Skip to content

Commit 6b5011a

Browse files
authored
Merge pull request #1587 from Hmbown/work/v0.8.35
chore(release): prepare v0.8.35
2 parents 81e4b93 + d505142 commit 6b5011a

139 files changed

Lines changed: 11294 additions & 4473 deletions

File tree

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

.gitignore

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -91,3 +91,7 @@ apps/
9191

9292
# Maintainer-internal design notes (trade-secret material, never published)
9393
.private/
94+
95+
# direnv
96+
.envrc
97+
.direnv

CHANGELOG.md

Lines changed: 129 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -7,6 +7,132 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
77

88
## [Unreleased]
99

10+
## [0.8.35] - 2026-05-13
11+
12+
A post-0.8.34 cleanup release focused on prompt hygiene, context-pressure
13+
guidance, and keeping the next release branch clearly separated from the
14+
already-published v0.8.34 tag.
15+
16+
> **Note on v0.8.34 contributor credits:** Horace Liu ([@liuhq](https://github.com/liuhq))
17+
> contributed Nix package support and install documentation in the v0.8.34
18+
> cycle but was inadvertently omitted from that release's changelog. The
19+
> README contributor list and this note correct the record.
20+
21+
### Changed
22+
23+
- **First-turn prompt context is leaner and easier to audit.** The
24+
generated project context pack now ignores hidden tool/cache state,
25+
balances top-level directories before descending, and `/context`
26+
shows named prompt layers instead of a single opaque system blob.
27+
- **Model-visible prompt policy de-conflicted.** The base and mode
28+
prompts no longer forbid useful `deepseek` CLI diagnostics, no
29+
longer require checklists for simple one-step work, and align
30+
long-session compaction guidance around the 60% suggestion threshold.
31+
- **Context-pressure guidance now has one split rule.** Manual
32+
`/compact` suggestions start around 60% during sustained work, while
33+
automatic replacement compaction remains an opt-in hard guardrail near
34+
80% so DeepSeek V4 prefix-cache economics stay intact.
35+
- **The Tasks sidebar now ages out stale live-tool noise.** Completed
36+
active tool rows linger briefly and then leave the right rail; very old
37+
running shell rows collapse to a single row instead of occupying the
38+
whole Tasks panel.
39+
40+
### Fixed
41+
42+
- **`auto_compact` settings help now reports the real default**, which
43+
has been off since v0.8.11 to avoid unnecessary cache-prefix rewrites.
44+
45+
## [0.8.34] - 2026-05-13
46+
47+
A polish, terminal-protocol, and internal-cleanup release. The model-facing
48+
surface is stable; this cycle focused on prefix-cache stability metrics,
49+
broader terminal protocol coverage, bundled skills, and shrinking the
50+
mega-files that had grown around the agent loop and TUI.
51+
52+
### Added
53+
54+
- **Prefix-cache stability tracking.** A footer chip surfaces how stable
55+
the cached prefix has been across recent turns (inspired by Reasonix),
56+
so users can spot cache-busting edits before cost climbs.
57+
- **Bundled DeepSeek-native workflow skills.** A starter set of skills
58+
ships in-binary so a fresh install has a usable `/skills` catalog
59+
without external assets.
60+
- **Native Kitty + Ghostty notification protocols.** `OSC 99` (Kitty)
61+
and `OSC 777` (Ghostty) are now first-class alongside the existing
62+
desktop notification fallback.
63+
- **Theme picker with more presets.** Catppuccin, Tokyo Night, Dracula,
64+
and Gruvbox join the built-in palette set; `/theme` now shows a
65+
live picker.
66+
- **Chunked parallel-safe tool execution.** The engine batches
67+
side-effect-free tool calls into a chunked dispatch so independent
68+
reads/searches finish in one turn instead of serialising round-trip
69+
by round-trip.
70+
- **Cancel-all shell jobs.** A single action stops every running
71+
background shell command instead of cancelling them one-by-one.
72+
- **`edit_file` tolerates typographic punctuation drift.** When the
73+
exact-match and leading-whitespace-fuzzy passes both fail and
74+
