chore(release): bump to v0.8.27, add CHANGELOG

2026-05-10 08:41:04 -05:00
parent cb084d1564
commit 6c25a18b42
30 changed files with 895 additions and 140 deletions
@@ -0,0 +1,642 @@
+# v0.8.27 — User-Issue Strategy Handoff
+
+**Audience:** the AI agent picking up post-v0.8.26 user-bug work.
+**Scope:** the issues filed by users in the 24–48 hours after v0.8.26
+shipped, plus older issues with concrete fix shapes that didn't make
+v0.8.26.
+**This is layered on top of the in-flight v0.8.27 cycle** — there are
+already 16 community-PR commits on `work/v0.8.27`. Don't start over;
+add to it.
+
+---
+
+## Where you are
+
+- **Working tree:** `/Volumes/VIXinSSD/whalebro/deepseek-tui`
+- **Active branch:** `work/v0.8.27` (off main at v0.8.26 tip)
+- **Already on the branch:** 16 commits — community PRs (#1316, #1317,
+  #1181, #1203, #1140, #1247, #1223, #1185, #1220, #1233, #1235,
+  #1197, plus a trackpad scroll fix and a card-rail UI tweak)
+- **Reference docs:** `.claude/HANDOFF_v0.8.26_security.md` for the
+  release flow steps 7–11 (same shape applies for v0.8.27)
+- **Issue board:** GitHub `Hmbown/DeepSeek-TUI`
+
+The previous agent only did community-PR cherry-picks. The strategic
+bug-fix work in this document is **not started**. Assume zero
+overlap.
+
+---
+
+## Hard rules
+
+1. **STOP and ask Hunter** before merging the v0.8.27 PR, tagging,
+   or publishing to crates.io / npm / Homebrew.
+2. No `--no-verify`, no `--no-gpg-sign`, no force push.
+3. Don't leak `.private/` content into PRs / CHANGELOG / release notes.
+4. v0.8.27 is **NOT** a security release. If a new GHSA arrives mid-
+   cycle, branch v0.8.28 — don't bundle.
+5. Time-box thorny items at 30 min. Defer to v0.8.28 instead of
+   sinking the cycle.
+
+---
+
+## P0 — ship these, they fix real user pain
+
+### 1. Cross-terminal flicker (#1119, #1352, #1356, #1363, #1366, #1260, #1295)
+
+**The most-reported bug since v0.8.26 shipped.** Five Ghostty / VSCode-
+terminal reports in 24 hours plus the existing Windows ones. **Same
+root cause, single fix.**
+
+**Diagnosis.** v0.8.22 added a viewport-reset escape sequence to fix
+viewport drift after focus/resize. The sequence is:
+
+```
+\x1b[r       set scroll region to entire screen
+\x1b[?6l     reset DECOM origin mode
+\x1b[H       cursor home
+\x1b[2J      erase entire screen
+\x1b[3J      erase saved lines
+```
+
+This fires on every redraw. `\x1b[2J\x1b[3J` is destructive — full
+clear. Terminals that don't optimize differential redraws (Ghostty,
+VSCode terminal in some configurations, Win10 conhost) blank-then-
+repaint every frame, producing visible flicker.
+
+The smoking-gun datapoint is **#1356**: "doesn't flicker on M4 Air,
+doesn't flicker for Claude Code / Codex / Gemini CLIs in the same
+VSCode terminal." Other CLIs use the alt-screen buffer's natural
+double-buffering and don't emit a destructive reset every frame.
+
+**Strategy — pick #1; #2 is the fallback.**
+
+#### 1.A — Replace destructive reset with lighter sequence (~30 min)
+
+In `crates/tui/src/tui/ui.rs` (search for the const that holds the
+reset sequence — likely named `VIEWPORT_RESET` or similar, check the
+v0.8.22 / v0.8.24 commits that mention `recover_terminal_modes` or
+`FocusGained`):
+
+```rust
+// before
+const VIEWPORT_RESET: &str = "\x1b[r\x1b[?6l\x1b[H\x1b[2J\x1b[3J";
+
+// after — drop the destructive 2J/3J; alt-screen buffer's existing
+// double-buffering handles the redraw without screen blanking.
+const VIEWPORT_RESET: &str = "\x1b[r\x1b[?6l\x1b[H";
+```
+
+Add a regression test that asserts the constant doesn't contain
+`2J` or `3J` (the destructive parts). The viewport-drift fix that
+the original sequence was added to address came from #1041 / similar
+— verify it's still working with the lighter sequence by manual
+smoke on macOS Terminal.app.
+
+#### 1.B — Audit redraw-rate and only emit on actual drift (~1 day)
+
+If 1.A reintroduces drift, fall back to this: track previous viewport
+state and only emit the reset when drift is detected (post-resize,
+post-focus-gain, post-pager-close). Search for where the constant is
+emitted; if it's in the per-frame draw path, that's the bug.
+
+#### 1.C — Per-terminal opt-out as belt-and-suspenders
+
+Detect `TERM_PROGRAM=ghostty`, `TERM_PROGRAM=vscode` (also covers
+VSCode terminal), and known-flaky `TERM` values. Skip the reset
+entirely on those. Document this as a fallback in the commit message;
+prefer 1.A as the primary fix.
+
+**Action:**
+1. File a tracking issue "Cross-terminal flicker survey" linking all
+   seven reports (#1119, #1352, #1356, #1363, #1366, #1260, #1295).
+2. Apply fix 1.A.
+3. Manual smoke on macOS Terminal.app + iTerm2 + Ghostty (Hunter has
+   Ghostty available; ask him).
+4. Comment on each linked issue: "Fixed in v0.8.27 — please update
+   and reopen if you still see flicker."
+
+---
+
+### 2. Long-text wrap (#1344, #1351, possibly #1359)
+
+**Diagnosis.** v0.8.25 fixed long markdown **table cells** (`wrap_cell_text`
+helper in `markdown_render.rs`). Long **paragraphs** and long **input
+lines** still clip at viewport width on some terminals, instead of
+wrapping. #1344 reports both directions; #1351 reports the same
+symptom plus a separate "table content shows `...`" issue. #1359 is
+"VSCode terminal won't wrap" — possibly the same root cause if VSCode
+reports terminal size differently.
+
+**Strategy.**
+
+1. Reproduce at narrow width: `COLUMNS=60 deepseek` → paste a 200-
+   character input line, ask for a 200-character paragraph response.
+   Confirm both clip rather than wrap.
+2. Trace the wrap paths:
+   - `crates/tui/src/tui/markdown_render.rs::render_message` →
+     `render_line_with_links` → `wrap_text` (paragraphs)
+   - `crates/tui/src/tui/composer.rs` (or wherever the input box
+     renders) — likely has its own wrap logic that diverged
+3. Unify on `wrap_text` from `markdown_render`. The composer should
+   use the same width-aware wrapper as the transcript.
+4. Add snapshot tests for both surfaces at widths 40, 60, 80, 120.
+5. For VSCode-terminal-specific size detection issues (#1359), verify
+   `crossterm::terminal::size()` returns the right value when run
+   inside VSCode terminal. If wrong, look at `--columns` override.
+
+**Cost:** 3-4 hours including tests.
+
+---
+
+### 3. Pager copy-out (#1354)
+
+**Diagnosis.** When users hit `Alt+V` (tool details) or `Ctrl+O`
+(thinking content), they get a pager view. The pager intercepts mouse
+capture, so terminal-native selection is disabled inside it. There's
+no in-app copy keybinding. Result: users can see the content but
+can't copy it. High-frustration UX gap — pager users are usually
+specifically there to copy something out.
+
+**Strategy.** Add a `c` (or `y`, vi-style) keybinding inside the
+pager view that copies the entire visible content to clipboard, with
+a status confirmation toast.
+
+In `crates/tui/src/tui/views/pager.rs` (or wherever `PagerView` is
+defined — search for `impl ModalView for PagerView`):
+
+```rust
+// inside handle_key, somewhere with the existing Esc/q/PgUp/PgDn handlers
+KeyCode::Char('c') | KeyCode::Char('y') => {
+    let text = self.body_text(); // whatever method gives the full body
+    if app.clipboard.write_text(&text).is_ok() {
+        app.status_message = Some("Pager content copied".to_string());
+    } else {
+        app.status_message = Some("Copy failed".to_string());
+    }
+    return Vec::new();
+}
+```
+
+Also surface the keybinding in the pager footer: append `[c copy]`
+to the existing affordance line.
+
+Add a regression test that constructs a pager, sends `c`, and
+asserts the clipboard mock saw the body text.
+
+**Cost:** ~45 minutes.
+
+---
+
+## P1 — should ship; clear shape, real impact
