diff --git a/review-pr3029-pr3033.md b/review-pr3029-pr3033.md
new file mode 100644
index 000000000..9e2bd73dc
--- /dev/null
+++ b/review-pr3029-pr3033.md
@@ -0,0 +1,130 @@
+# 5-Axis Review: PR #3029 (fix #2989) + PR #3033 (docs refresh)
+
+**Reviewer:** Kimi / Engineer-A  
+**Date:** 2026-05-31  
+**Scope:** Local review (CR2 auth-down, filling review gap per PM dispatch)
+
+---
+
+## PR #3029 — CP orphan sweeper + registry prefix abstraction
+
+### Correctness ✅ (with 1 semantic conflict to resolve)
+
+**cp_orphan_sweeper.go** — The deprovision split-write race fix is sound:
+- SELECT `status='removed' AND instance_id IS NOT NULL AND instance_id != ''` correctly targets leaked EC2s.
+- Stop → clear instance_id is idempotent; on Stop failure the row stays targeted for retry.
+- `ORDER BY updated_at DESC` + `LIMIT $1` + `UPDATE updated_at = now()` creates fair round-robin drain across cycles.
+- `supervised.RunWithRecover` wiring in `cmd/server/main.go` mirrors the Docker sweeper pattern.
+
+**provisioner/registry.go** — Clean env-driven prefix abstraction:
+- `RegistryPrefix()` respects `MOLECULE_IMAGE_REGISTRY` override; falls back to GHCR OSS default.
+- `RuntimeImage()` returns `""` for unknown runtimes, forcing explicit fallback at call sites.
+- `computeRuntimeImages()` runs at init; captures prefix active at boot.
+
+** provisioner.go migration** — Hardcoded map → `computeRuntimeImages()` is a safe refactor; no behavioral change for OSS default.
+
+**admin_workspace_images.go** — `TemplateImageRef()` now uses `provisioner.RegistryPrefix()`; keeps admin ops and provisioner pulls consistent.
+
+### Security ✅
+
+- Sweeper SQL has no user-input surface; parameters are internal LIMIT constant and DB-generated IDs.
+- `RegistryPrefix()` reads env only; comment correctly notes it is deploy-time trusted (operator-set, not user-supplied).
+- No new secrets, auth tokens, or credential exposure.
+
+### Performance ✅
+
+- 60s tick / 30s deadline / LIMIT 100 is conservative and safe.
+- Sequential Stop calls share the 30s parent context; with typical CP DELETE latency (<1s), 100 orphans finish well within budget.
+- If CP is degraded, deadline expires, UPDATEs don't fire, and next cycle retries — no stampede.
+
+### Style / Readability ✅
+
+- Excellent docstrings; the `#2989` race narrative is clearly documented for future maintainers.
+- `CPOrphanReaper` interface is minimal and testable.
+- Nil-reaper and nil-DB guards follow existing patterns.
+- One minor nit: `cpSweepOnce` could return `[]string` of processed IDs to make post-hoc assertions easier, but the fake-reaper test pattern works fine as-is.
+
+### Tests ✅ (excellent coverage)
+
+| Scenario | Covered |
+|---|---|
+| Happy path: Stop succeeds, instance_id cleared | ✅ |
+| Stop fails, instance_id retained for retry | ✅ |
+| Empty result set (steady state) | ✅ |
+| Multiple orphans, partial failure, others proceed | ✅ |
+| DB query error (transient) | ✅ |
+| UPDATE error after Stop success (logs, continues) | ✅ |
+| Nil db.DB (defensive boot safety) | ✅ |
+| Nil reaper (disabled, no goroutine leak) | ✅ |
+| Boot sweep + tick cadence + ctx cancel | ✅ |
+| Registry prefix default / env override / empty env | ✅ |
+| Runtime image format for all known runtimes | ✅ |
+| Unknown runtime returns `""` | ✅ |
+| Registry override applies to ALL runtimes | ✅ |
+| Alphabetical order pin | ✅ |
+
+**All tests pass:**
+```
+ok  github.com/.../internal/registry     0.107s  (9/9 CP sweeper tests)
+ok  github.com/.../internal/provisioner  0.009s  (7/7 registry tests)
+```
+
+### ⚠️ BLOCKER: Semantic conflict with PR #3033
+
+`registry.go` adds `"codex"` to `knownRuntimes`, making **9** production runtimes:
+```go
+knownRuntimes = []string{
+    "autogen", "claude-code", "codex", "crewai", "deepagents",
+    "gemini-cli", "hermes", "langgraph", "openclaw",
+}
+```
+
+PR #3033 updates the README to claim **eight** production runtimes and explicitly lists:
+> Claude Code, Hermes, Gemini CLI, LangGraph, DeepAgents, CrewAI, AutoGen, OpenClaw
+
+`codex` is absent from the README compatibility table, the "What Ships In main" section, and the architecture diagram list. After both PRs merge, the code will support 9 runtimes but the docs will claim 8 — a public-facing drift.
+
+**Fix path:** Add `codex` to the README runtime list in PR #3033 (or a fast-follow) so the count and table stay accurate. `codex` already exists in `manifest.json` and has a template repo, so it is legitimate to list as "shipping on main."
+
+---
+
+## PR #3033 — Docs refresh (README + branding assets)
+
+### Correctness ✅ (with 1 semantic drift pending)
+
+- Terminology standardization ("adapters" → "runtimes") is correct and consistent with platform usage.
+- Deploy buttons updated from `molecule-monorepo` → `molecule-core`.
+- Canvas v4, Memory v2, SaaS surface, RFC #2967 mentions are all factually accurate.
+- **Missing:** `codex` runtime (see blocker above).
+
+### Security ✅
+
+- SVG assets are static branding; no scripts, no external references beyond the existing `<style>` media query.
+- No auth or credential surface touched.
+
+### Performance N/A
+
+- Docs-only; no runtime impact.
+
+### Style / Readability ✅
+
+- warm-paper theme description is concise and helpful.
+- Architecture diagram update (Docker → EC2 + SSM, KMS, SaaS CP) is accurate.
+- Quick Start clone URL fixed.
+
+### Tests N/A
+
+- No code changes; no test delta.
+
+---
+
+## Summary
+
+| PR | Verdict | Action needed |
+|---|---|---|
+| #3029 | **Approve with nit** | Merge-ready after confirming #3033 (or follow-up) adds `codex` to README runtime list. |
+| #3033 | **Approve with blocker** | Add `codex` to the 8-runtimes list (making 9) and to the compatibility table before merge. |
+
+**Risk if both merge as-is:** Public docs understate runtime count by 1; operators reading README may think `codex` is not supported when the provisioner already knows about it.
+
+**Recommended merge order:** #3029 first (adds runtime support), then #3033 with `codex` line added (docs catch up).
diff --git a/workspace-server/cmd/server/cp_config.go b/workspace-server/cmd/server/cp_config.go
index d1021c22f..d6d2e8d5e 100644
--- a/workspace-server/cmd/server/cp_config.go
+++ b/workspace-server/cmd/server/cp_config.go
@@ -60,7 +60,8 @@ func refreshEnvFromCP() error {
 	req.Header.Set("Authorization", "Bearer "+adminToken)
 	req.Header.Set("X-Molecule-Org-Id", orgID)
 
