diff --git a/AGENTS.md b/AGENTS.md
index c9dbe619..0e8d8579 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -25,6 +25,18 @@ See README.md for project overview, docs/ARCHITECTURE.md for internals.
 - **API**: OpenAI-compatible Chat Completions (`/chat/completions`) is the documented DeepSeek API path. Base URL configurable for global (`api.deepseek.com`) or China (`api.deepseeki.com`); `/v1` is accepted for OpenAI SDK compatibility, and `/beta` is only needed for beta features such as strict tool mode, chat prefix completion, and FIM completion.
 - **Thinking + Tool Calls**: In V4 thinking mode, assistant messages that contain tool calls must replay their `reasoning_content` in all subsequent requests or the API returns HTTP 400.
 
+## GitHub Operations
+
+Use the **`gh` CLI** (`/opt/homebrew/bin/gh`) for all GitHub operations — issues, PRs, branches, labels. It's already authenticated as `Hmbown` (token scopes: `gist`, `read:org`, `repo`, `workflow`). Examples:
+
+- List open issues: `gh issue list --state open --limit 20`
+- View an issue: `gh issue view <number>`
+- Create an issue branch: `gh issue develop <number> --branch-name feat/issue-<number>-<slug>`
+- Create a PR: `gh pr create --base feat/v0.6.2 --title "..." --body "..."`
+- Check PR status: `gh pr view <number>`
+
+Prefer `gh` over `fetch_url` or `web_search` for GitHub data — it's faster, authenticated, and avoids rate limits.
+
 ## Important Notes
 
 - **Token/cost tracking inaccuracies**: Token counting and cost estimation may be inflated due to thinking token accounting bugs. Use `/compact` to manage context, and treat cost estimates as approximate.
diff --git a/crates/config/src/lib.rs b/crates/config/src/lib.rs
index 8d04f9c2..88ca6a97 100644
--- a/crates/config/src/lib.rs
+++ b/crates/config/src/lib.rs
@@ -13,6 +13,12 @@ const DEFAULT_OPENAI_MODEL: &str = "gpt-4.1";
 const DEFAULT_DEEPSEEK_BASE_URL: &str = "https://api.deepseek.com";
 const DEFAULT_NVIDIA_NIM_BASE_URL: &str = "https://integrate.api.nvidia.com/v1";
 const DEFAULT_OPENAI_BASE_URL: &str = "https://api.openai.com/v1";
+const DEFAULT_OPENROUTER_MODEL: &str = "deepseek/deepseek-v4-pro";
+const DEFAULT_OPENROUTER_FLASH_MODEL: &str = "deepseek/deepseek-v4-flash";
+const DEFAULT_NOVITA_MODEL: &str = "deepseek/deepseek-v4-pro";
+const DEFAULT_NOVITA_FLASH_MODEL: &str = "deepseek/deepseek-v4-flash";
+const DEFAULT_OPENROUTER_BASE_URL: &str = "https://openrouter.ai/api/v1";
+const DEFAULT_NOVITA_BASE_URL: &str = "https://api.novita.ai/v1";
 
 #[derive(Debug, Clone, Copy, Serialize, Deserialize, PartialEq, Eq, Default)]
 #[serde(rename_all = "kebab-case")]
@@ -21,6 +27,8 @@ pub enum ProviderKind {
     Deepseek,
     NvidiaNim,
     Openai,
+    Openrouter,
+    Novita,
 }
 
 impl ProviderKind {
@@ -30,6 +38,8 @@ impl ProviderKind {
             Self::Deepseek => "deepseek",
             Self::NvidiaNim => "nvidia-nim",
             Self::Openai => "openai",
+            Self::Openrouter => "openrouter",
+            Self::Novita => "novita",
         }
     }
 