+
+### 4. Ctrl+C context-sensitive (#1337, #1367)
+
+**Diagnosis.** Two related issues:
+- **#1337:** Windows users expect `Ctrl+C` to copy (legacy Windows
+  convention). Our binding is exit. They lose work copying.
+- **#1367:** Users don't know how to interrupt a long-running task.
+  `Esc` works but isn't discoverable.
+
+**Strategy — context-sensitive Ctrl+C (resolves both):**
+
+Three branches based on app state:
+
+| State | Ctrl+C behavior |
+|---|---|
+| **Selection active** | Copy + clear selection. No exit. |
+| **Turn in progress** | Interrupt the turn (same as Esc). No exit. |
+| **Idle, no selection** | First press: status hint "Press Ctrl+C again to exit". Second press within 2s: exit. |
+
+This pattern is well-precedented (htop, less, tmux) and addresses
+both issues in one change. Mirror Vim's "are you sure" pattern for
+the idle case.
+
+In `crates/tui/src/tui/ui.rs::handle_key_event`, find the
+`KeyCode::Char('c')` + `KeyModifiers::CONTROL` arm:
+
+```rust
+KeyCode::Char('c') if m.contains(KeyModifiers::CONTROL) => {
+    // Branch 1: selection active → copy
+    if app.viewport.transcript_selection.is_active() {
+        copy_active_selection(app);
+        app.viewport.transcript_selection.clear();
+        return Vec::new();
+    }
+    // Branch 2: turn in progress → interrupt
+    if app.is_loading {
+        // existing interrupt logic — same code path as Esc
+        return interrupt_current_turn(app);
+    }
+    // Branch 3: idle → first press shows hint, second press within 2s exits
+    let now = Instant::now();
+    let recent_ctrl_c = app.last_ctrl_c.is_some_and(|t| now.duration_since(t) < Duration::from_secs(2));
+    if recent_ctrl_c {
+        return vec![ViewEvent::Exit];
+    }
+    app.last_ctrl_c = Some(now);
+    app.status_message = Some("Press Ctrl+C again to exit".to_string());
+    Vec::new()
+}
+```
+
+Plus the discoverability hint for #1367: status bar during streaming
+shows `[Esc cancel · Ctrl+C twice exit]`.
+
+**Cost:** ~2 hours including tests for each branch.
+
+---
+
+### 5. `notify` tool (#1322)
+
+**Diagnosis.** Model-triggerable desktop notifications. Long agent
+runs would benefit from an "I'm done, look at me" pop-up. Other
+tools (Claude Code) have this.
+
+**Strategy.** Add a built-in `notify` tool spec.
+
+1. Add `notify-rust` to `crates/tui/Cargo.toml` (already cross-
+   platform: macOS Notification Center, Linux libnotify, Windows toast).
+2. New tool in `crates/tui/src/tools/notify.rs`:
+   ```rust
+   pub struct NotifyTool;
+   
+   #[async_trait]
+   impl ToolSpec for NotifyTool {
+       fn name(&self) -> &'static str { "notify" }
+       fn description(&self) -> &'static str {
+           "Display a desktop notification to the user. Use sparingly — only when a long-running task completes or needs the user's attention."
+       }
+       fn input_schema(&self) -> Value { /* {title: required, body: optional} */ }
+       fn capabilities(&self) -> Vec<ToolCapability> {
+           vec![ToolCapability::RequiresApproval]
+       }
+       fn approval_requirement(&self) -> ApprovalRequirement {
+           ApprovalRequirement::Auto  // notifications are low-risk
+       }
+       async fn execute(&self, input: Value, _ctx: &ToolContext) -> Result<ToolResult, ToolError> {
+           // truncate title to ~60 chars, body to ~200
+           // skip if app is currently focused (don't notify about
+           // the thing the user is watching) — read from
+           // app.focus_state if available
+           // call notify_rust::Notification::new()...
+       }
+   }
+   ```
+3. Wire up in `tool_setup.rs` (probably register conditional on a
+   `Feature::DesktopNotifications` feature flag, default-on).
+4. Add config opt-out: `[tools.notify] enabled = false`.
+
+Auto-suppress when terminal is focused — the user is watching, no
+notification needed.
+
+**Cost:** ~3-4 hours.
+
+---
+
+### 6. `/skills --remote` diagnostic (#1329)
+
+**Diagnosis.** "Failed to fetch" with no details. Could be TLS,
+network policy, auth, rate limit. Bare error → undiagnosable.
+
+**Strategy.** First fix is observability — surface the underlying
+error chain.
+
+In `crates/tui/src/commands/skills.rs` (or wherever `--remote` is
+handled), find the `.unwrap_err()` or `.context(...)` that's
+collapsing the chain:
+
+```rust
+// before
+return Err(anyhow!("Failed to fetch"));
+
+// after
+return Err(err.context("Failed to fetch remote skills"));
+// or, when surfacing to the user:
+return CommandResult::error(format!("Failed to fetch remote skills:\n{err:#}"));
+```
+
+Mirror the v0.8.23 #1244 fix shape (alternate `{err:#}` formatting
+for the full anyhow chain).
+
+Once the underlying error is visible, the actual bug becomes
+diagnosable. Likely either a TLS issue (rustls vs system trust store)
+or the network policy blocking the registry endpoint.
+
+**Cost:** ~30 minutes for the diagnostic improvement.
+
+---
+
+### 7. MCP lazy reload on config change (#1267 part 2)
+
+**Diagnosis.** v0.8.26 fixed the diagnostic side (stderr capture).
+The "auto-reload after config edit" piece is still missing — users
+have to manually run `/mcp reload` after editing `~/.deepseek/config.toml`.
+
+**Strategy — lazy hash check (no file watcher).** File watchers add
+long-lived tasks and have edge cases on remote / network filesystems.
+A lazy hash compare is bounded and cheap.
+
+In `crates/tui/src/mcp.rs::McpPool`:
+
+```rust
+pub struct McpPool {
+    // ... existing fields
+    config_hash: u64,  // hash of mcp config at last (re)connection
+}
+
+impl McpPool {
+    fn current_config_hash(&self, config: &McpConfig) -> u64 {
+        let mut hasher = std::hash::DefaultHasher::new();
+        // hash the relevant fields: servers map, timeouts, sandbox_mode
+        config.hash(&mut hasher);
+        hasher.finish()
+    }
+
+    pub async fn get_or_connect(&mut self, server: &str, config: &McpConfig) -> Result<&mut McpConnection> {
+        let new_hash = self.current_config_hash(config);
+        if new_hash != self.config_hash {
+            self.reload_all(config).await?;
+            self.config_hash = new_hash;
+        }
+        // existing get_or_connect logic
+    }
+}
+```
+
+`McpConfig` and adjacent types may need `Hash` derived. If hashing
+the whole config tree is expensive, hash just the `[mcp_servers]`
+section + `sandbox_mode`.
+
+**Cost:** ~2 hours including tests.
+
+---
+
+## P2 — nice-to-have if time permits
+
+### 8. Layout overlap (#1357)
+
+**Diagnosis.** Input box and inline runtime hint ("Cache: 99% hit |
+hit X | miss Y") render in adjacent rects but one isn't clearing its
+area properly when the other expands.
+
+**Strategy.** Inspect `crates/tui/src/tui/ui.rs::render` — find the
+composer's reserved-rows calculation. It probably doesn't account for
+the hint line on resize / long-content. Fix the rect math.
+
+**Cost:** ~2 hours (1 to repro, 1 to fix).
+
+### 9. `/skills` filter argument (#1318)
+
+**Diagnosis.** v0.8.26 added inter-row spacing (#1328 from @reidliu41).
+Reporter may want more.
+
+**Strategy.**
+1. Comment on #1318 asking if v0.8.26's spacing is enough.
+2. If not, add `/skills <prefix>` arg → filter to skills whose names
+   start with `<prefix>`. Mirror how `/help <topic>` works.
+
+**Cost:** Triage ping; 30 min if filter wanted.
+
+### 10. Status comments on partial fixes (#1112, #1267, #1318)
+
+Three issues that are partly addressed and need the reporter to
+confirm:
+
+- **#1112** — 1.2 TB snapshots. Cap added in v0.8.24. Comment:
+  "500 MB cap added in v0.8.24. Are you still seeing growth above
+  that? If so, please share `du -sh ~/.deepseek/snapshots`."
+- **#1267** — macOS Seatbelt blocks npx MCP. Already commented during