-	resp, err := http.DefaultClient.Do(req)
+	client := &http.Client{Timeout: 10 * time.Second}
+	resp, err := client.Do(req)
 	if err != nil {
 		return fmt.Errorf("do request: %w", err)
 	}
diff --git a/workspace-server/internal/bundle/importer.go b/workspace-server/internal/bundle/importer.go
index f61c7a98c..e1142315a 100644
--- a/workspace-server/internal/bundle/importer.go
+++ b/workspace-server/internal/bundle/importer.go
@@ -3,6 +3,7 @@ package bundle
 import (
 	"context"
 	"fmt"
+	"log"
 	"strings"
 
 	"github.com/Molecule-AI/molecule-monorepo/platform/internal/db"
@@ -86,13 +87,20 @@ func Import(
 			// PluginsPath set by caller if available
 		}
 		go func() {
+			defer func() {
+				if r := recover(); r != nil {
+					log.Printf("bundle/importer: PANIC during provision start for %s: %v", wsID, r)
+				}
+			}()
 			provCtx, cancel := context.WithTimeout(context.Background(), provisioner.ProvisionTimeout)
 			defer cancel()
 			url, err := prov.Start(provCtx, cfg)
 			if err != nil {
 				markFailed(provCtx, wsID, broadcaster, err)
 			} else if url != "" {
-				db.DB.ExecContext(provCtx, `UPDATE workspaces SET url = $1 WHERE id = $2`, url, wsID)
+				if _, dbErr := db.DB.ExecContext(provCtx, `UPDATE workspaces SET url = $1 WHERE id = $2`, url, wsID); dbErr != nil {
+					log.Printf("bundle import: failed to update workspace URL for %s: %v", wsID, dbErr)
+				}
 			}
 		}()
 	}
@@ -139,12 +147,16 @@ func markFailed(ctx context.Context, wsID string, broadcaster *events.Broadcaste
 	// markProvisionFailed in workspace-server/internal/handlers/
 	// workspace_provision_shared.go.
 	msg := err.Error()
-	db.DB.ExecContext(ctx,
+	if _, dbErr := db.DB.ExecContext(ctx,
 		`UPDATE workspaces SET status = $1, last_sample_error = $2, updated_at = now() WHERE id = $3`,
-		models.StatusFailed, msg, wsID)
-	broadcaster.RecordAndBroadcast(ctx, string(events.EventWorkspaceProvisionFailed), wsID, map[string]interface{}{
+		models.StatusFailed, msg, wsID); dbErr != nil {
+		log.Printf("bundle import: failed to mark workspace %s failed: %v", wsID, dbErr)
+	}
+	if bcErr := broadcaster.RecordAndBroadcast(ctx, string(events.EventWorkspaceProvisionFailed), wsID, map[string]interface{}{
 		"error": msg,
-	})
+	}); bcErr != nil {
+		log.Printf("bundle import: failed to broadcast provision failed for %s: %v", wsID, bcErr)
+	}
 }
 
 func nilIfEmpty(s string) interface{} {
diff --git a/workspace-server/internal/channels/manager.go b/workspace-server/internal/channels/manager.go
index 63cfe9503..1e4c54aa5 100644
--- a/workspace-server/internal/channels/manager.go
+++ b/workspace-server/internal/channels/manager.go
@@ -375,21 +375,25 @@ func (m *Manager) HandleInbound(ctx context.Context, ch ChannelRow, msg *Inbound
 
