gsd-pi 2.58.0 → 2.59.0-dev.023bd39

package/src/resources/extensions/gsd/workflow-manifest.ts

@@ -42,6 +42,23 @@ function requireDb() {
   return db;
 }
 
+/**
+ * Coerce a raw DB value to a number, returning `fallback` for
+ * null/undefined/non-numeric strings (e.g. "-", "N/A", "").
+ * SQLite can store TEXT in INTEGER columns after migrations or manual inserts.
+ */
+export function toNumeric(value: unknown, fallback: number | null = null): number | null {
+  if (value === null || value === undefined) return fallback;
+  if (typeof value === "number") return Number.isFinite(value) ? value : fallback;
+  if (typeof value === "string") {
+    const trimmed = value.trim();
+    if (trimmed === "" || trimmed === "-" || trimmed === "N/A") return fallback;
+    const n = Number(trimmed);
+    return Number.isFinite(n) ? n : fallback;
+  }
+  return fallback;
+}
+
 // ─── snapshotState ───────────────────────────────────────────────────────
 
 /**
@@ -99,7 +116,7 @@ export function snapshotState(): StateManifest {
     proof_level: (r["proof_level"] as string) ?? "",
     integration_closure: (r["integration_closure"] as string) ?? "",
     observability_impact: (r["observability_impact"] as string) ?? "",
-    sequence: (r["sequence"] as number) ?? 0,
+    sequence: toNumeric(r["sequence"], 0) as number,
     replan_triggered_at: (r["replan_triggered_at"] as string) ?? null,
   }));
 
@@ -129,12 +146,12 @@ export function snapshotState(): StateManifest {
     expected_output: JSON.parse((r["expected_output"] as string) || "[]"),
     observability_impact: (r["observability_impact"] as string) ?? "",
     full_plan_md: (r["full_plan_md"] as string) ?? "",
-    sequence: (r["sequence"] as number) ?? 0,
+    sequence: toNumeric(r["sequence"], 0) as number,
   }));
 
   const rawDecisions = db.prepare("SELECT * FROM decisions ORDER BY seq").all() as Record<string, unknown>[];
   const decisions: Decision[] = rawDecisions.map((r) => ({
-    seq: r["seq"] as number,
+    seq: toNumeric(r["seq"], 0) as number,
     id: r["id"] as string,
     when_context: (r["when_context"] as string) ?? "",
     scope: (r["scope"] as string) ?? "",
@@ -153,9 +170,9 @@ export function snapshotState(): StateManifest {
     slice_id: r["slice_id"] as string,
     milestone_id: r["milestone_id"] as string,
     command: r["command"] as string,
-    exit_code: (r["exit_code"] as number) ?? null,
+    exit_code: toNumeric(r["exit_code"]),
     verdict: (r["verdict"] as string) ?? "",
-    duration_ms: (r["duration_ms"] as number) ?? null,
+    duration_ms: toNumeric(r["duration_ms"]),
     created_at: r["created_at"] as string,
   }));
 

package/src/resources/extensions/gsd/workflow-projections.ts

@@ -9,8 +9,9 @@ import {
   getMilestone,
   getMilestoneSlices,
   getSliceTasks,
+  getVerificationEvidence,
 } from "./gsd-db.js";
-import type { MilestoneRow, SliceRow, TaskRow } from "./gsd-db.js";
+import type { MilestoneRow, SliceRow, TaskRow, VerificationEvidenceRow } from "./gsd-db.js";
 import { atomicWriteSync } from "./atomic-write.js";
 import { join } from "node:path";
 import { mkdirSync, existsSync } from "node:fs";
@@ -29,8 +30,10 @@ export function renderPlanContent(sliceRow: SliceRow, taskRows: TaskRow[]): stri
 
   lines.push(`# ${sliceRow.id}: ${sliceRow.title}`);
   lines.push("");
-  lines.push(`**Goal:** ${sliceRow.goal || sliceRow.full_summary_md || "TBD"}`);
-  lines.push(`**Demo:** After this: ${sliceRow.demo || sliceRow.full_uat_md || "TBD"}`);
+  // #2945: never use full_summary_md/full_uat_md as display fallbacks —
+  // they contain multi-line rendered markdown that corrupts single-line fields.
+  lines.push(`**Goal:** ${sliceRow.goal || "TBD"}`);
+  lines.push(`**Demo:** After this: ${sliceRow.demo || "TBD"}`);
   lines.push("");
   lines.push("## Tasks");
 
