExtensionCommandContext } from "@gsd/pi-coding-agent";
+import type { ExtensionAPI, ExtensionCommandContext } from "@gsd/pi-coding-agent";
 import type { GitServiceImpl } from "../git-service.js";
 import type { CaptureEntry } from "../captures.js";
 import type { BudgetAlertLevel } from "../auto-budget.js";
@@ -40,6 +40,8 @@ export interface StartModel {
   id: string;
 }
 
+export type ThinkingLevelSnapshot = ReturnType<ExtensionAPI["getThinkingLevel"]>;
+
 export interface PendingVerificationRetry {
   unitId: string;
   failureContext: string;
@@ -62,6 +64,15 @@ export interface SidecarItem {
   captureId?: string;
 }
 
+export interface PreExecFailure {
+  /** Milestone/slice that failed (e.g. "M001/S02"). */
+  unitId: string;
+  /** Verbatim blocking check strings from the failed gate run. */
+  blockingFindings: string[];
+  /** Condensed gate verdict excerpt for context (status + rationale). */
+  verdictExcerpt: string;
+}
+
 // ─── Constants ───────────────────────────────────────────────────────────────
 
 export const MAX_UNIT_DISPATCHES = 3;
@@ -120,6 +131,8 @@ export class AutoSession {
   currentDispatchedModelId: string | null = null;
   originalModelId: string | null = null;
   originalModelProvider: string | null = null;
+  autoModeStartThinkingLevel: ThinkingLevelSnapshot | null = null;
+  originalThinkingLevel: ThinkingLevelSnapshot | null = null;
   lastBudgetAlertLevel: BudgetAlertLevel = 0;
 
   // ── Recovery ─────────────────────────────────────────────────────────────
@@ -135,6 +148,18 @@ export class AutoSession {
   // ── Sidecar queue ─────────────────────────────────────────────────────
   sidecarQueue: SidecarItem[] = [];
 
+  // ── Pre-exec gate failure context (#4551) ───────────────────────────
+  /**
+   * Persisted when a pre-execution gate fails on a plan-slice or refine-slice
+   * unit. The planning → plan-slice dispatch rule reads this field and injects
+   * the failure details into the next re-dispatch prompt so the LLM can fix the
+   * specific issues instead of producing an identical plan.
+   *
+   * Cleared after it has been consumed (injected into the prompt) to avoid
+   * stale context bleeding into unrelated slices.
+   */
+  lastPreExecFailure: PreExecFailure | null = null;
+
   // ── Tool invocation errors (#2883) ──────────────────────────────────
   /** Set when a GSD tool execution ends with isError due to malformed/truncated
    *  JSON arguments. Checked by postUnitPreVerification to break retry loops. */
@@ -195,7 +220,12 @@ export class AutoSession {
   }
 
   get lockBasePath(): string {
-    return this.originalBasePath || this.basePath;
+    // Prefer originalBasePath (project root); fall back to basePath.
+    // Strip /.gsd/worktrees/ suffix if basePath is itself a worktree path
+    // to avoid reading/writing the lock inside the worktree (#3729).
+    const resolved = this.originalBasePath || this.basePath;
+    const markerIdx = resolved.indexOf("/.gsd/worktrees/");
+    return markerIdx !== -1 ? resolved.slice(0, markerIdx) : resolved;
   }
 
   reset(): void {
@@ -241,6 +271,8 @@ export class AutoSession {
     this.currentDispatchedModelId = null;
     this.originalModelId = null;
     this.originalModelProvider = null;
+    this.autoModeStartThinkingLevel = null;
+    this.originalThinkingLevel = null;
     this.lastBudgetAlertLevel = 0;
 
     // Recovery
@@ -261,6 +293,7 @@ export class AutoSession {
     this.sidecarQueue = [];
     this.rewriteAttemptCount = 0;
     this.consecutiveCompleteBootstraps = 0;
+    this.lastPreExecFailure = null;
     this.lastToolInvocationError = null;
     this.lastGitActionFailure = null;
     this.lastGitActionStatus = null;