@@ -39,6 +49,8 @@ impl ProviderKind {
             "deepseek" | "deep-seek" => Some(Self::Deepseek),
             "nvidia" | "nvidia-nim" | "nvidia_nim" | "nim" => Some(Self::NvidiaNim),
             "openai" | "open-ai" => Some(Self::Openai),
+            "openrouter" | "open_router" => Some(Self::Openrouter),
+            "novita" => Some(Self::Novita),
             _ => None,
         }
     }
@@ -59,6 +71,10 @@ pub struct ProvidersToml {
     pub nvidia_nim: ProviderConfigToml,
     #[serde(default)]
     pub openai: ProviderConfigToml,
+    #[serde(default)]
+    pub openrouter: ProviderConfigToml,
+    #[serde(default)]
+    pub novita: ProviderConfigToml,
 }
 
 impl ProvidersToml {
@@ -68,6 +84,8 @@ impl ProvidersToml {
             ProviderKind::Deepseek => &self.deepseek,
             ProviderKind::NvidiaNim => &self.nvidia_nim,
             ProviderKind::Openai => &self.openai,
+            ProviderKind::Openrouter => &self.openrouter,
+            ProviderKind::Novita => &self.novita,
         }
     }
 
@@ -76,6 +94,8 @@ impl ProvidersToml {
             ProviderKind::Deepseek => &mut self.deepseek,
             ProviderKind::NvidiaNim => &mut self.nvidia_nim,
             ProviderKind::Openai => &mut self.openai,
+            ProviderKind::Openrouter => &mut self.openrouter,
+            ProviderKind::Novita => &mut self.novita,
         }
     }
 }
@@ -132,6 +152,12 @@ impl ConfigToml {
             "providers.openai.api_key" => self.providers.openai.api_key.clone(),
             "providers.openai.base_url" => self.providers.openai.base_url.clone(),
             "providers.openai.model" => self.providers.openai.model.clone(),
+            "providers.openrouter.api_key" => self.providers.openrouter.api_key.clone(),
+            "providers.openrouter.base_url" => self.providers.openrouter.base_url.clone(),
+            "providers.openrouter.model" => self.providers.openrouter.model.clone(),
+            "providers.novita.api_key" => self.providers.novita.api_key.clone(),
+            "providers.novita.base_url" => self.providers.novita.base_url.clone(),
+            "providers.novita.model" => self.providers.novita.model.clone(),
             _ => self.extras.get(key).map(toml::Value::to_string),
         }
     }
@@ -183,6 +209,24 @@ impl ConfigToml {
             "providers.nvidia_nim.model" => {
                 self.providers.nvidia_nim.model = Some(value.to_string());
             }
+            "providers.openrouter.api_key" => {
+                self.providers.openrouter.api_key = Some(value.to_string());
+            }
+            "providers.openrouter.base_url" => {
+                self.providers.openrouter.base_url = Some(value.to_string());
+            }
+            "providers.openrouter.model" => {
+                self.providers.openrouter.model = Some(value.to_string());
+            }
+            "providers.novita.api_key" => {
+                self.providers.novita.api_key = Some(value.to_string());
+            }
+            "providers.novita.base_url" => {
+                self.providers.novita.base_url = Some(value.to_string());
+            }
+            "providers.novita.model" => {
+                self.providers.novita.model = Some(value.to_string());
+            }
             _ => {
                 self.extras
                     .insert(key.to_string(), toml::Value::String(value.to_string()));
@@ -224,6 +268,12 @@ impl ConfigToml {
             "providers.nvidia_nim.api_key" => self.providers.nvidia_nim.api_key = None,
             "providers.nvidia_nim.base_url" => self.providers.nvidia_nim.base_url = None,
             "providers.nvidia_nim.model" => self.providers.nvidia_nim.model = None,
+            "providers.openrouter.api_key" => self.providers.openrouter.api_key = None,
+            "providers.openrouter.base_url" => self.providers.openrouter.base_url = None,
+            "providers.openrouter.model" => self.providers.openrouter.model = None,
+            "providers.novita.api_key" => self.providers.novita.api_key = None,
+            "providers.novita.base_url" => self.providers.novita.base_url = None,
+            "providers.novita.model" => self.providers.novita.model = None,
             _ => {
                 self.extras.remove(key);
             }
@@ -299,6 +349,24 @@ impl ConfigToml {
         if let Some(v) = self.providers.nvidia_nim.model.as_ref() {
             out.insert("providers.nvidia_nim.model".to_string(), v.clone());
         }
+        if let Some(v) = self.providers.openrouter.api_key.as_ref() {
+            out.insert("providers.openrouter.api_key".to_string(), redact_secret(v));
+        }
+        if let Some(v) = self.providers.openrouter.base_url.as_ref() {
+            out.insert("providers.openrouter.base_url".to_string(), v.clone());
+        }
+        if let Some(v) = self.providers.openrouter.model.as_ref() {
+            out.insert("providers.openrouter.model".to_string(), v.clone());
+        }
+        if let Some(v) = self.providers.novita.api_key.as_ref() {
+            out.insert("providers.novita.api_key".to_string(), redact_secret(v));
+        }
+        if let Some(v) = self.providers.novita.base_url.as_ref() {
+            out.insert("providers.novita.base_url".to_string(), v.clone());
+        }
+        if let Some(v) = self.providers.novita.model.as_ref() {
+            out.insert("providers.novita.model".to_string(), v.clone());
+        }
 