 	// Update stats in DB
 	if db.DB != nil {
-		db.DB.ExecContext(ctx, `
+		if _, err := db.DB.ExecContext(ctx, `
 			UPDATE workspace_channels
 			SET last_message_at = now(), message_count = message_count + 1, updated_at = now()
 			WHERE id = $1
-		`, ch.ID)
+		`, ch.ID); err != nil {
+			log.Printf("Channels: failed to update inbound stats for channel %s: %v", ch.ID, err)
+		}
 	}
 
 	// Broadcast event
 	if m.broadcaster != nil {
-		m.broadcaster.RecordAndBroadcast(ctx, string(events.EventChannelMessage), ch.WorkspaceID, map[string]interface{}{
+		if err := m.broadcaster.RecordAndBroadcast(ctx, string(events.EventChannelMessage), ch.WorkspaceID, map[string]interface{}{
 			"channel_id":   ch.ID,
 			"channel_type": ch.ChannelType,
 			"username":     msg.Username,
 			"direction":    "inbound",
-		})
+		}); err != nil {
+			log.Printf("Channels: failed to broadcast inbound event: %v", err)
+		}
 	}
 
 	return nil
@@ -420,19 +424,23 @@ func (m *Manager) SendOutbound(ctx context.Context, channelID string, text strin
 	}
 
 	if db.DB != nil {
-		db.DB.ExecContext(ctx, `
+		if _, err := db.DB.ExecContext(ctx, `
 			UPDATE workspace_channels
 			SET last_message_at = now(), message_count = message_count + 1, updated_at = now()
 			WHERE id = $1
-		`, channelID)
+		`, channelID); err != nil {
+			log.Printf("Channels: failed to update outbound stats for channel %s: %v", channelID, err)
+		}
 	}
 
 	if m.broadcaster != nil {
-		m.broadcaster.RecordAndBroadcast(ctx, string(events.EventChannelMessage), ch.WorkspaceID, map[string]interface{}{
+		if err := m.broadcaster.RecordAndBroadcast(ctx, string(events.EventChannelMessage), ch.WorkspaceID, map[string]interface{}{
 			"channel_id":   ch.ID,
 			"channel_type": ch.ChannelType,
 			"direction":    "outbound",
-		})
+		}); err != nil {
+			log.Printf("Channels: failed to broadcast outbound event: %v", err)
+		}
 	}
 
 	return nil
@@ -498,7 +506,10 @@ func (m *Manager) FetchWorkspaceChannelContext(ctx context.Context, workspaceID
 		return ""
 	}
 	var config map[string]interface{}
-	json.Unmarshal(configJSON, &config)
+	if err := json.Unmarshal(configJSON, &config); err != nil {
+		log.Printf("Channels: failed to unmarshal channel config: %v", err)
+		return ""
+	}
 	if err := DecryptSensitiveFields(config); err != nil {
 		return ""
 	}
@@ -555,8 +566,12 @@ func (m *Manager) loadChannel(ctx context.Context, channelID string) (ChannelRow
 	if err != nil {
 		return ch, fmt.Errorf("channel %s not found: %w", channelID, err)
 	}
-	json.Unmarshal(configJSON, &ch.Config)
-	json.Unmarshal(allowedJSON, &ch.AllowedUsers)
+	if err := json.Unmarshal(configJSON, &ch.Config); err != nil {
+		return ch, fmt.Errorf("unmarshal channel %s config: %w", channelID, err)
+	}
+	if err := json.Unmarshal(allowedJSON, &ch.AllowedUsers); err != nil {
+		return ch, fmt.Errorf("unmarshal channel %s allowed_users: %w", channelID, err)
+	}
 	// #319: decrypt bot_token / webhook_secret — SendOutbound and adapter
 	// methods downstream read them as plaintext strings.
 	if err := DecryptSensitiveFields(ch.Config); err != nil {
diff --git a/workspace-server/internal/channels/telegram.go b/workspace-server/internal/channels/telegram.go
index 778afa5c0..596975997 100644
--- a/workspace-server/internal/channels/telegram.go
+++ b/workspace-server/internal/channels/telegram.go
@@ -482,10 +482,12 @@ func (t *TelegramAdapter) StartPolling(ctx context.Context, config map[string]in
 				if apiErr.Code == 429 {
 					retryAfter := time.Duration(apiErr.RetryAfter) * time.Second
 					log.Printf("Channels: Telegram poll rate-limited, sleeping %s", retryAfter)
+					timer := time.NewTimer(retryAfter)
 					select {
 					case <-ctx.Done():
+						timer.Stop()
 						return nil
-					case <-time.After(retryAfter):
+					case <-timer.C:
 						continue
 					}
 				}
@@ -495,10 +497,12 @@ func (t *TelegramAdapter) StartPolling(ctx context.Context, config map[string]in
 				}
 			}
 			log.Printf("Channels: Telegram poll error: %v", err)
+			timer := time.NewTimer(telegramPollInterval)
 			select {
 			case <-ctx.Done():
+				timer.Stop()
 				return nil
-			case <-time.After(telegramPollInterval):
+			case <-timer.C:
 				continue
 			}
 		}
@@ -513,7 +517,9 @@ func (t *TelegramAdapter) StartPolling(ctx context.Context, config map[string]in
 