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

@@ -568,15 +568,28 @@ export const DISPATCH_RULES: DispatchRule[] = [
   },
   {
     name: "planning → plan-slice",
-    match: async ({ state, mid, midTitle, basePath, sessionContextWindow, modelRegistry }) => {
+    match: async ({ state, mid, midTitle, basePath, sessionContextWindow, modelRegistry, session }) => {
       if (state.phase !== "planning") return null;
       if (!state.activeSlice) return missingSliceStop(mid, state.phase);
       const sid = state.activeSlice!.id;
       const sTitle = state.activeSlice!.title;
+      // #4551: Consume any persisted pre-exec failure for this slice so the
+      // re-dispatched prompt includes the exact blocked references. Clear the
+      // field immediately after reading to prevent stale context leaking into
+      // a later, unrelated plan-slice run.
+      const unitId = `${mid}/${sid}`;
+      let priorPreExecFailure: { blockingFindings: string[]; verdictExcerpt: string } | undefined;
+      if (session?.lastPreExecFailure?.unitId === unitId) {
+        priorPreExecFailure = {
+          blockingFindings: session.lastPreExecFailure.blockingFindings,
+          verdictExcerpt: session.lastPreExecFailure.verdictExcerpt,
+        };
+        session.lastPreExecFailure = null;
+      }
       return {
         action: "dispatch",
         unitType: "plan-slice",
-        unitId: `${mid}/${sid}`,
+        unitId,
         prompt: await buildPlanSlicePrompt(
           mid,
           midTitle,
@@ -584,7 +597,7 @@ export const DISPATCH_RULES: DispatchRule[] = [
           sTitle,
           basePath,
           undefined,
-          { sessionContextWindow, modelRegistry },
+          { sessionContextWindow, modelRegistry, priorPreExecFailure },
         ),
       };
     },

package/src/resources/extensions/gsd/auto-model-selection.ts

@@ -33,6 +33,14 @@ export interface PreferredModelConfig {
   source: "explicit" | "synthesized";
 }
 
+function reapplyThinkingLevel(
+  pi: ExtensionAPI,
+  level: ReturnType<ExtensionAPI["getThinkingLevel"]> | null | undefined,
+): void {
+  if (!level) return;
+  pi.setThinkingLevel(level);
+}
+
 export function resolvePreferredModelConfig(
   unitType: string,
   autoModeStartModel: { provider: string; id: string; flatRateCtx?: FlatRateContext } | null,
@@ -97,6 +105,8 @@ export async function selectAndApplyModel(
   isAutoMode = true,
   /** Explicit /gsd model pin captured at bootstrap for long-running auto loops. */
   sessionModelOverride?: { provider: string; id: string } | null,
+  /** Thinking level captured at auto-mode start and re-applied after model swaps. */
+  autoModeStartThinkingLevel?: ReturnType<ExtensionAPI["getThinkingLevel"]> | null,
 ): Promise<ModelSelectionResult> {
   const uokFlags = resolveUokFlags(prefs);
   const effectiveSessionModelOverride = sessionModelOverride === undefined
@@ -380,11 +390,12 @@ export async function selectAndApplyModel(
       const ok = await pi.setModel(model, { persist: false });
       if (ok) {
         appliedModel = model;
+        reapplyThinkingLevel(pi, autoModeStartThinkingLevel);
 
         // ADR-005: Adjust active tool set for the selected model's provider capabilities.
         // Hard-filter incompatible tools, then let extensions override via adjust_tool_set hook.
         const activeToolNames = pi.getActiveTools();
-        const { toolNames: compatibleTools, removedTools } = adjustToolSet(activeToolNames, model.api);
+        const { toolNames: compatibleTools, removedTools } = adjustToolSet(activeToolNames, model.api, model.provider);
         let finalToolNames = compatibleTools;
 