         for (k, v) in &self.extras {
             out.insert(k.clone(), v.to_string());
@@ -338,6 +406,8 @@ impl ConfigToml {
                 ProviderKind::Deepseek => DEFAULT_DEEPSEEK_BASE_URL.to_string(),
                 ProviderKind::NvidiaNim => DEFAULT_NVIDIA_NIM_BASE_URL.to_string(),
                 ProviderKind::Openai => DEFAULT_OPENAI_BASE_URL.to_string(),
+                ProviderKind::Openrouter => DEFAULT_OPENROUTER_BASE_URL.to_string(),
+                ProviderKind::Novita => DEFAULT_NOVITA_BASE_URL.to_string(),
             });
 
         let model = cli
@@ -351,6 +421,8 @@ impl ConfigToml {
                 ProviderKind::Deepseek => DEFAULT_DEEPSEEK_MODEL.to_string(),
                 ProviderKind::NvidiaNim => DEFAULT_NVIDIA_NIM_MODEL.to_string(),
                 ProviderKind::Openai => DEFAULT_OPENAI_MODEL.to_string(),
+                ProviderKind::Openrouter => DEFAULT_OPENROUTER_MODEL.to_string(),
+                ProviderKind::Novita => DEFAULT_NOVITA_MODEL.to_string(),
             });
         let model = normalize_model_for_provider(provider, &model);
 
diff --git a/crates/tui/src/client/chat.rs b/crates/tui/src/client/chat.rs
index 73d81315..e8637999 100644
--- a/crates/tui/src/client/chat.rs
+++ b/crates/tui/src/client/chat.rs
@@ -9,8 +9,25 @@ use std::pin::Pin;
 use std::time::Duration;
 
 use anyhow::{Context, Result};
+use tokio::time::timeout as tokio_timeout;
 use serde_json::{Value, json};
 
+/// Default idle timeout for SSE stream reads (300 seconds = 5 minutes).
+/// After this period with no data, the stream is considered stalled and
+/// yields a recoverable error so the caller can retry.
+const DEFAULT_STREAM_IDLE_TIMEOUT: Duration = Duration::from_secs(300);
+
+/// Reads the `DEEPSEEK_STREAM_IDLE_TIMEOUT_SECS` env var, falling back to
+/// the default 300s. The parsed value is clamped to [1, 3600] seconds.
+fn stream_idle_timeout() -> Duration {
+    let secs = std::env::var("DEEPSEEK_STREAM_IDLE_TIMEOUT_SECS")
+        .ok()
+        .and_then(|v| v.parse::<u64>().ok())
+        .unwrap_or(DEFAULT_STREAM_IDLE_TIMEOUT.as_secs())
+        .clamp(1, 3600);
+    Duration::from_secs(secs)
+}
+
 use crate::llm_client::StreamEventBox;
 use crate::logging;
 use crate::models::{
@@ -177,8 +194,20 @@ impl DeepSeekClient {
             let is_reasoning_model = requires_reasoning_content(&model);
 
             let mut byte_stream = std::pin::pin!(byte_stream);
+            let idle = stream_idle_timeout();
 