@@ -113,7 +116,10 @@ export function renderRoadmapContent(milestoneRow: MilestoneRow, sliceRows: Slic
     }
 
     const risk = (slice.risk || "low").toLowerCase();
-    const demo = slice.demo || slice.full_uat_md || "TBD";
+    // #2945 Bug 1: never use full_uat_md as a table cell fallback — it contains
+    // multi-line UAT content (preconditions, steps, expected results) that
+    // corrupts the markdown table and makes subsequent slices invisible.
+    const demo = slice.demo || "TBD";
 
     lines.push(`| ${slice.id} | ${slice.title} | ${risk} | ${depends} | ${done} | ${demo} |`);
   }
@@ -142,71 +148,93 @@ export function renderRoadmapProjection(basePath: string, milestoneId: string):
 
 /**
  * Render SUMMARY.md content from a task row.
- * Pure function — no side effects.
+ * Single source of truth for summary rendering — used both at completion
+ * time and at projection regeneration time (#2720).
+ *
+ * @param evidence - Optional verification evidence rows. When called from
+ *   complete-task, these are passed directly. When called from projection
+ *   regeneration, they are queried from the DB by renderSummaryProjection.
  */
-export function renderSummaryContent(taskRow: TaskRow, sliceId: string, milestoneId: string): string {
-  const lines: string[] = [];
-
-  // Frontmatter
-  lines.push("---");
-  lines.push(`id: ${taskRow.id}`);
-  lines.push(`parent: ${sliceId}`);
-  lines.push(`milestone: ${milestoneId}`);
-  lines.push("provides: []");
-  lines.push("requires: []");
-  lines.push("affects: []");
-
-  // key_files is already parsed to string[]
-  if (taskRow.key_files && taskRow.key_files.length > 0) {
-    lines.push(`key_files: [${taskRow.key_files.map(f => `"${f}"`).join(", ")}]`);
+export function renderSummaryContent(
+  taskRow: TaskRow,
+  sliceId: string,
+  milestoneId: string,
+  evidence?: Array<{ command: string; exitCode?: number; exit_code?: number; verdict: string; durationMs?: number; duration_ms?: number }>,
+): string {
+  // ── Frontmatter (YAML list format, matches parseSummary() expectations) ──
+  const keyFilesYaml = taskRow.key_files && taskRow.key_files.length > 0
+    ? taskRow.key_files.map(f => `  - ${f}`).join("\n")
+    : "  - (none)";
+  const keyDecisionsYaml = taskRow.key_decisions && taskRow.key_decisions.length > 0
+    ? taskRow.key_decisions.map(d => `  - ${d}`).join("\n")
+    : "  - (none)";
+
+  // Derive verification_result from evidence if available
+  const evidenceList = evidence ?? [];
+  const allPassed = evidenceList.length > 0 &&
+    evidenceList.every(e => {
+      const code = e.exitCode ?? e.exit_code ?? -1;
+      return code === 0 || e.verdict.includes("\u2705") || e.verdict.toLowerCase().includes("pass");
+    });
+  const verificationResult = taskRow.verification_result
+    ? (allPassed ? "passed" : (evidenceList.length === 0 ? "untested" : "mixed"))
+    : (allPassed ? "passed" : (evidenceList.length === 0 ? "untested" : "mixed"));
+
+  // Build verification evidence table
+  let evidenceTable = "| # | Command | Exit Code | Verdict | Duration |\n|---|---------|-----------|---------|----------|\n";
+  if (evidenceList.length > 0) {
+    evidenceList.forEach((e, i) => {
+      const code = e.exitCode ?? e.exit_code ?? 0;
+      const dur = e.durationMs ?? e.duration_ms ?? 0;
+      evidenceTable += `| ${i + 1} | \`${e.command}\` | ${code} | ${e.verdict} | ${dur}ms |\n`;
+    });
   } else {
-    lines.push("key_files: []");
+    evidenceTable += "| \u2014 | No verification commands discovered | \u2014 | \u2014 | \u2014 |\n";
   }
 
-  // key_decisions is already parsed to string[]
-  if (taskRow.key_decisions && taskRow.key_decisions.length > 0) {
-    lines.push(`key_decisions: [${taskRow.key_decisions.map(d => `"${d}"`).join(", ")}]`);
-  } else {
-    lines.push("key_decisions: []");
-  }
+  const title = taskRow.one_liner || taskRow.title || taskRow.id;
 