 				// Acknowledge the button press (removes loading spinner)
 				ackCfg := tgbotapi.NewCallback(cb.ID, "Received")
-				bot.Send(ackCfg)
+				if _, err := bot.Send(ackCfg); err != nil {
+					log.Printf("telegram: failed to send callback ack: %v", err)
+				}
 
 				// Update the message to show what was clicked
 				decision := "approved"
@@ -525,7 +531,9 @@ func (t *TelegramAdapter) StartPolling(ctx context.Context, config map[string]in
 					cb.Message.MessageID,
 					cb.Message.Text+"\n\n✅ CEO "+decision,
 				)
-				bot.Send(editMsg)
+				if _, err := bot.Send(editMsg); err != nil {
+					log.Printf("telegram: failed to send edit message: %v", err)
+				}
 
 				// Route the decision as an inbound message to the agent
 				inbound := &InboundMessage{
diff --git a/workspace-server/internal/handlers/a2a_proxy.go b/workspace-server/internal/handlers/a2a_proxy.go
index c06c553fd..16ff09ee8 100644
--- a/workspace-server/internal/handlers/a2a_proxy.go
+++ b/workspace-server/internal/handlers/a2a_proxy.go
@@ -932,7 +932,12 @@ func applyIdleTimeout(parent context.Context, b *events.Broadcaster, workspaceID
 	ctx, cancel := context.WithCancel(parent)
 	sub, unsub := b.SubscribeSSE(workspaceID)
 	go func() {
-		defer unsub()
+		defer func() {
+			if r := recover(); r != nil {
+				log.Printf("a2a_proxy: PANIC in SSE idle watcher for %s: %v", workspaceID, r)
+			}
+			unsub()
+		}()
 		timer := time.NewTimer(idle)
 		defer timer.Stop()
 		for {
diff --git a/workspace-server/internal/handlers/a2a_proxy_helpers.go b/workspace-server/internal/handlers/a2a_proxy_helpers.go
index 98c51bb7d..11916e6b1 100644
--- a/workspace-server/internal/handlers/a2a_proxy_helpers.go
+++ b/workspace-server/internal/handlers/a2a_proxy_helpers.go
@@ -407,15 +407,6 @@ func validateCallerToken(ctx context.Context, c *gin.Context, callerID string) e
 // matching (the wsauth errors are typed for the invalid case).
 var errInvalidCallerToken = errors.New("missing caller auth token")
 
-// canvasUserMessage holds the extracted user message extracted from an
-// A2A canvas request body for broadcasting to other sessions.
-type canvasUserMessage struct {
-	Message   string                               `json:"message,omitempty"`
-	Parts     []map[string]interface{}              `json:"parts,omitempty"`
-	MessageID string                               `json:"messageId,omitempty"`
-	Attachments []map[string]interface{}            `json:"attachments,omitempty"`
-}
-
 // extractCanvasUserMessage parses an A2A JSON-RPC request body and extracts
 // the user-authored text and attachments from a canvas-initiated message/send.
 // Returns nil when the body is not a canvas user message (empty, malformed,
diff --git a/workspace-server/internal/handlers/approvals.go b/workspace-server/internal/handlers/approvals.go
index dcce896d6..36e4c5e9c 100644
--- a/workspace-server/internal/handlers/approvals.go
+++ b/workspace-server/internal/handlers/approvals.go
@@ -51,23 +51,29 @@ func (h *ApprovalsHandler) Create(c *gin.Context) {
 		return
 	}
 
-	h.broadcaster.RecordAndBroadcast(ctx, string(events.EventApprovalRequested), workspaceID, map[string]interface{}{
+	if err := h.broadcaster.RecordAndBroadcast(ctx, string(events.EventApprovalRequested), workspaceID, map[string]interface{}{
 		"approval_id": approvalID,
 		"action":      body.Action,
 		"reason":      body.Reason,
 		"task_id":     body.TaskID,
-	})
+	}); err != nil {
+		log.Printf("approvals: failed to broadcast approval requested: %v", err)
+	}
 