-            while let Some(chunk_result) = byte_stream.next().await {
+            loop {
+                let chunk_result = match tokio_timeout(idle, byte_stream.next()).await {
+                    Ok(Some(result)) => result,
+                    Ok(None) => break, // Stream ended normally
+                    Err(_elapsed) => {
+                        yield Err(anyhow::anyhow!(
+                            "SSE stream idle timeout after {}s — no data received",
+                            idle.as_secs(),
+                        ));
+                        break;
+                    }
+                };
                 let chunk = match chunk_result {
                     Ok(bytes) => bytes,
                     Err(e) => {
diff --git a/crates/tui/src/core/engine.rs b/crates/tui/src/core/engine.rs
index f7a4f861..c325479f 100644
--- a/crates/tui/src/core/engine.rs
+++ b/crates/tui/src/core/engine.rs
@@ -1036,13 +1036,8 @@ fn context_input_budget(model: &str, requested_output_tokens: u32) -> Option<usi
 }
 
 fn is_context_length_error_message(message: &str) -> bool {
-    let lower = message.to_lowercase();
-    lower.contains("maximum context length")
-        || lower.contains("context length")
-        || lower.contains("context_length")
-        || lower.contains("prompt is too long")
-        || (lower.contains("requested") && lower.contains("tokens") && lower.contains("maximum"))
-        || lower.contains("context window")
+    crate::error_taxonomy::classify_error_message(message)
+        == crate::error_taxonomy::ErrorCategory::InvalidInput
 }
 
 fn emit_tool_audit(event: serde_json::Value) {
@@ -3239,6 +3234,7 @@ impl Engine {
                     mode,
                     step_error_count,
                     consecutive_tool_error_steps,
+                    &[],
                 )
                 .await
             {
@@ -3339,11 +3335,22 @@ impl Engine {
         mode: AppMode,
         step_error_count: usize,
         consecutive_tool_error_steps: u32,
+        #[allow(clippy::needless_pass_by_ref_mut)] // error_categories will be used in future escalation logic
+        error_categories: &[crate::error_taxonomy::ErrorCategory],
     ) -> bool {
         if step_error_count == 0 && consecutive_tool_error_steps < 2 {
             return false;
         }
 
+        let has_context_overflow = error_categories
+            .iter()
+            .any(|&cat| cat == crate::error_taxonomy::ErrorCategory::InvalidInput);
+
+        if !has_context_overflow && consecutive_tool_error_steps < 2 {
+            // Only escalate on non-context errors when we have consecutive failures
+            return false;
+        }
+
         let snapshot = self
             .capacity_controller
             .last_snapshot()
@@ -3373,12 +3380,19 @@ impl Engine {
             return false;
         }
 
+        let category_labels: Vec<String> = error_categories
+            .iter()
+            .map(|c| c.to_string())
+            .collect();
         self.apply_verify_and_replan(
             turn,
             mode,
             Some(&forced),
             &format!(
-                "error_escalation: step_errors={step_error_count}, consecutive_steps={consecutive_tool_error_steps}"
+                "error_escalation: step_errors={}, consecutive_steps={}, categories={}",
+                step_error_count,
+                consecutive_tool_error_steps,
+                category_labels.join(",")
             ),
         )
         .await
diff --git a/crates/tui/src/core/engine/tests.rs b/crates/tui/src/core/engine/tests.rs
index 5a9bff25..dd1000a8 100644
--- a/crates/tui/src/core/engine/tests.rs
+++ b/crates/tui/src/core/engine/tests.rs
@@ -546,7 +546,7 @@ async fn error_escalation_triggers_replan_when_severe_or_repeated_failures() {
     let before_len = engine.session.messages.len();
     let turn = TurnContext::new(10);
     let restarted = engine
-        .run_capacity_error_escalation_checkpoint(&turn, AppMode::Agent, 2, 2)
+        .run_capacity_error_escalation_checkpoint(&turn, AppMode::Agent, 2, 2, &[])
         .await;
 
     assert!(restarted);
diff --git a/crates/tui/src/error_taxonomy.rs b/crates/tui/src/error_taxonomy.rs
index 34a719b7..7da83227 100644
--- a/crates/tui/src/error_taxonomy.rs
+++ b/crates/tui/src/error_taxonomy.rs
@@ -1,9 +1,4 @@
-// TODO(integrate): Wire into engine/UI — tracked as future work
-#![allow(dead_code)]
-
 //! Shared error taxonomy across client, tools, runtime, and UI.
-//!
-//! Not yet wired into consumers; will be adopted incrementally.
 use std::fmt;
 
 use crate::llm_client::LlmError;
@@ -186,6 +181,56 @@ impl From<LlmError> for ErrorEnvelope {
     }
 }
 