         // Fire adjust_tool_set hook — extensions can override the filtered tool set
@@ -456,10 +467,14 @@ export async function selectAndApplyModel(
           );
           if (byId) {
             const fallbackOk = await pi.setModel(byId, { persist: false });
-            if (fallbackOk) appliedModel = byId;
+            if (fallbackOk) {
+              appliedModel = byId;
+              reapplyThinkingLevel(pi, autoModeStartThinkingLevel);
+            }
           }
         } else {
           appliedModel = startModel;
+          reapplyThinkingLevel(pi, autoModeStartThinkingLevel);
         }
       }
     }

package/src/resources/extensions/gsd/auto-post-unit.ts

@@ -55,7 +55,7 @@ import { hasPendingCaptures, loadPendingCaptures, revertExecutorResolvedCaptures
 import { debugLog } from "./debug-logger.js";
 import { runSafely } from "./auto-utils.js";
 import type { AutoSession, SidecarItem } from "./auto/session.js";
-import { getEvidence } from "./safety/evidence-collector.js";
+import { getEvidence, clearEvidenceFromDisk } from "./safety/evidence-collector.js";
 import { validateFileChanges } from "./safety/file-change-validator.js";
 // crossReferenceEvidence available for future use when verification_evidence is stored in DB
 // import { crossReferenceEvidence, type ClaimedEvidence } from "./safety/evidence-cross-ref.js";
@@ -107,7 +107,6 @@ import {
   updateProgressWidget as _updateProgressWidget,
   updateSliceProgressCache,
   unitVerb,
-  hideFooter,
   describeNextUnit,
 } from "./auto-dashboard.js";
 import { existsSync, unlinkSync } from "node:fs";
@@ -654,7 +653,7 @@ export async function postUnitPreVerification(pctx: PostUnitContext, opts?: PreV
             if (taskRow) {
               const expectedOutput = taskRow.expected_output ?? [];
               const plannedFiles = taskRow.files ?? [];
-              const audit = validateFileChanges(s.basePath, expectedOutput, plannedFiles);
+              const audit = validateFileChanges(s.basePath, expectedOutput, plannedFiles, safetyConfig.file_change_allowlist);
               if (audit && audit.violations.length > 0) {
                 const warnings = audit.violations.filter(v => v.severity === "warning");
                 for (const v of warnings) {
@@ -712,6 +711,16 @@ export async function postUnitPreVerification(pctx: PostUnitContext, opts?: PreV
             debugLog("postUnit", { phase: "safety-content-validation", error: String(e) });
           }
         }
+
+        // Clear persisted evidence file now that post-unit processing is complete
+        // (Bug #4385 — prevents stale evidence from affecting retries of same unit ID).
+        if (safetyConfig.evidence_collection && s.currentUnit.type === "execute-task" && sMid && sSid && sTid) {
+          try {
+            clearEvidenceFromDisk(s.basePath, sMid, sSid, sTid);
+          } catch (e) {
+            debugLog("postUnit", { phase: "safety-evidence-clear", error: String(e) });
+          }
+        }
       }
     } catch (e) {
       debugLog("postUnit", { phase: "safety-harness", error: String(e) });
@@ -1134,6 +1143,15 @@ export async function postUnitPostVerification(pctx: PostUnitContext): Promise<"
             `Pre-execution checks failed: ${blockingCount} blocking issue${blockingCount === 1 ? "" : "s"} found\n${details}${suffix}${evidenceNote}`,
             "error",
           );
+          // Persist failure context so the next plan-slice re-dispatch can inject
+          // it into the prompt and break the infinite loop (#4551).
+          s.lastPreExecFailure = {
+            unitId: currentUnit.id,
+            blockingFindings: blockingChecks.map(
+              c => `[${c.category}] ${c.target}: ${c.message}`,
+            ),
+            verdictExcerpt: `status=${result.status}; ${blockingCount} blocking issue${blockingCount === 1 ? "" : "s"} detected`,
+          };
           preExecPauseNeeded = true;
         } else if (result.status === "warn") {
           ctx.ui.notify(
@@ -1142,6 +1160,14 @@ export async function postUnitPostVerification(pctx: PostUnitContext): Promise<"
           );
           // Strict mode: treat warnings as blocking
           if (prefs?.enhanced_verification_strict === true) {
+            const warnChecks = result.checks.filter(c => !c.passed);
+            s.lastPreExecFailure = {
+              unitId: currentUnit.id,
+              blockingFindings: warnChecks.map(
+                c => `[${c.category}] ${c.target}: ${c.message}`,
+              ),
+              verdictExcerpt: `status=${result.status} (strict mode); ${warnChecks.length} warning${warnChecks.length === 1 ? "" : "s"} treated as blocking`,
+            };
             preExecPauseNeeded = true;
           }
         }