+  v0.8.26 cycle. Don't re-comment.
+- **#1318** — `/skills` crowded. Comment: "v0.8.26 added inter-row
+  spacing (#1328). Does this resolve it for you?"
+
+**Cost:** ~5 minutes total.
+
+---
+
+## P3 — investigate or defer
+
+### #1338 — Enter mid-run crashes Windows TUI
+
+**Defer unless you have Windows.** Add stack capture so the next
+reporter gets actionable output:
+
+```rust
+// in main.rs — add panic hook that logs to ~/.deepseek/last-panic.log
+std::panic::set_hook(Box::new(|info| {
+    let _ = std::fs::write(
+        dirs::home_dir().unwrap_or_default().join(".deepseek/last-panic.log"),
+        format!("{info}\n{}", std::backtrace::Backtrace::capture()),
+    );
+}));
+```
+
+**Cost:** 30 min for the diagnostic; actual fix needs Windows VM.
+
+### #1062 — Capacity-memory checkpoint cross-session recovery
+
+Old, complex. Don't pull into v0.8.27. Needs scope conversation with
+Hunter.
+
+### #1067 — glibc version required (older Linux distros)
+
+Static-link the deepseek binary or add a musl build to release.yml.
+**v0.8.27 candidate if anyone has time** — purely a build-config
+change.
+
+### #1364 — Hooks mutation rights + turn-end event
+
+**Defer to v0.9.0.** Real ask — Claude Code hooks have this. Worth
+doing as part of a hooks-v2 task. Out of scope for a polish release.
+
+### #1343 — Desktop GUI
+
+**Defer.** Recurring request. v0.9.x territory at the earliest.
+Comment with roadmap status if not already.
+
+---
+
+## Issues to close as fixed in v0.8.26
+
+These need a comment + close. Already verified by the previous agent:
+
+| # | Title | Fixed by |
+|---|---|---|
+| #1163 | Mouse drag-select / copy doesn't auto-scroll | PR #1239 |
+| #1169 | Selection crosses sidebar | Mouse-capture default-on for WT |
+| #1255 | Win10 conversation can't scroll | Mouse-capture default-on |
+| #1292 | Mac trackpad text selection broken | Drag-select rewrite |
+| #1298 | Wheel scrolls input history not transcript | Mouse-capture default-on |
+| #1308 | base_url for ollama/vllm ignored | Config-load warning |
+| #1331 | Mouse wheel changed in v0.8.24 | Mouse-capture default-on |
+
+**Action:** Run through with this comment template (translated for
+zh-CN issues #1255, #1292):
+
+```
+Fixed in [v0.8.26](https://github.com/Hmbown/DeepSeek-TUI/releases/tag/v0.8.26).
+Please update with:
+
+- npm: `npm install -g deepseek-tui@latest`
+- brew: `brew upgrade deepseek-tui`
+- cargo: `cargo install --force deepseek-tui-cli`
+
+Reopen if you still hit it. Thanks for the report!
+```
+
+---
+
+## Workflow
+
+### Step 1 — Branch state confirmation
+
+```bash
+cd /Volumes/VIXinSSD/whalebro/deepseek-tui
+git checkout work/v0.8.27
+git pull origin work/v0.8.27 || true
+git log --oneline main..HEAD | head -20
+```
+
+You should see ~16 commits already on the branch. Add to it; don't
+restart.
+
+### Step 2 — Tackle in priority order (P0 → P3)
+
+For each item:
+
+1. Read the issue thread on GitHub. Note any reporter clarifications.
+2. Implement per the strategy above.
+3. Add tests (TDD where the strategy specifies; verification snapshot
+   otherwise).
+4. After each commit:
+   ```bash
+   cargo fmt --all
+   cargo clippy -p deepseek-tui --all-targets --all-features --locked -- -D warnings
+   cargo test -p deepseek-tui --bin deepseek-tui --all-features --locked --no-fail-fast \
+     2>&1 | grep "test result:" | tail -3
+   ```
+5. Add a CHANGELOG entry under `## [0.8.27]` `### Fixed` or `### Added`,
+   crediting the issue number and original reporter.
+
+The known-flaky test is
+`mcp_connection_supports_streamable_http_event_stream_responses` —
+passes in isolation, intermittent under load. Don't chase.
+
+### Step 3 — Issue triage pass
+
+After each P0/P1 fix lands, close the corresponding issue with a
+comment template. Don't wait until the end of the cycle — closing as
+you go keeps the issue list visibly responsive.
+
+### Step 4 — Bump version when ready
+
+```bash
+sed -i '' 's|^version = "0.8.26"|version = "0.8.27"|' Cargo.toml
+find crates -maxdepth 2 -name Cargo.toml -exec sed -i '' \
+  's|version = "0.8.26"|version = "0.8.27"|g' {} +
+sed -i '' 's|"version": "0.8.26"|"version": "0.8.27"|' \
+  npm/deepseek-tui/package.json
+sed -i '' 's|"deepseekBinaryVersion": "0.8.26"|"deepseekBinaryVersion": "0.8.27"|' \
+  npm/deepseek-tui/package.json
+cargo update --workspace --offline
+./scripts/release/check-versions.sh
+```
+
+Add `## [0.8.27] - YYYY-MM-DD` heading at the top of CHANGELOG.md.
+
+### Step 5 — Full preflight + install
+
+```bash
+cargo fmt --all -- --check
+cargo clippy --workspace --all-targets --all-features --locked -- -D warnings
+cargo test --workspace --all-features --locked --no-fail-fast \
+  2>&1 | grep "test result:" | tail -10
+./scripts/release/check-versions.sh
+./scripts/release/publish-crates.sh dry-run
+cargo build --release --locked -p deepseek-tui-cli -p deepseek-tui
+node scripts/release/npm-wrapper-smoke.js
+
+cargo install --path crates/cli --force --locked
+cargo install --path crates/tui --force --locked
+deepseek --version  # confirm: deepseek 0.8.27 (<sha>)
+```
+
+### Step 6 — STOP-FOR-MAINTAINER
+
+Push the branch and open the release PR. Hand back to Hunter:
+
+- PR number + link
+- Bullet list of all P0/P1/P2 items completed
+- Items deferred (P3 items) with reason
+- Preflight summary
+- "deepseek 0.8.27 installed at ~/.cargo/bin/, ready for testing"
+- Issues closed with v0.8.26 fixed-in comment
+
+WAIT for Hunter's "go" before merging, tagging, or publishing.
+
+### Step 7 — Release flow
+
+Same as v0.8.26 — see `.claude/HANDOFF_v0.8.26_security.md` steps 8–11.
+Concretely: merge PR → auto-tag fires → release.yml builds matrix +
+GitHub Release → crates.io publish → npm publish → Homebrew formula
+update → verify GHCR → README post-merge bookkeeping.
+
+**No GHSA flow this cycle.** If a new advisory comes in, branch
+v0.8.28 — don't bundle.
+
+### Step 8 — CNB mirror (new for v0.8.27)
+
+After GitHub Release is live:
+
+```bash
+# If CNB_TOKEN is in repo secrets, the GitHub Action handles it
+# automatically on tag push. Verify:
+#   https://cnb.cool/deepseek-tui.com/DeepSeek-TUI/-/tags
+
+# Otherwise (one-time bring-up was done manually) push from local:
+git remote add cnb https://<token>@cnb.cool/deepseek-tui.com/DeepSeek-TUI 2>/dev/null || true
+git push cnb v0.8.27 main
+```
+
+Add a banner to README.md and README.zh-CN.md if not already there:
+
+```
+> 🇨🇳 国内镜像 / Mainland China mirror:
+>   https://cnb.cool/deepseek-tui.com/DeepSeek-TUI
+> Issues and PRs: please use GitHub.
+```
+
+---
+
+## Quality bar
+
+Apply to every change:
+
+- CI green (modulo documented flaky)
+- No new `unwrap()` / `expect()` outside test code
+- No new external network surfaces without `validate_network_policy`
+- New env vars or config keys → `config.example.toml` entry + CHANGELOG note
+- Behavior changes user-visible → CHANGELOG entry calling out the change
+
+When in doubt, defer to v0.8.28. A clean release of 8 P0/P1 items beats
+a cluttered release of 15 with one regression.
+
+---
+
+## Output expectation
+
+Realistic v0.8.27 landing zone on top of the existing 16 commits:
+
+- **All 7 closable v0.8.26 issues** closed with comments
+- **P0 #1, #2, #3** fully shipped (flicker, wrap, pager copy)
+- **P1 #4, #5, #6, #7** at least 2 of 4 shipped
+- **P2 #8, #9, #10** at least the comment-pings
+- **CNB mirror** wired in
+
+That's a substantial v0.8.27 that respects the "post-v0.8.26 inflow"
+framing. Users see real responsiveness to their reports.
+
+If at any point something looks materially harder than this document
+suggests, STOP and surface to Hunter with the specifics. Don't
+freelance scope.
@@ -5,6 +5,96 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).