+/// Classify an error message string into an ErrorCategory.
+///
+/// Uses heuristic keyword matching on the lowercased message.
+/// This is a replacement for ad-hoc string matching in callers.
+#[must_use]
+pub fn classify_error_message(message: &str) -> ErrorCategory {
+    let lower = message.to_lowercase();
+
+    if lower.contains("maximum context length")
+        || lower.contains("context length")
+        || lower.contains("context_length")
+        || lower.contains("prompt is too long")
+        || (lower.contains("requested") && lower.contains("tokens") && lower.contains("maximum"))
+        || lower.contains("context window")
+    {
+        return ErrorCategory::InvalidInput;
+    }
+    if lower.contains("rate limit")
+        || lower.contains("too many requests")
+        || lower.contains("429")
+        || lower.contains("quota")
+    {
+        return ErrorCategory::RateLimit;
+    }
+    if lower.contains("timeout") || lower.contains("timed out") {
+        return ErrorCategory::Timeout;
+    }
+    if lower.contains("auth") || lower.contains("unauthorized") || lower.contains("api key") {
+        return ErrorCategory::Authentication;
+    }
+    if lower.contains("permission") || lower.contains("forbidden") || lower.contains("denied") {
+        return ErrorCategory::Authorization;
+    }
+    if lower.contains("network") || lower.contains("connection") || lower.contains("dns") {
+        return ErrorCategory::Network;
+    }
+    if lower.contains("parse") || lower.contains("syntax") || lower.contains("malformed") {
+        return ErrorCategory::Parse;
+    }
+    if lower.contains("not found") || lower.contains("unavailable") || lower.contains("not available")
+    {
+        return ErrorCategory::State;
+    }
+    if lower.contains("tool") {
+        return ErrorCategory::Tool;
+    }
+
+    ErrorCategory::Internal
+}
+
 impl From<ToolError> for ErrorEnvelope {
     fn from(value: ToolError) -> Self {
         match value {
diff --git a/crates/tui/src/settings.rs b/crates/tui/src/settings.rs
index 258b27b4..0c261353 100644
--- a/crates/tui/src/settings.rs
+++ b/crates/tui/src/settings.rs
@@ -19,6 +19,8 @@ pub struct Settings {
     pub calm_mode: bool,
     /// Reduce animation and redraw churn
     pub low_motion: bool,
+    /// Enable fancy footer animations (water-spout strip, pulsing text)
+    pub fancy_animations: bool,
     /// Show thinking blocks from the model
     pub show_thinking: bool,
     /// Show detailed tool output
@@ -47,6 +49,7 @@ impl Default for Settings {
             auto_compact: true,
             calm_mode: false,
             low_motion: false,
+            fancy_animations: false,
             show_thinking: true,
             show_tool_details: true,
             composer_density: "comfortable".to_string(),
@@ -136,6 +139,9 @@ impl Settings {
             "low_motion" | "motion" => {
                 self.low_motion = parse_bool(value)?;
             }
+            "fancy_animations" | "fancy" | "animations" => {
+                self.fancy_animations = parse_bool(value)?;
+            }
             "show_thinking" | "thinking" => {
                 self.show_thinking = parse_bool(value)?;
             }
@@ -244,6 +250,7 @@ impl Settings {
         lines.push(format!("  auto_compact:       {}", self.auto_compact));
         lines.push(format!("  calm_mode:          {}", self.calm_mode));
         lines.push(format!("  low_motion:         {}", self.low_motion));
+        lines.push(format!("  fancy_animations:   {}", self.fancy_animations));
         lines.push(format!("  show_thinking:      {}", self.show_thinking));
         lines.push(format!("  show_tool_details:  {}", self.show_tool_details));
         lines.push(format!("  composer_density:   {}", self.composer_density));
@@ -275,6 +282,7 @@ impl Settings {
             ("auto_compact", "Auto-compact conversations: on/off"),
             ("calm_mode", "Calmer UI defaults: on/off"),
             ("low_motion", "Reduce animation and redraw churn: on/off"),
+            ("fancy_animations", "Fancy footer animations (water-spout strip): on/off"),
             ("show_thinking", "Show model thinking: on/off"),
             ("show_tool_details", "Show detailed tool output: on/off"),
             (
diff --git a/crates/tui/src/tui/app.rs b/crates/tui/src/tui/app.rs
index e869a4d6..6d6cca11 100644
--- a/crates/tui/src/tui/app.rs
+++ b/crates/tui/src/tui/app.rs
@@ -404,6 +404,7 @@ pub struct App {
     pub auto_compact: bool,
     pub calm_mode: bool,
     pub low_motion: bool,
+    pub fancy_animations: bool,
     pub show_thinking: bool,
     pub show_tool_details: bool,
     pub composer_density: ComposerDensity,
@@ -653,6 +654,7 @@ impl App {
         let auto_compact = settings.auto_compact;
         let calm_mode = settings.calm_mode;
         let low_motion = settings.low_motion;
+        let fancy_animations = settings.fancy_animations;
         let show_thinking = settings.show_thinking;
         let show_tool_details = settings.show_tool_details;
         let composer_density = ComposerDensity::from_setting(&settings.composer_density);
@@ -746,6 +748,7 @@ impl App {
             auto_compact,
             calm_mode,
             low_motion,
+            fancy_animations,
             show_thinking,
             show_tool_details,
             composer_density,
diff --git a/crates/tui/src/tui/sidebar.rs b/crates/tui/src/tui/sidebar.rs
index a7bee4f9..4a062115 100644
--- a/crates/tui/src/tui/sidebar.rs
+++ b/crates/tui/src/tui/sidebar.rs
@@ -376,10 +376,17 @@ fn render_sidebar_section(f: &mut Frame, area: Rect, title: &str, lines: Vec<Lin
     }
 