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

@@ -1380,7 +1380,18 @@ async function renderSlicePrompt(options: {
 
 export async function buildPlanSlicePrompt(
   mid: string, _midTitle: string, sid: string, sTitle: string, base: string, level?: InlineLevel,
-  options?: { softScopeHint?: string; sessionContextWindow?: number; modelRegistry?: MinimalModelRegistry },
+  options?: {
+    softScopeHint?: string;
+    sessionContextWindow?: number;
+    modelRegistry?: MinimalModelRegistry;
+    /** Failure context from a prior pre-exec gate run (#4551). When present, a
+     *  "Fix these specific issues" section is appended so the LLM addresses the
+     *  exact problems instead of producing an identical plan that fails again. */
+    priorPreExecFailure?: {
+      blockingFindings: string[];
+      verdictExcerpt: string;
+    };
+  },
 ): Promise<string> {
   const prependBlocks: string[] = [];
   // ADR-011: when the refining-phase dispatch rule gracefully downgrades to
@@ -1393,6 +1404,22 @@ export async function buildPlanSlicePrompt(
       `This scope was captured during an earlier progressive-planning pass that was later disabled. Treat it as context only — you may plan beyond it if the work genuinely requires more scope. Do NOT treat this as a hard boundary.`,
     );
   }
+  // #4551: inject pre-exec failure context so the re-dispatched plan-slice
+  // addresses the exact blocked references rather than reproducing the same plan.
+  if (options?.priorPreExecFailure) {
+    const { blockingFindings, verdictExcerpt } = options.priorPreExecFailure;
+    const findingsList = blockingFindings.length > 0
+      ? blockingFindings.map(f => `- ${f}`).join("\n")
+      : "- (no specific findings recorded)";
+    prependBlocks.push(
+      `## Fix these specific issues from the prior pre-exec check\n\n` +
+      `The previous plan-slice attempt was blocked by pre-execution validation.\n` +
+      `Gate verdict: ${verdictExcerpt}\n\n` +
+      `Blocked references that must be resolved in this plan:\n${findingsList}\n\n` +
+      `Revise the plan so that every reference listed above is satisfied before execution begins. ` +
+      `Do not reproduce the same file paths, package names, or task ordering that caused these failures.`,
+    );
+  }
   return renderSlicePrompt({
     mid, sid, sTitle, base,
     level: level ?? resolveInlineLevel(),