-  lines.push("patterns_established: []");
-  lines.push("drill_down_paths: []");
-  lines.push("observability_surfaces: []");
-  lines.push(`duration: "${taskRow.duration || ""}"`);
-  lines.push(`verification_result: "${taskRow.verification_result || ""}"`);
-  lines.push(`completed_at: ${taskRow.completed_at || ""}`);
-  lines.push(`blocker_discovered: ${taskRow.blocker_discovered ? "true" : "false"}`);
-  lines.push("---");
-  lines.push("");
-  lines.push(`# ${taskRow.id}: ${taskRow.title}`);
-  lines.push("");
+  return `---
+id: ${taskRow.id}
+parent: ${sliceId}
+milestone: ${milestoneId}
+key_files:
+${keyFilesYaml}
+key_decisions:
+${keyDecisionsYaml}
+duration: ${taskRow.duration || ""}
+verification_result: ${verificationResult}
+completed_at: ${taskRow.completed_at || ""}
+blocker_discovered: ${taskRow.blocker_discovered ? "true" : "false"}
+---
 
-  // One-liner (if present)
-  if (taskRow.one_liner) {
-    lines.push(`> ${taskRow.one_liner}`);
-    lines.push("");
-  }
+# ${taskRow.id}: ${title}
 
-  lines.push("## What Happened");
-  lines.push(taskRow.full_summary_md || taskRow.narrative || "No summary recorded.");
-  lines.push("");
+**${taskRow.one_liner || ""}**
 
-  // Deviations (if present)
-  if (taskRow.deviations) {
-    lines.push("## Deviations");
-    lines.push(taskRow.deviations);
-    lines.push("");
-  }
+## What Happened
 
-  // Known issues (if present)
-  if (taskRow.known_issues) {
-    lines.push("## Known Issues");
-    lines.push(taskRow.known_issues);
-    lines.push("");
-  }
+${taskRow.narrative || "No summary recorded."}
 
-  return lines.join("\n");
+## Verification
+
+${taskRow.verification_result || "No verification recorded."}
+
+## Verification Evidence
+
+${evidenceTable}
+## Deviations
+
+${taskRow.deviations || "None."}
+
+## Known Issues
+
+${taskRow.known_issues || "None."}
+
+## Files Created/Modified
+
+${taskRow.key_files && taskRow.key_files.length > 0 ? taskRow.key_files.map(f => `- \`${f}\``).join("\n") : "None."}
+`;
 }
 
 /**
@@ -218,7 +246,8 @@ export function renderSummaryProjection(basePath: string, milestoneId: string, s
   const taskRow = taskRows.find(t => t.id === taskId);
   if (!taskRow) return;
 
-  const content = renderSummaryContent(taskRow, sliceId, milestoneId);
+  const evidenceRows = getVerificationEvidence(milestoneId, sliceId, taskId);
+  const content = renderSummaryContent(taskRow, sliceId, milestoneId, evidenceRows);
   const dir = join(basePath, ".gsd", "milestones", milestoneId, "slices", sliceId, "tasks");
   mkdirSync(dir, { recursive: true });
   atomicWriteSync(join(dir, `${taskId}-SUMMARY.md`), content);
@@ -235,14 +264,18 @@ export function renderStateContent(state: GSDState): string {
   const lines: string[] = [];
   lines.push("# GSD State", "");
 
-  const activeMilestone = state.activeMilestone
-    ? `${state.activeMilestone.id}: ${state.activeMilestone.title}`
-    : "None";
   const activeSlice = state.activeSlice
     ? `${state.activeSlice.id}: ${state.activeSlice.title}`
     : "None";
 