 	// Auto-escalate to parent
 	var parentID *string
-	db.DB.QueryRowContext(ctx, `SELECT parent_id FROM workspaces WHERE id = $1`, workspaceID).Scan(&parentID)
+	if err := db.DB.QueryRowContext(ctx, `SELECT parent_id FROM workspaces WHERE id = $1`, workspaceID).Scan(&parentID); err != nil {
+		log.Printf("approvals: failed to lookup parent for escalation: %v", err)
+	}
 	if parentID != nil {
-		h.broadcaster.RecordAndBroadcast(ctx, string(events.EventApprovalEscalated), *parentID, map[string]interface{}{
+		if err := h.broadcaster.RecordAndBroadcast(ctx, string(events.EventApprovalEscalated), *parentID, map[string]interface{}{
 			"approval_id":       approvalID,
 			"from_workspace_id": workspaceID,
 			"action":            body.Action,
 			"reason":            body.Reason,
-		})
+		}); err != nil {
+			log.Printf("approvals: failed to broadcast approval escalated: %v", err)
+		}
 	}
 
 	c.JSON(http.StatusCreated, gin.H{"approval_id": approvalID, "status": "pending"})
@@ -80,10 +86,12 @@ func (h *ApprovalsHandler) ListAll(c *gin.Context) {
 	ctx := c.Request.Context()
 
 	// Auto-expire stale approvals (older than 10 min)
-	db.DB.ExecContext(ctx, `
+	if _, err := db.DB.ExecContext(ctx, `
 		UPDATE approval_requests SET status = 'denied', decided_by = 'auto-expired', decided_at = now()
 		WHERE status = 'pending' AND created_at < now() - interval '10 minutes'
-	`)
+	`); err != nil {
+		log.Printf("approvals: failed to auto-expire stale approvals: %v", err)
+	}
 
 	rows, err := db.DB.QueryContext(ctx, `
 		SELECT a.id, a.workspace_id, w.name, a.action, a.reason, a.status, a.created_at
@@ -211,11 +219,13 @@ func (h *ApprovalsHandler) Decide(c *gin.Context) {
 		eventType = "APPROVAL_DENIED"
 	}
 
-	h.broadcaster.RecordAndBroadcast(ctx, eventType, workspaceID, map[string]interface{}{
+	if err := h.broadcaster.RecordAndBroadcast(ctx, eventType, workspaceID, map[string]interface{}{
 		"approval_id": approvalID,
 		"decision":    body.Decision,
 		"decided_by":  decidedBy,
-	})
+	}); err != nil {
+		log.Printf("approvals: failed to broadcast approval decision: %v", err)
+	}
 
 	c.JSON(http.StatusOK, gin.H{"status": body.Decision, "approval_id": approvalID})
 }
diff --git a/workspace-server/internal/handlers/channels.go b/workspace-server/internal/handlers/channels.go
index 6d9008bf5..6fe36348d 100644
--- a/workspace-server/internal/handlers/channels.go
+++ b/workspace-server/internal/handlers/channels.go
@@ -558,6 +558,11 @@ func (h *ChannelHandler) Webhook(c *gin.Context) {
 
 	// Process asynchronously — don't block the webhook response
 	go func() {
+		defer func() {
+			if r := recover(); r != nil {
+				log.Printf("Channels: PANIC in async HandleInbound for workspace %s: %v", ch.WorkspaceID[:12], r)
+			}
+		}()
 		bgCtx := context.Background()
 		if err := h.manager.HandleInbound(bgCtx, ch, msg); err != nil {
 			log.Printf("Channels: async HandleInbound error for workspace %s: %v", ch.WorkspaceID[:12], err)
diff --git a/workspace-server/internal/handlers/discovery.go b/workspace-server/internal/handlers/discovery.go
index 33ec12637..7fa816383 100644
--- a/workspace-server/internal/handlers/discovery.go
+++ b/workspace-server/internal/handlers/discovery.go
@@ -239,7 +239,7 @@ func (h *DiscoveryHandler) Peers(c *gin.Context) {
 
 	// Siblings
 	if parentID.Valid {
-		siblings, _ := queryPeerMaps(`
+		siblings, _ := queryPeerMaps(ctx, `
 			SELECT w.id, w.name, COALESCE(w.role, ''), w.tier, w.status,
 				   COALESCE(w.agent_card, 'null'::jsonb), COALESCE(w.url, ''),
 				   w.parent_id, w.active_tasks
@@ -247,7 +247,7 @@ func (h *DiscoveryHandler) Peers(c *gin.Context) {
 			parentID.String, workspaceID)
 		peers = append(peers, siblings...)
 	} else {
-		siblings, _ := queryPeerMaps(`
+		siblings, _ := queryPeerMaps(ctx, `
 			SELECT w.id, w.name, COALESCE(w.role, ''), w.tier, w.status,
 				   COALESCE(w.agent_card, 'null'::jsonb), COALESCE(w.url, ''),
 				   w.parent_id, w.active_tasks
@@ -257,7 +257,7 @@ func (h *DiscoveryHandler) Peers(c *gin.Context) {
 	}
 