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

@@ -13,7 +13,7 @@ import { appendEvent } from "./workflow-events.js";
 import { atomicWriteSync } from "./atomic-write.js";
 import { clearParseCache } from "./files.js";
 import { parseRoadmap as parseLegacyRoadmap, parsePlan as parseLegacyPlan } from "./parsers-legacy.js";
-import { isDbAvailable, getTask, getSlice, getSliceTasks, getPendingGates, updateTaskStatus, updateSliceStatus } from "./gsd-db.js";
+import { isDbAvailable, getTask, getSlice, getSliceTasks, getPendingGates, updateTaskStatus, updateSliceStatus, insertSlice } from "./gsd-db.js";
 import { isValidationTerminal } from "./state.js";
 import { getErrorMessage } from "./error-utils.js";
 import { logWarning, logError } from "./workflow-logger.js";
@@ -268,11 +268,26 @@ export function verifyExpectedArtifact(
   // RESEARCH file. Without this, resolveExpectedArtifactPath returns null and
   // the retry/escalation machinery silently re-dispatches forever.
   //
+  // #4068: Also treat a PARALLEL-BLOCKER placeholder as a terminal completion
+  // so that timeout-recovery can write the blocker, have verifyExpectedArtifact
+  // return true, and let the dispatch loop advance past this unit.  Without
+  // this, the blocker is written but verification still returns false, the unit
+  // is never cleared from unitDispatchCount, and on the next iteration the
+  // dispatch rule (which correctly skips parallel-research when PARALLEL-BLOCKER
+  // exists) returns null — leaving the loop stuck re-deriving indefinitely.
+  //
   // NOTE: this predicate mirrors the dispatch rule at
   // auto-dispatch.ts parallel-research-slices — keep the two in sync.
   if (unitType === "research-slice" && unitId.endsWith("/parallel-research")) {
     const { milestone: mid } = parseUnitId(unitId);
     if (!mid) return false;
+
+    // #4068: PARALLEL-BLOCKER written by timeout-recovery is a terminal state.
+    const blockerPath = resolveExpectedArtifactPath(unitType, unitId, base);
+    if (blockerPath && existsSync(blockerPath)) {
+      return true;
+    }
+
     const roadmapFile = resolveMilestoneFile(base, mid, "ROADMAP");
     if (!roadmapFile || !existsSync(roadmapFile)) {
       logWarning("recovery", `verify-fail ${unitType} ${unitId}: roadmap missing`);
@@ -532,6 +547,16 @@ export function writeBlockerPlaceholder(
       try { updateSliceStatus(mid, sid, "complete", ts); } catch (e) { logWarning("recovery", `updateSliceStatus failed during context exhaustion: ${e instanceof Error ? e.message : String(e)}`); }
       try { appendEvent(base, { cmd: "complete-slice", params: { milestoneId: mid, sliceId: sid }, ts, actor: "system", trigger_reason: "blocker-placeholder-recovery" }); } catch (e) { logWarning("recovery", `appendEvent failed for slice recovery: ${e instanceof Error ? e.message : String(e)}`); }
     }
+    // Insert a placeholder complete slice so deriveState sees activeMilestoneSlices.length > 0
+    // and exits the pre-planning phase. Without this, activeMilestoneSlices stays empty
+    // after the blocker ROADMAP.md is written, causing deriveState to return phase:'pre-planning'
+    // indefinitely and re-dispatching plan-milestone in an infinite loop (#4378).
+    if (unitType === "plan-milestone" && mid) {
+      try {
+        insertSlice({ id: "S00-blocker", milestoneId: mid, title: "Blocker placeholder — planning failed", status: "complete", sequence: 0 });
+      } catch (e) { logWarning("recovery", `insertSlice placeholder failed for plan-milestone recovery: ${e instanceof Error ? e.message : String(e)}`); }
+      try { appendEvent(base, { cmd: "plan-milestone", params: { milestoneId: mid }, ts, actor: "system", trigger_reason: "blocker-placeholder-recovery" }); } catch (e) { logWarning("recovery", `appendEvent failed for plan-milestone recovery: ${e instanceof Error ? e.message : String(e)}`); }
+    }
   }
 
   return diagnoseExpectedArtifact(unitType, unitId, base);

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

@@ -60,8 +60,8 @@ import { initRoutingHistory } from "./routing-history.js";
 import { restoreHookState, resetHookState } from "./post-unit-hooks.js";
 import { resetProactiveHealing, setLevelChangeCallback } from "./doctor-proactive.js";
 import { snapshotSkills } from "./skill-discovery.js";
-import { isDbAvailable, getMilestone, openDatabase } from "./gsd-db.js";
-import { hideFooter } from "./auto-dashboard.js";
+import { isDbAvailable, getMilestone, openDatabase, getDbStatus } from "./gsd-db.js";
+
 import {
   debugLog,
   enableDebug,
@@ -92,7 +92,7 @@ import type { WorktreeResolver } from "./worktree-resolver.js";
 import { getSessionModelOverride } from "./session-model-override.js";
 