-  lines.push(`**Active Milestone:** ${activeMilestone}`);
+  if (state.phase === 'complete' && state.lastCompletedMilestone) {
+    lines.push(`**Last Completed Milestone:** ${state.lastCompletedMilestone.id}: ${state.lastCompletedMilestone.title}`);
+  } else {
+    const activeMilestone = state.activeMilestone
+      ? `${state.activeMilestone.id}: ${state.activeMilestone.title}`
+      : "None";
+    lines.push(`**Active Milestone:** ${activeMilestone}`);
+  }
   lines.push(`**Active Slice:** ${activeSlice}`);
   lines.push(`**Phase:** ${state.phase}`);
   if (state.requirements) {

package/src/resources/extensions/gsd/workflow-reconcile.ts

@@ -1,19 +1,53 @@
 import { join } from "node:path";
 import { mkdirSync, existsSync, readFileSync, unlinkSync } from "node:fs";
+import { logWarning, logError } from "./workflow-logger.js";
 import { readEvents, findForkPoint, appendEvent, getSessionId } from "./workflow-events.js";
 import type { WorkflowEvent } from "./workflow-events.js";
 import {
   transaction,
   updateTaskStatus,
   updateSliceStatus,
+  getSliceTasks,
   insertVerificationEvidence,
   upsertDecision,
   openDatabase,
 } from "./gsd-db.js";
+import { isClosedStatus } from "./status-guards.js";
 import { writeManifest } from "./workflow-manifest.js";
 import { atomicWriteSync } from "./atomic-write.js";
 import { acquireSyncLock, releaseSyncLock } from "./sync-lock.js";
 
+// ─── Replay Helpers ──────────────────────────────────────────────────────────
+
+/**
+ * Replay a complete_slice event with task validation.
+ *
+ * #2945 Bug 2: The original replay blindly called updateSliceStatus("done")
+ * without checking whether all tasks in the slice are actually complete.
+ * During API overload or partial execution, a complete_slice event could
+ * be logged even when tasks were skipped, causing the milestone completion
+ * guard to see the slice as "done" and allow premature milestone completion.
+ *
+ * This function validates that every task in the slice has a closed status
+ * before marking the slice as done. If any task is still pending, the slice
+ * status is left unchanged.
+ */
+export function replaySliceComplete(milestoneId: string, sliceId: string, ts: string): void {
+  const tasks = getSliceTasks(milestoneId, sliceId);
+  // If there are tasks and any are not closed, skip the status update
+  if (tasks.length > 0) {
+    const incompleteTasks = tasks.filter(t => !isClosedStatus(t.status));
+    if (incompleteTasks.length > 0) {
+      process.stderr.write(
+        `[gsd] reconcile: skipping complete_slice replay for ${sliceId} — ` +
+        `${incompleteTasks.length} task(s) still pending\n`,
+      );
+      return;
+    }
+  }
+  updateSliceStatus(milestoneId, sliceId, "done", ts);
+}
+
 // ─── Public Types ─────────────────────────────────────────────────────────────
 
 export interface ConflictEntry {
@@ -82,7 +116,8 @@ function replayEvents(events: WorkflowEvent[]): void {
       case "complete_slice": {
         const milestoneId = p["milestoneId"] as string;
         const sliceId = p["sliceId"] as string;
-        updateSliceStatus(milestoneId, sliceId, "done", event.ts);
+        // #2945 Bug 2: validate tasks before marking slice done
+        replaySliceComplete(milestoneId, sliceId, event.ts);
         break;
       }
       case "plan_slice": {
@@ -274,9 +309,7 @@ export function reconcileWorktreeLogs(
   // Acquire advisory lock to prevent concurrent reconcile + append races
   const lock = acquireSyncLock(mainBasePath);
   if (!lock.acquired) {
-    process.stderr.write(
-      `[gsd] reconcile: could not acquire sync lock — another reconciliation may be in progress\n`,
-    );
+    logWarning("reconcile", "could not acquire sync lock — another reconciliation may be in progress");
     return { autoMerged: 0, conflicts: [] };
   }
 
@@ -315,9 +348,7 @@ function _reconcileWorktreeLogsInner(
   if (conflicts.length > 0) {
     // D-04: atomic all-or-nothing — block entire merge
     writeConflictsFile(mainBasePath, conflicts, worktreeBasePath);
-    process.stderr.write(
-      `[gsd] reconcile: ${conflicts.length} conflict(s) detected — see ${join(mainBasePath, ".gsd", "CONFLICTS.md")}\n`,
-    );
+    logError("reconcile", `${conflicts.length} conflict(s) detected`, { count: String(conflicts.length), path: join(mainBasePath, ".gsd", "CONFLICTS.md") });
     return { autoMerged: 0, conflicts };
   }
 