+## [0.8.27] - 2026-05-10
+
+A substantial polish release bundling 17 community PRs and a small
+internal fix. Big thanks to every contributor below.
+
+### Added
+
+- **Unified `/mode` command** (#1247) — `/mode [agent|plan|yolo|1|2|3]`
+  replaces the separate `/agent`, `/plan`, and `/yolo` commands. Running
+  `/mode` without arguments opens a picker modal. The legacy aliases
+  (`/yolo`, `/agent`, `/plan`) are kept as compatibility shorthands.
+  Thanks **@reidliu41**.
+- **`/status` runtime diagnostics** (#1223) — shows version, provider,
+  model, workspace, mode, permissions, context-window usage, cache
+  hit/miss, and session cost. Previously `/status` was an alias for
+  `/statusline` (footer config); that alias is now `/statusline` only.
+  Thanks **@reidliu41**.
+- **`/feedback` command** (#1185) — opens the matching GitHub issue
+  template (bug report, feature request) in the browser. Security
+  vulnerability reports route through the project's security policy
+  page first. Thanks **@reidliu41**.
+- **Session artifact metadata** (#1220) — large tool outputs spilled to
+  the session artifacts directory are now tracked in a durable metadata
+  index, so saved sessions retain references across save/restore cycles.
+  Thanks **@THINKER-ONLY**.
+- **Subagent results are self-reports** (#1140) — the compacted result
+  summary now notes that child-agent outputs are unverified self-reports.
+  The parent model should verify side effects with tools like `read_file`
+  or `list_dir` before claiming success. Thanks **@THINKER-ONLY**.
+- **Global AGENTS.md fallback** (#1197) — when the workspace and its
+  parents don't provide project instructions, the TUI now loads
+  `~/.deepseek/AGENTS.md` before falling back to auto-generated
+  instructions. Repo-local context still takes priority.
+  Thanks **@manaskarra**.
+- **`--yolo` forwarded from CLI to TUI** (#1233) — the `deepseek --yolo`
+  flag now propagates through the dispatcher to the TUI binary via
+  `DEEPSEEK_YOLO=true`. Previously the flag set `yolo` in the CLI
+  process but the TUI session started in its default mode.
+  Thanks **@fuleinist**.
+- **`composer_arrows_scroll` config** (#1211) — a new
+  `tui.composer_arrows_scroll` option (default `false`) makes plain
+  Up/Down arrow keys scroll the transcript when the composer is empty,
+  instead of navigating input history. Helpful for terminals that map
+  trackpad gestures to arrow keys. Thanks **@lbcheng888**.
+- **Session cost persistence** (#1192) — accumulated costs (session +
+  sub-agents, both USD and CNY) and the displayed-cost high-water mark
+  now survive session save/restore, so the monotonic cost guarantee
+  (#244) holds across restarts. Thanks **@lbcheng888**.
+- **Provider-aware model picker and provider persistence** (#1320) —
+  switching providers now persists the choice to
+  `~/.deepseek/settings.toml` so it survives restarts. The model
+  picker hides DeepSeek-specific models when a non-DeepSeek provider
+  is active. `OPENAI_MODEL` env var now overrides the per-provider
+  model rather than the global `default_text_model`. Bailian / ZhiPu
+  Coding Plan endpoints are now supported.
+  Thanks **@imkingjh999**.
+- **HTTP User-Agent header** (#1320) — all outbound API requests now
+  carry `deepseek-tui/{version}` in the User-Agent, matching the format
+  `fetch_url` already uses. Thanks **@imkingjh999**.
+
+### Fixed
+
+- **HTTP 400 quota errors retried** (#1203) — some OpenAI-compatible
+  gateways return quota/rate-limit errors as HTTP 400 instead of 429.
+  These are now classified as retryable `RateLimited` errors.
+  Thanks **@dst1213**.
+- **Explicit hidden/ignored file completions** (#1270) — when the user
+  types an explicit path starting with `.` (e.g., `.deepseek/commands/`),
+  the file-completion system now surfaces hidden and gitignored entries
+  while still respecting `.deepseekignore`. Thanks **@SamhandsomeLee**.
+
+### Changed
+
+- **Windows mouse capture docs** (#1181) — the `--mouse-capture` help
+  text and the configuration docs now mention scrollbar dragging and
+  note that raw terminal selection on Windows may cross the sidebar.
+  Thanks **@Oliver-ZPLiu**.
+- **README zh-CN sync** (#1235) — the Chinese README's quickstart section
+  now shows `deepseek run pr <N>` instead of the outdated
+  `deepseek pr <N>`. Thanks **@whtis**.
+- **Tool output render perf** (#1098) — tool output summaries and the
+  "is this a diff?" check are now pre-computed once at cell creation
+  instead of re-parsed every frame. Tool output cells also got a visual
+  card-rail (`╭ │ ╰`) for clearer grouping. Thanks **@lbcheng888**.
+
+### Internal
+
+- Test coverage for approval decision branches (@tuohai666, #1316)
+- Test coverage for hook event dispatch paths (@tuohai666, #1317)
+
 ## [0.8.26] - 2026-05-09