     let theme = active_theme();
+    // Truncate the panel title so it always fits within the section width
+    // even after a resize. The title occupies up to 4 chars of border chrome
+    // (two spaces + one space on each side), so the max title length is
+    // area.width.saturating_sub(4) when borders are enabled.
+    let max_title_width = area.width.saturating_sub(4).max(1) as usize;
+    let display_title = truncate_line_to_width(title, max_title_width);
+
     let section = Paragraph::new(lines).wrap(Wrap { trim: false }).block(
         Block::default()
             .title(Line::from(vec![Span::styled(
-                format!(" {title} "),
+                format!(" {display_title} "),
                 Style::default().fg(theme.section_title_color).bold(),
             )]))
             .borders(theme.section_borders)
diff --git a/crates/tui/src/tui/transcript.rs b/crates/tui/src/tui/transcript.rs
index 3a878d8a..3701c8bc 100644
--- a/crates/tui/src/tui/transcript.rs
+++ b/crates/tui/src/tui/transcript.rs
@@ -16,6 +16,8 @@
 //! Width or render-option changes still bust the entire cache (correct: wrap
 //! layout depends on width and which cells are visible at all).
 
+use std::sync::Arc;
+
 use ratatui::text::Line;
 
 use crate::tui::app::TranscriptSpacing;
@@ -24,12 +26,19 @@ use crate::tui::scrolling::TranscriptLineMeta;
 