@@ -341,9 +372,7 @@ function _reconcileWorktreeLogsInner(
   try {
     writeManifest(mainBasePath);
   } catch (err) {
-    process.stderr.write(
-      `[gsd] reconcile: manifest write failed (non-fatal): ${(err as Error).message}\n`,
-    );
+    logWarning("reconcile", "manifest write failed (non-fatal)", { error: (err as Error).message });
   }
 
   return { autoMerged: merged.length, conflicts: [] };

package/src/resources/extensions/gsd/workspace-index.ts

@@ -11,6 +11,7 @@ import {
   resolveTasksDir,
 } from "./paths.js";
 import { deriveState } from "./state.js";
+import { extractVerdict } from "./verdict-parser.js";
 import { milestoneIdSort, findMilestoneIds } from "./guided-flow.js";
 import type { RiskLevel } from "./types.js";
 import { getSliceBranchName, detectWorktreeName } from "./worktree.js";
@@ -42,6 +43,10 @@ export interface WorkspaceMilestoneTarget {
   id: string;
   title: string;
   roadmapPath?: string;
+  /** Authoritative milestone lifecycle status from the GSD state registry. */
+  status?: "complete" | "active" | "pending" | "parked";
+  /** Milestone validation verdict, when validation has been performed. */
+  validationVerdict?: "pass" | "needs-attention" | "needs-remediation";
   slices: WorkspaceSliceTarget[];
 }
 
@@ -192,6 +197,31 @@ export async function indexWorkspace(basePath: string, opts: IndexWorkspaceOptio
     phase: state.phase,
   };
 
+  // Enrich milestones with authoritative status from state registry (#2807)
+  if (state.registry) {
+    const registryMap = new Map(state.registry.map(e => [e.id, e]));
+    for (const milestone of milestones) {
+      const entry = registryMap.get(milestone.id);
+      if (entry) {
+        milestone.status = entry.status;
+      }
+    }
+  }
+
+  // Populate validationVerdict from VALIDATION files (#2807)
+  for (const milestone of milestones) {
+    const validationPath = resolveMilestoneFile(basePath, milestone.id, "VALIDATION");
+    if (validationPath) {
+      const validationContent = await loadFile(validationPath);
+      if (validationContent) {
+        const verdict = extractVerdict(validationContent);
+        if (verdict === "pass" || verdict === "needs-attention" || verdict === "needs-remediation") {
+          milestone.validationVerdict = verdict;
+        }
+      }
+    }
+  }
+
   const scopes: WorkspaceScopeTarget[] = [{ scope: "project", label: "project", kind: "project" }];
   for (const milestone of milestones) {
     scopes.push({ scope: milestone.id, label: `${milestone.id}: ${milestone.title}`, kind: "milestone" });

package/src/resources/extensions/gsd/worktree-manager.ts

@@ -15,7 +15,7 @@
  *   4. remove()  — git worktree remove + branch cleanup
  */
 
-import { existsSync, mkdirSync, readFileSync, realpathSync, rmSync } from "node:fs";
+import { existsSync, lstatSync, mkdirSync, readdirSync, readFileSync, realpathSync, rmSync } from "node:fs";
 import { execFileSync } from "node:child_process";
 import { join, resolve, sep } from "node:path";
 import { GSDError, GSD_PARSE_ERROR, GSD_STALE_STATE, GSD_LOCK_HELD, GSD_GIT_ERROR, GSD_MERGE_CONFLICT } from "./errors.js";
@@ -277,6 +277,78 @@ export function listWorktrees(basePath: string): WorktreeInfo[] {
   return worktrees;
 }
 