 A security + polish release. Two responsibly-disclosed issues were
@@ -19,7 +19,7 @@ default-members = ["crates/cli", "crates/app-server", "crates/tui"]
 resolver = "2"

 [workspace.package]
-version = "0.8.26"
+version = "0.8.27"
 edition = "2024"
 # Rust 1.88 stabilized `let_chains` in `if`/`while` conditions, which the
 # codebase relies on extensively. Cargo enforces this so users on older
@@ -7,5 +7,5 @@ repository.workspace = true
 description = "Model/provider registry and fallback strategy for DeepSeek workspace architecture"

 [dependencies]
-deepseek-config = { path = "../config", version = "0.8.26" }
+deepseek-config = { path = "../config", version = "0.8.27" }
 serde.workspace = true
@@ -10,15 +10,15 @@ description = "Codex-style app-server transport for DeepSeek workspace architect
 anyhow.workspace = true
 axum.workspace = true
 clap.workspace = true
-deepseek-agent = { path = "../agent", version = "0.8.26" }
-deepseek-config = { path = "../config", version = "0.8.26" }
-deepseek-core = { path = "../core", version = "0.8.26" }
-deepseek-execpolicy = { path = "../execpolicy", version = "0.8.26" }
-deepseek-hooks = { path = "../hooks", version = "0.8.26" }
-deepseek-mcp = { path = "../mcp", version = "0.8.26" }
-deepseek-protocol = { path = "../protocol", version = "0.8.26" }
-deepseek-state = { path = "../state", version = "0.8.26" }
-deepseek-tools = { path = "../tools", version = "0.8.26" }
+deepseek-agent = { path = "../agent", version = "0.8.27" }
+deepseek-config = { path = "../config", version = "0.8.27" }
+deepseek-core = { path = "../core", version = "0.8.27" }
+deepseek-execpolicy = { path = "../execpolicy", version = "0.8.27" }
+deepseek-hooks = { path = "../hooks", version = "0.8.27" }
+deepseek-mcp = { path = "../mcp", version = "0.8.27" }
+deepseek-protocol = { path = "../protocol", version = "0.8.27" }
+deepseek-state = { path = "../state", version = "0.8.27" }
+deepseek-tools = { path = "../tools", version = "0.8.27" }
 serde.workspace = true
 serde_json.workspace = true
 tokio.workspace = true
@@ -14,13 +14,13 @@ path = "src/main.rs"
 anyhow.workspace = true
 clap.workspace = true
 clap_complete.workspace = true
-deepseek-agent = { path = "../agent", version = "0.8.26" }
-deepseek-app-server = { path = "../app-server", version = "0.8.26" }
-deepseek-config = { path = "../config", version = "0.8.26" }
-deepseek-execpolicy = { path = "../execpolicy", version = "0.8.26" }
-deepseek-mcp = { path = "../mcp", version = "0.8.26" }
-deepseek-secrets = { path = "../secrets", version = "0.8.26" }
-deepseek-state = { path = "../state", version = "0.8.26" }
+deepseek-agent = { path = "../agent", version = "0.8.27" }
+deepseek-app-server = { path = "../app-server", version = "0.8.27" }
+deepseek-config = { path = "../config", version = "0.8.27" }
+deepseek-execpolicy = { path = "../execpolicy", version = "0.8.27" }
+deepseek-mcp = { path = "../mcp", version = "0.8.27" }
+deepseek-secrets = { path = "../secrets", version = "0.8.27" }
+deepseek-state = { path = "../state", version = "0.8.27" }
 chrono.workspace = true
 dirs.workspace = true
 serde.workspace = true
@@ -8,7 +8,7 @@ description = "Config schema and precedence model for DeepSeek workspace archite

 [dependencies]
 anyhow.workspace = true
-deepseek-secrets = { path = "../secrets", version = "0.8.26" }
+deepseek-secrets = { path = "../secrets", version = "0.8.27" }
 dirs.workspace = true
 serde.workspace = true
 toml.workspace = true
@@ -962,9 +962,7 @@ impl ConfigToml {
            .clone()
            .or_else(|| env.sandbox_mode.clone())
            .or_else(|| self.sandbox_mode.clone());
-        let yolo = cli
-            .yolo
-            .or(env.yolo);
+        let yolo = cli.yolo.or(env.yolo);

        ResolvedRuntimeOptions {
            provider,
@@ -9,13 +9,13 @@ description = "Core runtime boundaries for DeepSeek workspace architecture"
 [dependencies]
 anyhow.workspace = true
 chrono.workspace = true
-deepseek-agent = { path = "../agent", version = "0.8.26" }
-deepseek-config = { path = "../config", version = "0.8.26" }
-deepseek-execpolicy = { path = "../execpolicy", version = "0.8.26" }
-deepseek-hooks = { path = "../hooks", version = "0.8.26" }
-deepseek-mcp = { path = "../mcp", version = "0.8.26" }
-deepseek-protocol = { path = "../protocol", version = "0.8.26" }
-deepseek-state = { path = "../state", version = "0.8.26" }
-deepseek-tools = { path = "../tools", version = "0.8.26" }
+deepseek-agent = { path = "../agent", version = "0.8.27" }
+deepseek-config = { path = "../config", version = "0.8.27" }
+deepseek-execpolicy = { path = "../execpolicy", version = "0.8.27" }
+deepseek-hooks = { path = "../hooks", version = "0.8.27" }
+deepseek-mcp = { path = "../mcp", version = "0.8.27" }
+deepseek-protocol = { path = "../protocol", version = "0.8.27" }
+deepseek-state = { path = "../state", version = "0.8.27" }
+deepseek-tools = { path = "../tools", version = "0.8.27" }
 serde_json.workspace = true
 uuid.workspace = true
@@ -8,5 +8,5 @@ description = "Execution policy and approval model parity for DeepSeek workspace

 [dependencies]
 anyhow.workspace = true
-deepseek-protocol = { path = "../protocol", version = "0.8.26" }
+deepseek-protocol = { path = "../protocol", version = "0.8.27" }
 serde.workspace = true
@@ -10,7 +10,7 @@ description = "Hook dispatch and notifications parity for DeepSeek workspace arc
 anyhow.workspace = true
 async-trait.workspace = true
 chrono.workspace = true
-deepseek-protocol = { path = "../protocol", version = "0.8.26" }
+deepseek-protocol = { path = "../protocol", version = "0.8.27" }
 reqwest.workspace = true
 serde.workspace = true
 serde_json.workspace = true
@@ -9,7 +9,7 @@ description = "Tool invocation lifecycle, schema validation, and scheduler paral
 [dependencies]
 anyhow.workspace = true
 async-trait.workspace = true
-deepseek-protocol = { path = "../protocol", version = "0.8.26" }
+deepseek-protocol = { path = "../protocol", version = "0.8.27" }
 serde.workspace = true
 serde_json.workspace = true
 tokio.workspace = true
@@ -21,8 +21,8 @@ path = "src/main.rs"
 [dependencies]
 anyhow = "1.0.100"
 arboard = "3.4"
-deepseek-secrets = { path = "../secrets", version = "0.8.26" }
-deepseek-tools = { path = "../tools", version = "0.8.26" }
+deepseek-secrets = { path = "../secrets", version = "0.8.27" }
+deepseek-tools = { path = "../tools", version = "0.8.27" }
 schemaui = { version = "0.12.0", default-features = false, optional = true }
 async-stream = "0.3.6"
 async-trait = "0.1"
@@ -514,7 +514,11 @@ impl DeepSeekClient {
        let headers = build_default_headers(api_key, extra_headers)?;
        let mut builder = reqwest::Client::builder()
            .default_headers(headers)
-            .user_agent(concat!("Mozilla/5.0 (compatible; deepseek-tui/", env!("CARGO_PKG_VERSION"), "; +https://github.com/Hmbown/DeepSeek-TUI)"))
+            .user_agent(concat!(
+                "Mozilla/5.0 (compatible; deepseek-tui/",
+                env!("CARGO_PKG_VERSION"),
+                "; +https://github.com/Hmbown/DeepSeek-TUI)"
+            ))
            .connect_timeout(Duration::from_secs(30))
            .tcp_keepalive(Some(Duration::from_secs(30)))
            .http2_keep_alive_interval(Some(Duration::from_secs(15)))
@@ -2380,7 +2380,7 @@ fn normalize_model_for_provider(provider: ApiProvider, model: &str) -> Option<St
    normalize_model_name(model).map(|normalized| model_for_provider(provider, normalized))
 }

-fn provider_passes_model_through(provider: ApiProvider) -> bool {
+pub(crate) fn provider_passes_model_through(provider: ApiProvider) -> bool {
    matches!(provider, ApiProvider::Openai | ApiProvider::Ollama)
 }