 	// Children
-	children, _ := queryPeerMaps(`
+	children, _ := queryPeerMaps(ctx, `
 		SELECT w.id, w.name, COALESCE(w.role, ''), w.tier, w.status,
 			   COALESCE(w.agent_card, 'null'::jsonb), COALESCE(w.url, ''),
 			   w.parent_id, w.active_tasks
@@ -266,7 +266,7 @@ func (h *DiscoveryHandler) Peers(c *gin.Context) {
 
 	// Parent
 	if parentID.Valid {
-		parent, _ := queryPeerMaps(`
+		parent, _ := queryPeerMaps(ctx, `
 			SELECT w.id, w.name, COALESCE(w.role, ''), w.tier, w.status,
 				   COALESCE(w.agent_card, 'null'::jsonb), COALESCE(w.url, ''),
 				   w.parent_id, w.active_tasks
@@ -303,8 +303,8 @@ func filterPeersByQuery(peers []map[string]interface{}, q string) []map[string]i
 }
 
 // queryPeerMaps returns clean JSON-serializable maps instead of Workspace structs.
-func queryPeerMaps(query string, args ...interface{}) ([]map[string]interface{}, error) {
-	rows, err := db.DB.Query(query, args...)
+func queryPeerMaps(ctx context.Context, query string, args ...interface{}) ([]map[string]interface{}, error) {
+	rows, err := db.DB.QueryContext(ctx, query, args...)
 	if err != nil {
 		log.Printf("queryPeerMaps error: %v", err)
 		return nil, err
diff --git a/workspace-server/internal/handlers/mcp_tools.go b/workspace-server/internal/handlers/mcp_tools.go
index 24e991bb9..1d135ea25 100644
--- a/workspace-server/internal/handlers/mcp_tools.go
+++ b/workspace-server/internal/handlers/mcp_tools.go
@@ -15,6 +15,7 @@ import (
 	"fmt"
 	"io"
 	"log"
+	"net"
 	"net/http"
 	"os"
 	"strings"
@@ -54,6 +55,22 @@ func updateMCPDelegationStatus(ctx context.Context, db *sql.DB, workspaceID, del
 	}
 }
 
+// ─────────────────────────────────────────────────────────────────────────────
+// mcpHTTPClient is a dedicated client for MCP bridge A2A calls.
+// Per-request deadlines are enforced via context (30 s sync, 8 s async).
+// Transport-level timeouts ensure dead workspaces fail fast instead of
+// hanging on OS default TCP timeouts (~75 s Linux).
+var mcpHTTPClient = &http.Client{
+	Transport: &http.Transport{
+		DialContext: (&net.Dialer{
+			Timeout:   5 * time.Second,
+			KeepAlive: 30 * time.Second,
+		}).DialContext,
+		ResponseHeaderTimeout: 30 * time.Second,
+		TLSHandshakeTimeout:   5 * time.Second,
+	},
+}
+
 // ─────────────────────────────────────────────────────────────────────────────
 // Tool implementations
 // ─────────────────────────────────────────────────────────────────────────────
@@ -231,7 +248,7 @@ func (h *MCPHandler) toolDelegateTask(ctx context.Context, callerID string, args
 	// so this header reflects a verified caller identity, not a spoofable value.
 	httpReq.Header.Set("X-Workspace-ID", callerID)
 
-	resp, err := http.DefaultClient.Do(httpReq)
+	resp, err := mcpHTTPClient.Do(httpReq)
 	if err != nil {
 		updateMCPDelegationStatus(ctx, h.database, callerID, delegationID, "failed", err.Error())
 		return "", fmt.Errorf("A2A call failed: %w", err)
@@ -279,6 +296,11 @@ func (h *MCPHandler) toolDelegateTaskAsync(ctx context.Context, callerID string,
 	// Fire and forget in a detached goroutine. Use a background context so
 	// the call is not cancelled when the HTTP request completes.
 	go func() {
+		defer func() {
+			if r := recover(); r != nil {
+				log.Printf("MCPHandler.delegate_task_async: PANIC for %s → %s: %v", callerID, targetID, r)
+			}
+		}()
 		bgCtx, cancel := context.WithTimeout(context.Background(), mcpAsyncCallTimeout)
 		defer cancel()
 
@@ -314,7 +336,7 @@ func (h *MCPHandler) toolDelegateTaskAsync(ctx context.Context, callerID string,
 		httpReq.Header.Set("Content-Type", "application/json")
 		httpReq.Header.Set("X-Workspace-ID", callerID)
 