+// ─── Nested .git Detection (#2616) ──────────────────────────────────────
+//
+// Scaffolding tools (create-next-app, cargo init, etc.) create nested .git
+// directories inside worktrees. Git records these as gitlinks (mode 160000)
+// without a .gitmodules entry — so worktree cleanup destroys the only copy
+// of their object database, causing permanent silent data loss.
+
+/** Directories to skip when scanning for nested .git dirs. */
+const NESTED_GIT_SKIP_DIRS = new Set([
+  ".git", ".gsd", "node_modules", ".next", ".nuxt", "dist", "build",
+  "__pycache__", ".tox", ".venv", "venv", "target", "vendor",
+]);
+
+/**
+ * Recursively find nested .git directories inside a worktree root.
+ * Returns paths to directories that contain their own .git (directory, not file).
+ * Skips node_modules, .gsd, and other non-project directories for performance.
+ *
+ * A nested .git *directory* (not a .git file — which is a legitimate worktree
+ * pointer) indicates a scaffolded repo that will become an orphaned gitlink.
+ */
+export function findNestedGitDirs(rootPath: string): string[] {
+  const results: string[] = [];
+
+  function walk(dir: string, depth: number): void {
+    // Cap recursion depth to avoid runaway scanning
+    if (depth > 10) return;
+
+    let entries: string[];
+    try {
+      entries = readdirSync(dir);
+    } catch {
+      return; // Permission denied, broken symlink, etc.
+    }
+
+    for (const entry of entries) {
+      if (NESTED_GIT_SKIP_DIRS.has(entry)) continue;
+
+      const fullPath = join(dir, entry);
+
+      // Only follow real directories, not symlinks
+      let stat;
+      try {
+        stat = lstatSync(fullPath);
+      } catch {
+        continue;
+      }
+      if (!stat.isDirectory()) continue;
+
+      // Check if this directory contains a .git *directory* (not a .git file).
+      // A .git file is a worktree pointer and is legitimate.
+      // A .git directory is a standalone repo created by scaffolding.
+      const innerGit = join(fullPath, ".git");
+      try {
+        const innerStat = lstatSync(innerGit);
+        if (innerStat.isDirectory()) {
+          results.push(fullPath);
+          // Don't recurse into the nested repo — we found what we need
+          continue;
+        }
+      } catch {
+        // No .git here — continue scanning
+      }
+
+      walk(fullPath, depth + 1);
+    }
+  }
+
+  walk(rootPath, 0);
+  return results;
+}
+
 /**
  * Remove a worktree and optionally delete its branch.
  * If the process is currently inside the worktree, chdir out first.
@@ -355,6 +427,30 @@ export function removeWorktree(
     }
   }
 
+  // Nested .git safety (#2616): detect nested .git directories created by
+  // scaffolding tools (create-next-app, cargo init, etc.). These produce
+  // gitlink entries (mode 160000) without .gitmodules — cleanup would destroy
+  // the only copy of the nested object database, causing permanent data loss.
+  // Fix: remove the nested .git dirs so git tracks the files as regular content.
+  const nestedGitDirs = findNestedGitDirs(resolvedWtPath);
+  if (nestedGitDirs.length > 0) {
+    for (const nestedDir of nestedGitDirs) {
+      const nestedGitPath = join(nestedDir, ".git");
+      try {
+        rmSync(nestedGitPath, { recursive: true, force: true });
+        logWarning("reconcile",
+          `Removed nested .git directory from scaffolded project to prevent data loss (#2616)`,
+          { worktree: name, nestedRepo: nestedDir },
+        );
+      } catch {
+        logWarning("reconcile",
+          `Failed to remove nested .git directory — files may be lost as orphaned gitlink`,
+          { worktree: name, nestedRepo: nestedDir },
+        );
+      }
+    }
+  }
+
   // Remove worktree: try non-force first when submodules have changes,
   // falling back to force only after submodule state has been preserved.
   const useForce = hasSubmoduleChanges ? false : force;
@@ -365,6 +461,29 @@ export function removeWorktree(
     try { nativeWorktreeRemove(basePath, resolvedWtPath, true); } catch { /* may fail */ }
   }
 
+  // (#2821) If the worktree directory STILL exists after both native removal
+  // attempts (e.g. untracked files like ASSESSMENT/UAT-RESULT prevent git
+  // worktree remove), force-remove the git internal worktree metadata first,
+  // then remove the filesystem directory. Without this, the .git/worktrees/<name>
+  // lock prevents rmSync from cleaning up, and the orphaned worktree directory
+  // causes every subsequent `/gsd auto` to re-enter the stale worktree.
+  if (existsSync(resolvedWtPath)) {
+    try {
+      const wtInternalDir = join(basePath, ".git", "worktrees", name);
+      if (existsSync(wtInternalDir)) {
+        rmSync(wtInternalDir, { recursive: true, force: true });
+      }
+      rmSync(resolvedWtPath, { recursive: true, force: true });
+    } catch {
+      logWarning(
+        "reconcile",
+        `Worktree directory could not be removed after git internal cleanup: ${resolvedWtPath}. ` +
+          `Manual cleanup: rm -rf "${resolvedWtPath.replaceAll("\\", "/")}"`,
+        { worktree: name },
+      );
+    }
+  }
+
   // Prune stale entries so git knows the worktree is gone
   nativeWorktreePrune(basePath);
 

package/src/resources/extensions/gsd/worktree-resolver.ts

@@ -350,7 +350,13 @@ export class WorktreeResolver {
       data: { milestoneId, mode },
     });
 