@@ -683,6 +683,7 @@ pub fn create_saved_session_with_id_and_mode(
            model: model.to_string(),
            workspace: workspace.to_path_buf(),
            mode: mode.map(str::to_string),
+            cost: SessionCostSnapshot::default(),
        },
        messages: capped_messages,
        system_prompt: merge_truncation_note(
@@ -335,7 +335,7 @@ mod tests {
            duration_ms: None,
            source: ExecSource::Assistant,
            interaction: None,
-                output_summary: None,
+            output_summary: None,
        }))
    }

@@ -356,8 +356,8 @@ mod tests {
            output: None,
            prompts: None,
            spillover_path: None,
-                output_summary: None,
-                is_diff: false,
+            output_summary: None,
+            is_diff: false,
        }))
    }

@@ -2258,11 +2258,11 @@ fn wrap_card_rail(mut lines: Vec<Line<'static>>) -> Vec<Line<'static>> {
    }
    for (i, line) in lines.iter_mut().enumerate() {
        let rail = if i == 0 {
-            "\u{256D} "  // ╭
+            "\u{256D} " // ╭
        } else if i == n - 1 {
-            "\u{2570} "  // ╰
+            "\u{2570} " // ╰
        } else {
-            "\u{2502} "  // │
+            "\u{2502} " // │
        };
        line.spans.insert(0, Span::raw(rail));
    }
@@ -3154,8 +3154,8 @@ mod tests {
            spillover_path: Some(PathBuf::from(
                "/Users/dev/.deepseek/tool_outputs/call-abc12.txt",
            )),
-                output_summary: None,
-                is_diff: false,
+            output_summary: None,
+            is_diff: false,
        };
        let lines = cell.lines_with_mode(120, true, super::RenderMode::Live);
        let joined: String = lines
@@ -3184,8 +3184,8 @@ mod tests {
            output: Some("output".to_string()),
            prompts: None,
            spillover_path: Some(PathBuf::from("/tmp/spill.txt")),
-                output_summary: None,
-                is_diff: false,
+            output_summary: None,
+            is_diff: false,
        };
        let lines = cell.lines_with_mode(120, true, super::RenderMode::Transcript);
        let joined: String = lines
@@ -3208,8 +3208,8 @@ mod tests {
            output: Some("contents".to_string()),
            prompts: None,
            spillover_path: None,
-                output_summary: None,
-                is_diff: false,
+            output_summary: None,
+            is_diff: false,
        };
        let lines = cell.lines_with_mode(80, true, super::RenderMode::Live);
        let joined: String = lines
@@ -3230,8 +3230,8 @@ mod tests {
            output: Some("output".to_string()),
            prompts: None,
            spillover_path: Some(PathBuf::from(long_path)),
-                output_summary: None,
-                is_diff: false,
+            output_summary: None,
+            is_diff: false,
        };
        let lines = cell.lines_with_mode(40, true, super::RenderMode::Live);
        let annotation_line = lines
@@ -3342,8 +3342,8 @@ mod tests {
            output: None,
            prompts: None,
            spillover_path: None,
-                output_summary: None,
-                is_diff: false,
+            output_summary: None,
+            is_diff: false,
        };
        let lines = cell.lines_with_mode(80, true, super::RenderMode::Live);
        assert_eq!(lines.len(), 1);
@@ -3364,8 +3364,8 @@ mod tests {
            ),
            prompts: None,
            spillover_path: None,
-                output_summary: None,
-                is_diff: false,
+            output_summary: None,
+            is_diff: false,
        };
        let lines = cell.lines_with_mode(80, true, super::RenderMode::Transcript);
        // Transcript mode emits header + name kv + (no args, output present)
@@ -3384,8 +3384,8 @@ mod tests {
            output: Some("first line\nsecond line\nthird line".to_string()),
            prompts: None,
            spillover_path: None,
-                output_summary: None,
-                is_diff: false,
+            output_summary: None,
+            is_diff: false,
        };
        let lines = cell.lines_with_mode(80, true, super::RenderMode::Live);
        assert!(
@@ -3640,7 +3640,7 @@ mod tests {
            duration_ms: None,
            source: ExecSource::Assistant,
            interaction: None,
-                output_summary: None,
+            output_summary: None,
        }));

        let animated = cell.lines_with_options(80, TranscriptRenderOptions::default());
@@ -3857,7 +3857,7 @@ mod tests {
            duration_ms: Some(10),
            source: ExecSource::Assistant,
            interaction: None,
-                output_summary: None,
+            output_summary: None,
        };
        let header = &cell.lines_with_motion(80, true)[0];
        let visible: String = header
@@ -3887,7 +3887,7 @@ mod tests {
            duration_ms: None,
            source: ExecSource::Assistant,
            interaction: None,
-                output_summary: None,
+            output_summary: None,
        };

        let header = &cell.lines_with_motion(80, true)[0];
@@ -3912,8 +3912,8 @@ mod tests {
            output: None,
            prompts: None,
            spillover_path: None,
-                output_summary: None,
-                is_diff: false,
+            output_summary: None,
+            is_diff: false,
        };
        let lines = cell.lines_with_mode(80, true, super::RenderMode::Live);
        let header_visible: String = lines[0]
@@ -3941,8 +3941,8 @@ mod tests {
            output: None,
            prompts: None,
            spillover_path: None,
-                output_summary: None,
-                is_diff: false,
+            output_summary: None,
+            is_diff: false,
        };
        let lines = cell.lines_with_mode(80, true, super::RenderMode::Live);
        let header_visible: String = lines[0]
@@ -4187,7 +4187,7 @@ mod tests {
            duration_ms: Some(42),
            source: ExecSource::Assistant,
            interaction: None,
-                output_summary: None,
+            output_summary: None,
        };

        let lines = cell.lines_with_motion(80, true);
@@ -4410,8 +4410,8 @@ mod tests {
                "Diff this commit against main".to_string(),
            ]),
            spillover_path: None,
-                output_summary: None,
-                is_diff: false,
+            output_summary: None,
+            is_diff: false,
        }));
        let text = lines_text(&cell.lines(80));

@@ -4437,8 +4437,8 @@ mod tests {
            output: None,
            prompts: None,
            spillover_path: None,
-                output_summary: None,
-                is_diff: false,
+            output_summary: None,
+            is_diff: false,
        }));
        let text = lines_text(&cell.lines(80));
        assert!(text.contains("query: foo"));
@@ -4462,8 +4462,8 @@ mod tests {
            output: Some(diff_stat.to_string()),
            prompts: None,
            spillover_path: None,
-                output_summary: None,
-                is_diff: false,
+            output_summary: None,
+            is_diff: false,
        }));

        let transcript_text = lines_text(&cell.transcript_lines(80));
@@ -4514,8 +4514,8 @@ mod tests {
            output: Some(output),
            prompts: None,
            spillover_path: None,
-                output_summary: None,
-                is_diff: false,
+            output_summary: None,
+            is_diff: false,
        }));

        let live = cell.lines_with_options(80, TranscriptRenderOptions::default());
@@ -4595,8 +4595,10 @@ mod tests {
            lines_text(&cell.lines_with_options(80, TranscriptRenderOptions::default()));

        // Live mode: one-line summary + expand affordance.
-        assert!(live_text.contains("Alt+V for details"),
-            "live view must show expand affordance: {live_text}");
+        assert!(
+            live_text.contains("Alt+V for details"),
+            "live view must show expand affordance: {live_text}"
+        );
        // The pre-computed summary captures the first meaningful content.
        assert!(
            live_text.contains("Error:") || live_text.contains("fatal:"),
@@ -683,13 +683,18 @@ fn render_context_panel(f: &mut Frame, area: Rect, app: &App) {
    // Only show the additive breakdown when it matches the displayed
    // total; when the high-water mark is in effect (post-reconciliation),
    // the breakdown would not sum to the displayed value (#244).
-    let cost_line = if (displayed_total - real_total).abs() < COST_EQ_TOLERANCE {
+    let cost_line = if (total_cost - real_total).abs() < COST_EQ_TOLERANCE {
        format!(
            "cost: {} (session {} + agents {})",
            app.format_cost_amount(total_cost),
            app.format_cost_amount(session_cost),
            app.format_cost_amount(agent_cost)
-        ),
+        )
+    } else {
+        format!("cost: {}", app.format_cost_amount(total_cost),)
+    };
+    lines.push(Line::from(Span::styled(
+        cost_line,
        Style::default().fg(palette::TEXT_MUTED),
    )));