-		resp, err := http.DefaultClient.Do(httpReq)
+		resp, err := mcpHTTPClient.Do(httpReq)
 		if err != nil {
 			log.Printf("MCPHandler.delegate_task_async: A2A call to %s: %v", targetID, err)
 			return
diff --git a/workspace-server/internal/handlers/plugins_install.go b/workspace-server/internal/handlers/plugins_install.go
index c335bf507..a5368b047 100644
--- a/workspace-server/internal/handlers/plugins_install.go
+++ b/workspace-server/internal/handlers/plugins_install.go
@@ -201,6 +201,11 @@ func (h *PluginsHandler) uninstallViaDocker(ctx context.Context, c *gin.Context,
 	// Auto-restart (small delay to ensure fs writes are flushed)
 	if h.restartFunc != nil {
 		go func() {
+			defer func() {
+				if r := recover(); r != nil {
+					log.Printf("plugins_install: PANIC in delayed restart for %s: %v", workspaceID, r)
+				}
+			}()
 			time.Sleep(2 * time.Second)
 			h.restartFunc(workspaceID)
 		}()
diff --git a/workspace-server/internal/handlers/restart_context.go b/workspace-server/internal/handlers/restart_context.go
index 3ae2f41d2..0b24bbcd1 100644
--- a/workspace-server/internal/handlers/restart_context.go
+++ b/workspace-server/internal/handlers/restart_context.go
@@ -133,24 +133,24 @@ func loadRestartContextData(ctx context.Context, workspaceID string) restartCont
 	// message bus.
 	keySet := map[string]struct{}{}
 	if rows, err := db.DB.QueryContext(ctx, `SELECT key FROM global_secrets`); err == nil {
+		defer rows.Close()
 		for rows.Next() {
 			var k string
 			if rows.Scan(&k) == nil {
 				keySet[k] = struct{}{}
 			}
 		}
-		rows.Close()
 	}
 	if rows, err := db.DB.QueryContext(ctx,
 		`SELECT key FROM workspace_secrets WHERE workspace_id = $1`, workspaceID,
 	); err == nil {
+		defer rows.Close()
 		for rows.Next() {
 			var k string
 			if rows.Scan(&k) == nil {
 				keySet[k] = struct{}{}
 			}
 		}
-		rows.Close()
 	}
 	for k := range keySet {
 		d.EnvKeys = append(d.EnvKeys, k)
@@ -163,6 +163,8 @@ func loadRestartContextData(ctx context.Context, workspaceID string) restartCont
 // workspace's status flips to 'online' or the deadline expires.
 // Returns true on success; callers log+drop on false.
 func waitForWorkspaceOnline(ctx context.Context, workspaceID string, timeout time.Duration) bool {
+	ticker := time.NewTicker(restartContextOnlinePollInterval)
+	defer ticker.Stop()
 	deadline := time.Now().Add(timeout)
 	for time.Now().Before(deadline) {
 		var status string
@@ -174,7 +176,7 @@ func waitForWorkspaceOnline(ctx context.Context, workspaceID string, timeout tim
 		select {
 		case <-ctx.Done():
 			return false
-		case <-time.After(restartContextOnlinePollInterval):
+		case <-ticker.C:
 		}
 	}
 	return false
diff --git a/workspace-server/internal/handlers/terminal.go b/workspace-server/internal/handlers/terminal.go
index 8007391a8..313a92fd4 100644
--- a/workspace-server/internal/handlers/terminal.go
+++ b/workspace-server/internal/handlers/terminal.go
@@ -213,7 +213,12 @@ func (h *TerminalHandler) handleLocalConnect(c *gin.Context, workspaceID string)
 	// Bridge: container stdout → WebSocket
 	done := make(chan struct{})
 	go func() {
-		defer close(done)
+		defer func() {
+			if r := recover(); r != nil {
+				log.Printf("Terminal: PANIC in stdout bridge: %v", r)
+			}
+			close(done)
+		}()
 		buf := make([]byte, 4096)
 		for {
 			n, err := resp.Reader.Read(buf)
@@ -434,7 +439,12 @@ func (h *TerminalHandler) handleRemoteConnect(c *gin.Context, workspaceID, insta
 
 	// PTY → WebSocket
 	go func() {
-		defer close(done)
+		defer func() {
+			if r := recover(); r != nil {
+				log.Printf("Terminal: PANIC in PTY bridge: %v", r)
+			}
+			close(done)
+		}()
 		buf := make([]byte, 4096)
 		for {
 			n, err := ptmx.Read(buf)
@@ -456,6 +466,11 @@ func (h *TerminalHandler) handleRemoteConnect(c *gin.Context, workspaceID, insta
 
 	// WebSocket → PTY (stdin)
 	go func() {
+		defer func() {
+			if r := recover(); r != nil {
+				log.Printf("Terminal: PANIC in stdin loop: %v", r)
+			}
+		}()
 		for {
 			_, msg, rErr := conn.ReadMessage()
 			if rErr != nil {
diff --git a/workspace-server/internal/handlers/workspace.go b/workspace-server/internal/handlers/workspace.go
index 781741aad..0be412352 100644
--- a/workspace-server/internal/handlers/workspace.go
+++ b/workspace-server/internal/handlers/workspace.go
@@ -296,6 +296,7 @@ func (h *WorkspaceHandler) Create(c *gin.Context) {
 		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to create workspace"})
 		return
 	}
+	defer func() { _ = tx.Rollback() }()
 