 export interface BootstrapDeps {
-  shouldUseWorktreeIsolation: () => boolean;
+  shouldUseWorktreeIsolation: (basePath?: string) => boolean;
   registerSigtermHandler: (basePath: string) => void;
   lockBase: () => string;
   buildResolver: () => WorktreeResolver;
@@ -273,8 +273,8 @@ export async function bootstrapAutoSession(
   //
   // Precedence:
   // 1) Explicit session override via /gsd model (this session)
-  // 2) GSD model preferences from PREFERENCES.md (validated against live auth)
-  // 3) Current session model from settings/session restore (if provider ready)
+  // 2) Current session model from settings/session restore (if provider ready)
+  // 3) GSD model preferences from PREFERENCES.md (validated against live auth)
   //
   // This preserves #3517 defaults while honoring explicit runtime model
   // selection for subsequent /gsd runs in the same session.
@@ -314,11 +314,14 @@ export async function bootstrapAutoSession(
   }
   const sessionModelReady =
     ctx.model && ctx.modelRegistry.isProviderRequestReady(ctx.model.provider);
+  const currentSessionModel = (sessionModelReady && ctx.model)
+    ? { provider: ctx.model.provider, id: ctx.model.id }
+    : null;
+  const startThinkingSnapshot = pi.getThinkingLevel();
   const startModelSnapshot = manualSessionOverride
+    ?? currentSessionModel
     ?? validatedPreferredModel
-    ?? (sessionModelReady && ctx.model
-      ? { provider: ctx.model.provider, id: ctx.model.id }
-      : null);
+    ?? null;
 
   try {
     // Validate GSD_PROJECT_ID early so the user gets immediate feedback
@@ -340,7 +343,7 @@ export async function bootstrapAutoSession(
     const hasLocalGit = existsSync(join(base, ".git"));
     if (!hasLocalGit || isInheritedRepo(base)) {
       const mainBranch =
-        loadEffectiveGSDPreferences()?.preferences?.git?.main_branch || "main";
+        loadEffectiveGSDPreferences(base)?.preferences?.git?.main_branch || "main";
       nativeInit(base, mainBranch);
     }
 
@@ -358,7 +361,7 @@ export async function bootstrapAutoSession(
     // Ensure .gitignore has baseline patterns.
     // ensureGitignore checks for git-tracked .gsd/ files and skips the
     // ".gsd" pattern if the project intentionally tracks .gsd/ in git.
-    const gitPrefs = loadEffectiveGSDPreferences()?.preferences?.git;
+    const gitPrefs = loadEffectiveGSDPreferences(base)?.preferences?.git;
     const manageGitignore = gitPrefs?.manage_gitignore;
     ensureGitignore(base, { manageGitignore });
     if (manageGitignore !== false) untrackRuntimeFiles(base);
@@ -387,7 +390,7 @@ export async function bootstrapAutoSession(
     // Initialize GitServiceImpl
     s.gitService = new GitServiceImpl(
       s.basePath,
-      loadEffectiveGSDPreferences()?.preferences?.git ?? {},
+      loadEffectiveGSDPreferences(base)?.preferences?.git ?? {},
     );
 
     // ── Debug mode ──
@@ -431,7 +434,7 @@ export async function bootstrapAutoSession(
     // was lost due to session ending between completion and teardown.
     // Must run after DB open and before worktree entry.
     try {
-      const auditResult = auditOrphanedMilestoneBranches(base, getIsolationMode());
+      const auditResult = auditOrphanedMilestoneBranches(base, getIsolationMode(base));
       for (const msg of auditResult.recovered) {
         ctx.ui.notify(`Orphan audit: ${msg}`, "info");
       }
@@ -451,7 +454,7 @@ export async function bootstrapAutoSession(
     // Stale worktree state recovery (#654)
     if (
       state.activeMilestone &&
-      shouldUseWorktreeIsolation() &&
+      shouldUseWorktreeIsolation(base) &&
       !detectWorktreeName(base)
     ) {
       const wtPath = getAutoWorktreePath(base, state.activeMilestone.id);
@@ -469,7 +472,7 @@ export async function bootstrapAutoSession(
     if (
       state.activeMilestone &&
       (state.phase === "pre-planning" || state.phase === "complete") &&
-      getIsolationMode() !== "none" &&
+      getIsolationMode(base) !== "none" &&
       !detectWorktreeName(base) &&
       !base.includes(`${pathSep}.gsd${pathSep}worktrees${pathSep}`)
     ) {
@@ -550,37 +553,15 @@ export async function bootstrapAutoSession(
         const { showSmartEntry } = await import("./guided-flow.js");
         await showSmartEntry(ctx, pi, base, { step: requestedStepMode });
 