@@ -624,15 +624,12 @@ pub(super) fn handle_tool_call_complete(
                match result.as_ref() {
                    Ok(tool_result) => {
                        generic.output = Some(tool_result.content.clone());
-                        generic.output_summary =
-                            Some(summarize_tool_output(&tool_result.content));
-                        generic.is_diff =
-                            output_looks_like_diff(&tool_result.content);
+                        generic.output_summary = Some(summarize_tool_output(&tool_result.content));
+                        generic.is_diff = output_looks_like_diff(&tool_result.content);
                    }
                    Err(err) => {
                        generic.output = Some(err.to_string());
-                        generic.output_summary =
-                            Some(summarize_tool_output(&err.to_string()));
+                        generic.output_summary = Some(summarize_tool_output(&err.to_string()));
                        generic.is_diff = false;
                    }
                }
@@ -715,12 +712,8 @@ fn push_orphan_tool_completion(
        .and_then(|m| m.get("spillover_path"))
        .and_then(serde_json::Value::as_str)
        .map(std::path::PathBuf::from);
-    let output_summary = output
-        .as_deref()
-        .map(|o| summarize_tool_output(o));
-    let is_diff = output
-        .as_deref()
-        .is_some_and(|o| output_looks_like_diff(o));
+    let output_summary = output.as_deref().map(|o| summarize_tool_output(o));
+    let is_diff = output.as_deref().is_some_and(|o| output_looks_like_diff(o));
    app.add_message(HistoryCell::Tool(ToolCell::Generic(GenericToolCell {
        name: name.to_string(),
        status,
@@ -564,7 +564,7 @@ mod tests {
            duration_ms: None,
            source: ExecSource::Assistant,
            interaction: None,
-                output_summary: None,
+            output_summary: None,
        }))
    }

@@ -6279,10 +6279,16 @@ fn apply_loaded_session(app: &mut App, session: &SavedSession) -> bool {
    // the restored total with no regression.
    let total_restored_usd = session.metadata.cost.total_usd();
    let total_restored_cny = session.metadata.cost.total_cny();
-    app.session.displayed_cost_high_water =
-        session.metadata.cost.displayed_cost_high_water_usd.max(total_restored_usd);
-    app.session.displayed_cost_high_water_cny =
-        session.metadata.cost.displayed_cost_high_water_cny.max(total_restored_cny);
+    app.session.displayed_cost_high_water = session
+        .metadata
+        .cost
+        .displayed_cost_high_water_usd
+        .max(total_restored_usd);
+    app.session.displayed_cost_high_water_cny = session
+        .metadata
+        .cost
+        .displayed_cost_high_water_cny
+        .max(total_restored_cny);
    app.session.last_prompt_tokens = None;
    app.session.last_completion_tokens = None;
    app.session.last_prompt_cache_hit_tokens = None;
@@ -246,8 +246,8 @@ fn selection_to_text_copies_rendered_transcript_block() {
            output: Some("tool output line".to_string()),
            prompts: None,
            spillover_path: None,
-                output_summary: None,
-                is_diff: false,
+            output_summary: None,
+            is_diff: false,
        })),
        HistoryCell::Assistant {
            content: "copy assistant".to_string(),
@@ -1197,7 +1197,7 @@ fn active_tool_status_label_summarizes_live_tool_group() {
            duration_ms: None,
            source: ExecSource::Assistant,
            interaction: None,
-                output_summary: None,
+            output_summary: None,
        })),
    );
    active.push_tool(
@@ -1209,8 +1209,8 @@ fn active_tool_status_label_summarizes_live_tool_group() {
            output: Some("done".to_string()),
            prompts: None,
            spillover_path: None,
-                output_summary: None,
-                is_diff: false,
+            output_summary: None,
+            is_diff: false,
        })),
    );
    app.active_cell = Some(active);
@@ -1237,8 +1237,8 @@ fn active_tool_status_label_counts_foreground_rlm_work() {
            output: None,
            prompts: None,
            spillover_path: None,
-                output_summary: None,
-                is_diff: false,
+            output_summary: None,
+            is_diff: false,
        })),
    );
    app.active_cell = Some(active);
@@ -2502,8 +2502,8 @@ fn jump_to_adjacent_tool_cell_finds_next_and_previous() {
            output: Some("done".to_string()),
            prompts: None,
            spillover_path: None,
-                output_summary: None,
-                is_diff: false,
+            output_summary: None,
+            is_diff: false,
        })),
        HistoryCell::Assistant {
            content: "ok".to_string(),
@@ -2516,8 +2516,8 @@ fn jump_to_adjacent_tool_cell_finds_next_and_previous() {
            output: Some("...".to_string()),
            prompts: None,
            spillover_path: None,
-                output_summary: None,
-                is_diff: false,
+            output_summary: None,
+            is_diff: false,
        })),
    ];
    app.mark_history_updated();
@@ -2574,8 +2574,8 @@ fn detail_target_prefers_visible_tool_card() {
            output: Some("done".to_string()),
            prompts: None,
            spillover_path: None,
-                output_summary: None,
-                is_diff: false,
+            output_summary: None,
+            is_diff: false,
        })),
        HistoryCell::Assistant {
            content: "ok".to_string(),
@@ -2588,8 +2588,8 @@ fn detail_target_prefers_visible_tool_card() {
            output: Some("...".to_string()),
            prompts: None,
            spillover_path: None,
-                output_summary: None,
-                is_diff: false,
+            output_summary: None,
+            is_diff: false,
        })),
    ];
    app.tool_details_by_cell.insert(
@@ -2681,8 +2681,8 @@ fn spillover_pager_section_returns_none_when_no_spillover() {
        output: Some("hi".to_string()),
        prompts: None,
        spillover_path: None,
-            output_summary: None,
-            is_diff: false,
+        output_summary: None,
+        is_diff: false,
    }))];
    app.resync_history_revisions();
    assert!(spillover_pager_section(&app, 0).is_none());
@@ -2704,8 +2704,8 @@ fn spillover_pager_section_loads_file_when_present() {
        output: Some("(truncated head)".to_string()),
        prompts: None,
        spillover_path: Some(path.clone()),
-            output_summary: None,
-            is_diff: false,
+        output_summary: None,
+        is_diff: false,
    }))];
    app.resync_history_revisions();

@@ -2729,8 +2729,8 @@ fn spillover_pager_section_returns_notice_when_file_missing() {
        output: Some("(truncated head)".to_string()),
        prompts: None,
        spillover_path: Some(bogus),
-            output_summary: None,
-            is_diff: false,
+        output_summary: None,
+        is_diff: false,
    }))];
    app.resync_history_revisions();

@@ -2754,7 +2754,7 @@ fn terminal_pause_has_live_owner_only_for_running_exec_cells() {
            duration_ms: None,
            source: ExecSource::Assistant,
            interaction: Some("interactive".to_string()),
-                output_summary: None,
+            output_summary: None,
        })),
    );
    app.active_cell = Some(active);
@@ -2770,8 +2770,8 @@ fn terminal_pause_has_live_owner_only_for_running_exec_cells() {
            output: None,
            prompts: None,
            spillover_path: None,
-                output_summary: None,
-                is_diff: false,
+            output_summary: None,
+            is_diff: false,
        })),
    );
    app.active_cell = Some(active);
@@ -2795,8 +2795,8 @@ fn active_rlm_task_entries_surface_foreground_rlm_work() {
            output: None,
            prompts: None,
            spillover_path: None,
-                output_summary: None,
-                is_diff: false,
+            output_summary: None,
+            is_diff: false,
        })),
    );
    app.active_cell = Some(active);
@@ -2687,8 +2687,8 @@ mod tests {
            output: Some("hello world ".repeat(420)),
            prompts: None,
            spillover_path: None,
-                output_summary: None,
-                is_diff: false,
+            output_summary: None,
+            is_diff: false,
        }));
        for width in [40u16, 80, 111, 165] {
            let lines = cell.lines(width);
@@ -2742,8 +2742,8 @@ mod tests {
                output: Some(output),
                prompts: None,
                spillover_path: None,
-                    output_summary: None,
-                    is_diff: false,
+                output_summary: None,
+                is_diff: false,
            })));