 	maxConcurrent := payload.MaxConcurrentTasks
 	if maxConcurrent <= 0 {
@@ -478,7 +479,9 @@ func (h *WorkspaceHandler) Create(c *gin.Context) {
 			// from the external agent (with this token + its URL)
 			// flips the row to online.
 			// Preserve BYO-compute runtime label (kimi, kimi-cli, external).
-			db.DB.ExecContext(ctx, `UPDATE workspaces SET status = $1, runtime = $2, updated_at = now() WHERE id = $3`, models.StatusAwaitingAgent, normalizeExternalRuntime(payload.Runtime), id)
+			if _, err := db.DB.ExecContext(ctx, `UPDATE workspaces SET status = $1, runtime = $2, updated_at = now() WHERE id = $3`, models.StatusAwaitingAgent, normalizeExternalRuntime(payload.Runtime), id); err != nil {
+				log.Printf("External workspace %s: status update failed: %v", id, err)
+			}
 			tok, tokErr := wsauth.IssueToken(ctx, db.DB, id)
 			if tokErr != nil {
 				log.Printf("External workspace %s: token issuance failed: %v", id, tokErr)
diff --git a/workspace-server/internal/handlers/workspace_provision.go b/workspace-server/internal/handlers/workspace_provision.go
index 24135bf6e..06bc6bb5e 100644
--- a/workspace-server/internal/handlers/workspace_provision.go
+++ b/workspace-server/internal/handlers/workspace_provision.go
@@ -261,7 +261,7 @@ func (h *WorkspaceHandler) buildProvisionerConfig(
 	workspaceAccess := payload.WorkspaceAccess
 	if (workspacePath == "" || workspaceAccess == "") && db.DB != nil {
 		var dbDir, dbAccess string
-		if err := db.DB.QueryRow(
+		if err := db.DB.QueryRowContext(ctx,
 			`SELECT COALESCE(workspace_dir, ''), COALESCE(workspace_access, 'none') FROM workspaces WHERE id = $1`,
 			workspaceID,
 		).Scan(&dbDir, &dbAccess); err == nil {
diff --git a/workspace-server/internal/middleware/mcp_ratelimit.go b/workspace-server/internal/middleware/mcp_ratelimit.go
index c8f76b578..3133e2358 100644
--- a/workspace-server/internal/middleware/mcp_ratelimit.go
+++ b/workspace-server/internal/middleware/mcp_ratelimit.go
@@ -4,6 +4,7 @@ import (
 	"context"
 	"crypto/sha256"
 	"fmt"
+	"log"
 	"net/http"
 	"strconv"
 	"strings"
@@ -41,6 +42,11 @@ func NewMCPRateLimiter(rate int, interval time.Duration, ctx context.Context) *M
 		interval: interval,
 	}
 	go func() {
+		defer func() {
+			if r := recover(); r != nil {
+				log.Printf("mcp_ratelimit: PANIC in bucket cleanup: %v", r)
+			}
+		}()
 		ticker := time.NewTicker(5 * time.Minute)
 		defer ticker.Stop()
 		for {
diff --git a/workspace-server/internal/middleware/ratelimit.go b/workspace-server/internal/middleware/ratelimit.go
index e01324d39..c3c09dbfc 100644
--- a/workspace-server/internal/middleware/ratelimit.go
+++ b/workspace-server/internal/middleware/ratelimit.go
@@ -3,6 +3,7 @@ package middleware
 
 import (
 	"context"
+	"log"
 	"net/http"
 	"strconv"
 	"strings"
@@ -35,6 +36,11 @@ func NewRateLimiter(rate int, interval time.Duration, ctx context.Context) *Rate
 		interval: interval,
 	}
 	go func() {
+		defer func() {
+			if r := recover(); r != nil {
+				log.Printf("ratelimit: PANIC in bucket cleanup: %v", r)
+			}
+		}()
 		ticker := time.NewTicker(5 * time.Minute)
 		defer ticker.Stop()
 		for {
diff --git a/workspace-server/internal/middleware/session_auth.go b/workspace-server/internal/middleware/session_auth.go
index 3f6d058d6..62c7a970b 100644
--- a/workspace-server/internal/middleware/session_auth.go
+++ b/workspace-server/internal/middleware/session_auth.go
@@ -116,6 +116,11 @@ func sessionCachePut(key string, ok bool) {
 
 func init() {
 	go func() {
+		defer func() {
+			if r := recover(); r != nil {
+				log.Printf("session_auth: PANIC in cache sweeper: %v", r)
+			}
+		}()
 		// Jitter startup so restarts don't align sweeps.
 		time.Sleep(time.Duration(rand.Int64N(int64(sessionCacheSweepEvery))))
 		t := time.NewTicker(sessionCacheSweepEvery)
diff --git a/workspace-server/internal/supervised/supervised.go b/workspace-server/internal/supervised/supervised.go
index 8de84258f..e9c6a40f3 100644
--- a/workspace-server/internal/supervised/supervised.go
+++ b/workspace-server/internal/supervised/supervised.go
@@ -60,10 +60,12 @@ func RunWithRecover(ctx context.Context, name string, fn func(context.Context))
 		}
 
 		// Panic → back off and restart.
+		timer := time.NewTimer(backoff)
 		select {
 		case <-ctx.Done():
+			timer.Stop()
 			return
-		case <-time.After(backoff):
+		case <-timer.C:
 		}
 		if backoff < maxBackoff {
 			backoff *= 2