-        invalidateAllCaches();
-        const postState = await deriveState(base);
-        if (
-          postState.activeMilestone &&
-          postState.phase !== "complete" &&
-          postState.phase !== "pre-planning"
-        ) {
-          s.consecutiveCompleteBootstraps = 0; // Successfully advanced past "complete"
-          state = postState;
-        } else if (
-          postState.activeMilestone &&
-          postState.phase === "pre-planning"
-        ) {
-          const contextFile = resolveMilestoneFile(
-            base,
-            postState.activeMilestone.id,
-            "CONTEXT",
-          );
-          const hasContext = !!(contextFile && (await loadFile(contextFile)));
-          if (hasContext) {
-            state = postState;
-          } else {
-            ctx.ui.notify(
-              "Discussion completed but no milestone context was written. Run /gsd to try the discussion again, or /gsd auto after creating the milestone manually.",
-              "warning",
-            );
-            return releaseLockAndReturn();
-          }
-        } else {
-          return releaseLockAndReturn();
-        }
+        // showSmartEntry dispatches via pi.sendMessage() which is fire-and-forget:
+        // it queues the message and returns immediately, before the LLM turn runs.
+        // Checking postState here would always see the pre-dispatch state, causing
+        // the premature "Discussion completed but..." warning (#3420).
+        //
+        // checkAutoStartAfterDiscuss (in guided-flow.ts) already handles re-entering
+        // auto-mode by calling startAutoDetached after the discussion completes.
+        // Release the lock and let the async dispatch proceed.
+        return releaseLockAndReturn();
       }
 
       // Active milestone exists but has no roadmap
@@ -592,17 +573,16 @@ export async function bootstrapAutoSession(
           const { showSmartEntry } = await import("./guided-flow.js");
           await showSmartEntry(ctx, pi, base, { step: requestedStepMode });
 
-          invalidateAllCaches();
-          const postState = await deriveState(base);
-          if (postState.activeMilestone && postState.phase !== "pre-planning") {
-            state = postState;
-          } else {
-            ctx.ui.notify(
-              "Discussion completed but milestone context is still missing. Run /gsd to try again.",
-              "warning",
-            );
-            return releaseLockAndReturn();
-          }
+          // showSmartEntry dispatches via pi.sendMessage() which is fire-and-forget:
+          // it queues the message and returns immediately, before the LLM turn runs.
+          // Checking postState here fires before the LLM has had a turn, so the
+          // pre-planning phase would still appear unchanged and a premature warning
+          // would be emitted (#3420).
+          //
+          // checkAutoStartAfterDiscuss (in guided-flow.ts) already handles re-entering
+          // auto-mode by calling startAutoDetached after the discussion completes.
+          // Release the lock and let the async dispatch proceed.
+          return releaseLockAndReturn();
         }
       }
 
@@ -664,15 +644,16 @@ export async function bootstrapAutoSession(
     s.pendingQuickTasks = [];
     s.currentUnit = null;
     s.currentMilestoneId = state.activeMilestone?.id ?? null;
-    s.originalModelId = ctx.model?.id ?? null;
-    s.originalModelProvider = ctx.model?.provider ?? null;
+    s.originalModelId = startModelSnapshot?.id ?? ctx.model?.id ?? null;
+    s.originalModelProvider = startModelSnapshot?.provider ?? ctx.model?.provider ?? null;
+    s.originalThinkingLevel = startThinkingSnapshot ?? null;
 
     // Register SIGTERM handler
     registerSigtermHandler(base);
 