-    if (mode === "none") {
+    // #2625: If we are physically inside an auto-worktree, we MUST merge
+    // regardless of the current isolation config. This prevents data loss when
+    // the default isolation mode changes between versions (e.g., "worktree" ->
+    // "none"): the worktree branch still holds real commits that need merging.
+    const inWorktree = this.deps.isInAutoWorktree(this.s.basePath) && this.s.originalBasePath;
+
+    if (mode === "none" && !inWorktree) {
       debugLog("WorktreeResolver", {
         action: "mergeAndExit",
         milestoneId,
@@ -361,8 +367,7 @@ export class WorktreeResolver {
     }
 
     if (
-      mode === "worktree" ||
-      (this.deps.isInAutoWorktree(this.s.basePath) && this.s.originalBasePath)
+      mode === "worktree" || inWorktree
     ) {
       this._mergeWorktreeMode(milestoneId, ctx);
     } else if (mode === "branch") {
@@ -432,6 +437,20 @@ export class WorktreeResolver {
           milestoneId,
           roadmapContent,
         );
+
+        // #2945 Bug 3: mergeMilestoneToMain performs best-effort worktree
+        // cleanup internally (step 12), but it can silently fail on Windows
+        // or when the worktree directory is locked. Perform a secondary
+        // teardown here to ensure the worktree is properly cleaned up.
+        // This is idempotent — if the worktree was already removed,
+        // teardownAutoWorktree handles the no-op case gracefully.
+        try {
+          this.deps.teardownAutoWorktree(originalBase, milestoneId);
+        } catch {
+          // Best-effort — the primary cleanup in mergeMilestoneToMain may
+          // have already removed the worktree.
+        }
+
         if (mergeResult.codeFilesChanged) {
           ctx.notify(
             `Milestone ${milestoneId} merged to main.${mergeResult.pushed ? " Pushed to remote." : ""}`,

package/src/resources/extensions/mcp-client/index.ts

@@ -111,7 +111,11 @@ function readConfigs(): McpServerConfig[] {
 }
 
 function getServerConfig(name: string): McpServerConfig | undefined {
-	return readConfigs().find((s) => s.name === name);
+	const trimmed = name.trim();
+	return readConfigs().find((s) =>
+		s.name === trimmed ||
+		s.name.toLowerCase() === trimmed.toLowerCase(),
+	);
 }
 
 /** Resolve ${VAR} references in env values against process.env. */
@@ -131,12 +135,14 @@ function resolveEnv(env: Record<string, string>): Record<string, string> {
 }
 
 async function getOrConnect(name: string, signal?: AbortSignal): Promise<Client> {
-	const existing = connections.get(name);
-	if (existing) return existing.client;
-
 	const config = getServerConfig(name);
 	if (!config) throw new Error(`Unknown MCP server: "${name}". Use mcp_servers to list available servers.`);
 
+	// Always use config.name as the canonical cache key so that variant
+	// casing / whitespace still hits the same connection.
+	const existing = connections.get(config.name);
+	if (existing) return existing.client;
+
 	const client = new Client({ name: "gsd", version: "1.0.0" });
 	let transport: StdioClientTransport | StreamableHTTPClientTransport;
 
@@ -151,15 +157,15 @@ async function getOrConnect(name: string, signal?: AbortSignal): Promise<Client>
 	} else if (config.transport === "http" && config.url) {
 		const resolvedUrl = config.url.replace(
 			/\$\{([^}]+)\}/g,
-			(_, name) => process.env[name] ?? "",
+			(_, varName) => process.env[varName] ?? "",
 		);
 		transport = new StreamableHTTPClientTransport(new URL(resolvedUrl));
 	} else {
-		throw new Error(`Server "${name}" has unsupported transport: ${config.transport}`);
+		throw new Error(`Server "${config.name}" has unsupported transport: ${config.transport}`);
 	}
 
 	await client.connect(transport, { signal, timeout: 30000 });
-	connections.set(name, { client, transport });
+	connections.set(config.name, { client, transport });
 	return client;
 }