            let height: u16 = 30;
@@ -3122,8 +3122,8 @@ mod tests {
                    output: Some(format!("found 12 matches in cell-{i}")),
                    prompts: None,
                    spillover_path: None,
-                        output_summary: None,
-                        is_diff: false,
+                    output_summary: None,
+                    is_diff: false,
                }))
            } else if i % 2 == 0 {
                HistoryCell::User {
@@ -1,7 +1,7 @@
 {
  "name": "deepseek-tui",
-  "version": "0.8.26",
-  "deepseekBinaryVersion": "0.8.26",
+  "version": "0.8.27",
+  "deepseekBinaryVersion": "0.8.27",
  "description": "Install and run deepseek and deepseek-tui binaries from GitHub release artifacts.",
  "author": "Hmbown",
  "license": "MIT",
@@ -55,6 +55,7 @@ export async function runTriage(env: AgentEnv): Promise<Record<string, unknown>>
        headers: {
          Accept: "application/vnd.github+json",
          "X-GitHub-Api-Version": "2022-11-28",
+          "User-Agent": "deepseek-tui-web",
          ...(env.GITHUB_TOKEN ? { Authorization: `Bearer ${env.GITHUB_TOKEN}` } : {}),
        },
      }
@@ -119,6 +120,7 @@ export async function runPrReview(env: AgentEnv): Promise<Record<string, unknown
        headers: {
          Accept: "application/vnd.github+json",
          "X-GitHub-Api-Version": "2022-11-28",
+          "User-Agent": "deepseek-tui-web",
          ...(env.GITHUB_TOKEN ? { Authorization: `Bearer ${env.GITHUB_TOKEN}` } : {}),
        },
      }
@@ -142,6 +144,7 @@ export async function runPrReview(env: AgentEnv): Promise<Record<string, unknown
            headers: {
              Accept: "application/vnd.github+json",
              "X-GitHub-Api-Version": "2022-11-28",
+              "User-Agent": "deepseek-tui-web",
              ...(env.GITHUB_TOKEN ? { Authorization: `Bearer ${env.GITHUB_TOKEN}` } : {}),
            },
          });
@@ -200,6 +203,7 @@ export async function runStale(env: AgentEnv): Promise<Record<string, unknown>>
        headers: {
          Accept: "application/vnd.github+json",
          "X-GitHub-Api-Version": "2022-11-28",
+          "User-Agent": "deepseek-tui-web",
          ...(env.GITHUB_TOKEN ? { Authorization: `Bearer ${env.GITHUB_TOKEN}` } : {}),
        },
      }
@@ -264,6 +268,7 @@ export async function runDupes(env: AgentEnv): Promise<Record<string, unknown>>
        headers: {
          Accept: "application/vnd.github+json",
          "X-GitHub-Api-Version": "2022-11-28",
+          "User-Agent": "deepseek-tui-web",
          ...(env.GITHUB_TOKEN ? { Authorization: `Bearer ${env.GITHUB_TOKEN}` } : {}),
        },
      }
@@ -325,6 +330,7 @@ export async function runDigest(env: AgentEnv): Promise<Record<string, unknown>>
          headers: {
            Accept: "application/vnd.github+json",
            "X-GitHub-Api-Version": "2022-11-28",
+            "User-Agent": "deepseek-tui-web",
            ...(env.GITHUB_TOKEN ? { Authorization: `Bearer ${env.GITHUB_TOKEN}` } : {}),
          },
        }
@@ -335,6 +341,7 @@ export async function runDigest(env: AgentEnv): Promise<Record<string, unknown>>
          headers: {
            Accept: "application/vnd.github+json",
            "X-GitHub-Api-Version": "2022-11-28",
+            "User-Agent": "deepseek-tui-web",
            ...(env.GITHUB_TOKEN ? { Authorization: `Bearer ${env.GITHUB_TOKEN}` } : {}),
          },
        }
@@ -9,10 +9,9 @@
 * - Cites specific files / line numbers / linked issues when discussing code.
 * - Always ends with the draft disclaimer.
 */
-const BASE = process.env.DEEPSEEK_BASE_URL ?? "https://api.deepseek.com";
-const MODEL = process.env.DEEPSEEK_MODEL ?? "deepseek-v4-flash";
-
 const MAX_OUTPUT_TOKENS = 2_000;
+const FALLBACK_BASE = "https://api.deepseek.com";
+const FALLBACK_MODEL = "deepseek-v4-flash";

 interface ChatMessage {
  role: "system" | "user" | "assistant";
@@ -47,17 +46,20 @@ export async function agentChat(
  apiKey: string,
  jsonMode = false
 ): Promise<{ content: string; usage: { input: number; output: number } }> {
-  const res = await fetch(`${BASE}/v1/chat/completions`, {
+  const base = process.env.DEEPSEEK_BASE_URL ?? FALLBACK_BASE;
+  const model = process.env.DEEPSEEK_MODEL ?? FALLBACK_MODEL;
+  const res = await fetch(`${base}/v1/chat/completions`, {
    method: "POST",
    headers: {
      "Content-Type": "application/json",
      Authorization: `Bearer ${apiKey}`,
    },
    body: JSON.stringify({
-      model: MODEL,
+      model,
      messages,
      temperature: 0.3,
      max_tokens: MAX_OUTPUT_TOKENS,
+      reasoning_effort: "high",
      ...(jsonMode ? { response_format: { type: "json_object" } } : {}),
    }),
  });
@@ -1,7 +1,7 @@
 import type { CuratedDispatch, FeedItem, RepoStats } from "./types";

-const BASE = process.env.DEEPSEEK_BASE_URL ?? "https://api.deepseek.com";
-const MODEL = process.env.DEEPSEEK_MODEL ?? "deepseek-v4-flash";
+const FALLBACK_BASE = "https://api.deepseek.com";
+const FALLBACK_MODEL = "deepseek-v4-flash";

 interface ChatMessage {
  role: "system" | "user" | "assistant";
@@ -13,16 +13,20 @@ interface ChatResponse {
 }

 export async function chat(messages: ChatMessage[], apiKey: string, jsonMode = false): Promise<string> {
-  const res = await fetch(`${BASE}/v1/chat/completions`, {
+  const base = process.env.DEEPSEEK_BASE_URL ?? FALLBACK_BASE;
+  const model = process.env.DEEPSEEK_MODEL ?? FALLBACK_MODEL;
+  const res = await fetch(`${base}/v1/chat/completions`, {
    method: "POST",
    headers: {
      "Content-Type": "application/json",
      Authorization: `Bearer ${apiKey}`,
    },
    body: JSON.stringify({
-      model: MODEL,
+      model,
      messages,
      temperature: 0.4,
+      max_tokens: 4096,
+      reasoning_effort: "high",
      ...(jsonMode ? { response_format: { type: "json_object" } } : {}),
    }),
  });
@@ -18,8 +18,8 @@ export interface RepoFacts {
 }

 export const FACTS: RepoFacts = {
-  "generatedAt": "2026-05-08T19:10:07.269Z",
-  "version": "0.8.22",
+  "generatedAt": "2026-05-10T13:27:03.912Z",
+  "version": "0.8.26",
  "crates": [
    "agent",
    "app-server",
@@ -21,7 +21,8 @@
  ],
  "vars": {
    "GITHUB_REPO": "Hmbown/deepseek-tui",
-    "DEEPSEEK_MODEL": "deepseek-v4-flash"
+    "DEEPSEEK_MODEL": "deepseek-v4-flash",
+    "DEEPSEEK_BASE_URL": "https://gateway.ai.cloudflare.com/v1/cf50f793171d7cb3b2ce23368b69cdcb/deepseek-tui-web/deepseek"
  },
  "triggers": {
    "crons": [