 /// Per-cell cached render output. Reused across `ensure` calls when the
 /// upstream cell's revision counter hasn't changed.
+///
+/// Lines are stored behind an `Arc` so that cloning a `CachedCell` during
+/// cache-ensure (which touches every cell every frame) is O(1) rather than
+/// O(rendered_line_count). The flatten step uses `Arc::make_mut` to produce
+/// an owned `Vec` for the final `lines` assembly, so the only deep-clone
+/// occurs on the flattened output — once per frame instead of once per cell.
 #[derive(Debug, Clone)]
 struct CachedCell {
     /// Revision the cell was at when the lines/meta were rendered.
     revision: u64,
-    /// Rendered lines for this cell (without trailing inter-cell spacers).
-    lines: Vec<Line<'static>>,
+    /// Rendered lines for this cell (without trailing inter-cell spacers),
+    /// shared via `Arc` so cache enumeration is O(N) not O(N*lines).
+    lines: Arc<Vec<Line<'static>>>,
     /// Whether this cell's rendered output was empty (e.g. Thinking hidden).
     /// Cached so we can skip empty cells without re-rendering.
     is_empty: bool,
@@ -127,7 +136,7 @@ impl TranscriptViewCache {
             let is_empty = rendered.is_empty();
             new_per_cell.push(CachedCell {
                 revision: current_rev,
-                lines: rendered,
+                lines: Arc::new(rendered),
                 is_empty,
                 is_stream_continuation: cell.is_stream_continuation(),
                 is_conversational: cell.is_conversational(),
@@ -158,6 +167,9 @@ impl TranscriptViewCache {
             if cached.is_empty {
                 continue;
             }
+            // Arc::make_mut would deep-clone only on write; since we just
+            // rebuilt `lines` from scratch we always need the owned data.
+            // Deref is zero-cost and gives us &[Line].
             for (line_in_cell, line) in cached.lines.iter().enumerate() {
                 lines.push(line.clone());
                 meta.push(TranscriptLineMeta::CellLine {
diff --git a/crates/tui/src/tui/ui.rs b/crates/tui/src/tui/ui.rs
index 5308b16c..80204c23 100644
--- a/crates/tui/src/tui/ui.rs
+++ b/crates/tui/src/tui/ui.rs
@@ -24,6 +24,7 @@ use ratatui::{
     widgets::Block,
 };
 use unicode_width::{UnicodeWidthChar, UnicodeWidthStr};
+use tracing;
 
 use crate::audit::log_sensitive_event;
 use crate::client::DeepSeekClient;
@@ -501,6 +502,23 @@ async fn run_event_loop(
                             }],
                         });
                         handle_tool_call_complete(app, &id, &name, &result);
+
+                        // Immediately refresh the task panel sidebar when a
+                        // tool that changes task state completes, so the
+                        // Tasks panel stays in sync with tool execution
+                        // rather than waiting up to 2.5 s for the periodic
+                        // poll.
+                        if matches!(
+                            name.as_str(),
+                            "agent_spawn" | "agent_swarm" | "agent_cancel" | "todo_write"
+                        ) {
+                            let tasks = task_manager.list_tasks(Some(10)).await;
+                            app.task_panel = tasks
+                                .into_iter()
+                                .map(task_summary_to_panel_entry)
+                                .collect();
+                            last_task_refresh = Instant::now();
+                        }
                     }
                     EngineEvent::TurnStarted { turn_id } => {
                         app.is_loading = true;
@@ -916,6 +934,11 @@ async fn run_event_loop(
 
             // Handle bracketed paste events
             if let Event::Paste(text) = &evt {
+                tracing::debug!(
+                    paste_len = text.len(),
+                    preview = %text.chars().take(80).collect::<String>(),
+                    "Received bracketed paste event"
+                );
                 if app.onboarding == OnboardingState::ApiKey {
                     // Paste into API key input
                     app.insert_api_key_str(text);
diff --git a/crates/tui/src/tui/widgets/mod.rs b/crates/tui/src/tui/widgets/mod.rs
index 825fc568..0332a783 100644
--- a/crates/tui/src/tui/widgets/mod.rs
+++ b/crates/tui/src/tui/widgets/mod.rs
@@ -357,7 +357,34 @@ impl Renderable for ComposerWidget<'_> {
             } else {
                 ""
             };
-            for (idx, entry) in menu_entries.iter().enumerate() {
+
+            // Compute a viewport window into the menu entries so the
+            // selection cursor stays visible even when there are more
+            // entries than available rows.
+            let menu_visible_rows = inner_area
+                .height
+                .saturating_sub(visual_rows as u16)
+                .saturating_sub(top_padding as u16)
+                .saturating_sub(1) // at least one row for the cursor
+                .max(1) as usize;
+            let menu_total = menu_entries.len();
+            let menu_top = if menu_total <= menu_visible_rows {
+                0
+            } else {
+                // Keep the selection centered in the viewport.
+                let half = menu_visible_rows / 2;
+                if selected <= half {
+                    0
+                } else if selected + half >= menu_total {
+                    menu_total.saturating_sub(menu_visible_rows)
+                } else {
+                    selected.saturating_sub(half)
+                }
+            };
+            let menu_bottom = (menu_top + menu_visible_rows).min(menu_total);
+
+            for idx in menu_top..menu_bottom {
+                let entry = &menu_entries[idx];
                 let is_selected = idx == selected;
                 let style = if is_selected {
                     Style::default()