     // Capture integration branch
     if (s.currentMilestoneId) {
-      if (getIsolationMode() !== "none") {
+      if (getIsolationMode(base) !== "none") {
         captureIntegrationBranch(base, s.currentMilestoneId);
       }
       setActiveMilestoneId(base, s.currentMilestoneId);
@@ -681,7 +662,7 @@ export async function bootstrapAutoSession(
     // Guard against stale milestone branch when isolation:none (#3613).
     // A prior session with isolation:branch/worktree may have left HEAD on
     // milestone/<MID>. Auto-checkout back to the integration branch.
-    if (getIsolationMode() === "none" && nativeIsRepo(base)) {
+    if (getIsolationMode(base) === "none" && nativeIsRepo(base)) {
       try {
         const currentBranch = nativeGetCurrentBranch(base);
         if (currentBranch.startsWith("milestone/")) {
@@ -712,7 +693,7 @@ export async function bootstrapAutoSession(
 
     if (
       s.currentMilestoneId &&
-      getIsolationMode() !== "none" &&
+      getIsolationMode(base) !== "none" &&
       !detectWorktreeName(base) &&
       !isUnderGsdWorktrees(base)
     ) {
@@ -758,9 +739,22 @@ export async function bootstrapAutoSession(
     // call returns "db_unavailable", triggering artifact-retry which
     // re-dispatches the same task — producing an infinite loop (#2419).
     if (existsSync(gsdDbPath) && !isDbAvailable()) {
+      const dbStatus = getDbStatus();
+      const phaseHint = dbStatus.lastPhase === "open"
+        ? "The database file could not be opened"
+        : dbStatus.lastPhase === "initSchema"
+          ? "The database schema could not be initialized"
+          : dbStatus.lastPhase === "vacuum-recovery"
+            ? "Corruption recovery (VACUUM) failed"
+            : dbStatus.attempted
+              ? "The database could not be opened (phase unknown)"
+              : "The database provider could not be loaded";
+      const errorDetail = dbStatus.lastError ? ` (${dbStatus.lastError.message})` : "";
+      const providerHint = dbStatus.provider
+        ? ` Provider: ${dbStatus.provider}.`
+        : " No SQLite provider available — check Node >= 22 or install better-sqlite3.";
       ctx.ui.notify(
-        "SQLite database exists but failed to open. Auto-mode cannot proceed without a working database provider. " +
-          "Check for corrupt gsd.db or missing native SQLite bindings.",
+        `SQLite database exists but failed to open: ${gsdDbPath}. ${phaseHint}${errorDetail}.${providerHint}`,
         "error",
       );
       return releaseLockAndReturn();
@@ -779,6 +773,7 @@ export async function bootstrapAutoSession(
         id: startModelSnapshot.id,
       };
     }
+    s.autoModeStartThinkingLevel = startThinkingSnapshot ?? null;
     s.manualSessionModelOverride = manualSessionOverride ?? null;
 
     // Apply worker model override from parallel orchestrator (#worker-model).
@@ -801,14 +796,11 @@ export async function bootstrapAutoSession(
     }
 
     // Snapshot installed skills
-    if (resolveSkillDiscoveryMode() !== "off") {
+    if (resolveSkillDiscoveryMode(base) !== "off") {
       snapshotSkills();
     }
 
     ctx.ui.setStatus("gsd-auto", s.stepMode ? "next" : "auto");
-    ctx.ui.setFooter(hideFooter);
-    // Hide gsd-health during AUTO — gsd-progress is the single source of truth
-    // for last-commit / cost / health signal while auto is running.
     ctx.ui.setWidget("gsd-health", undefined);
     const modeLabel = s.stepMode ? "Step-mode" : "Auto-mode";
     const pendingCount = (state.registry ?? []).filter(
@@ -835,7 +827,7 @@ export async function bootstrapAutoSession(
     // FlatRateContext used by selectAndApplyModel so user-declared
     // flat-rate providers and externalCli auto-detection are respected.
     const { isFlatRateProvider, buildFlatRateContext } = await import("./auto-model-selection.js");
-    const bannerPrefs = loadEffectiveGSDPreferences()?.preferences;
+    const bannerPrefs = loadEffectiveGSDPreferences(base)?.preferences;
     const effectiveProvider = s.autoModeStartModel?.provider ?? ctx.model?.provider;
     const effectivelyEnabled = routingConfig.enabled
       && (routingConfig.allow_flat_rate_providers