`fuzz: true` is set, the tool retries with smart quotes (`"`/`"`
75+
`"`, `'`/`'``'`), en/em-dashes (``/```-`), and non-breaking
76+
spaces (U+00A0 → space) normalized to ASCII. Catches the copy-paste
77+
failure mode where a browser or chat client substituted Unicode
78+
punctuation for the ASCII the file actually contains.
79+
80+
### Changed
81+
82+
- **`crates/tui/src/tui/ui.rs` split into focused modules.** The
83+
former 10k-line single-file TUI dispatcher is decomposed into smaller
84+
modules with clearer responsibilities so reviewing a UI change does
85+
not require holding the entire surface in head.
86+
- **`crates/tui/src/core/engine.rs` reduced.** Helper clusters moved
87+
into the existing `core/engine/` submodule directory next to the
88+
turn loop and tool execution code, making the agent-loop core
89+
easier to read end-to-end.
90+
- **Structured tracing on tool dispatch.** Tool entry, exit, duration,
91+
and result/error are emitted through `tracing` events so
92+
`RUST_LOG=engine.tool_execution=debug` produces a coherent timeline
93+
instead of scattered ad-hoc prints.
94+
- **`/init` updates `AGENTS.md` in place** instead of refusing when
95+
the file already exists, so adding new project guidance does not
96+
require manual stitching.
97+
- **Reasoning tokens included in cost calculations**, and the cost
98+
display auto-switches to CNY when the session locale is `zh-Hans`.
99+
- **Stale repo-root development docs removed.** `TAKEOVER_PROMPT.md`
100+
(v0.8.6 era), `PROMPT_ANALYSIS.md`, and the redundant
101+
`DEPENDENCY_GRAPH.md` no longer ship in releases; `docs/ARCHITECTURE.md`
102+
remains the canonical crate-layout reference.
103+
104+
### Fixed
105+
106+
- **Auth keys checked against the saved provider on startup**, so a
107+
stored DeepSeek key is no longer rejected after switching providers
108+
mid-session.
109+
- **Auto router skipped for decisive local routes**, removing an
110+
extra model round-trip on prompts the dispatcher can route directly.
111+
- **Reasoning content stripped for generic providers** that do not
112+
understand the `reasoning_content` field, preventing HTTP 400s when
113+
pointing at an OpenAI-compatible gateway that lacks DeepSeek
114+
thinking semantics.
115+
- **`FocusGained` debounced** so terminals (Tabby) that emit rapid
116+
focus events no longer trigger a repaint flicker loop.
117+
- **MCP HTTP transport defaults `Accept: application/json,
118+
text/event-stream`** and persists `Mcp-Session-Id` across requests,
119+
matching the spec for resumable streams.
120+
- **Shell output tail preserved when truncating**, so the last lines
121+
of a long command output (usually the error trailer) survive the
122+
in-transcript summary.
123+
- **Prefix cache preserved while pruning tool results.** Old
124+
side-effect tool payloads no longer invalidate the prefix that
125+
the next turn would otherwise reuse.
126+
- **Review sub-agents prevented from spawning further sub-agents**
127+
(#1489), keeping recursive depth bounded.
128+
- **Help overlay closes cleanly** and repaints without a stale frame.
129+
- **Pinyin `/skills` alias dispatched correctly** so Chinese-locale
130+
users reach the same surface.
131+
- **VTE flicker terminals get reduced motion** by default to avoid
132+
thrashing on terminals that mishandle frequent partial redraws.
133+
- **Composer border no longer shows the derived session title**, keeping
134+
the composer chrome reserved for editor and mode state.
135+
10136
## [0.8.33] - 2026-05-12
11137

12138
A sub-agent and RLM renovation release. The model-facing delegation
@@ -3979,7 +4105,9 @@ Welcome — and thank you.
39794105
- Hooks system and config profiles
39804106
- Example skills and launch assets
39814107

3982-
[Unreleased]: https://github.com/Hmbown/DeepSeek-TUI/compare/v0.8.33...HEAD
4108+
[Unreleased]: https://github.com/Hmbown/DeepSeek-TUI/compare/v0.8.35...HEAD
4109+
[0.8.35]: https://github.com/Hmbown/DeepSeek-TUI/compare/v0.8.34...v0.8.35
4110+
[0.8.34]: https://github.com/Hmbown/DeepSeek-TUI/compare/v0.8.33...v0.8.34
39834111
[0.8.33]: https://github.com/Hmbown/DeepSeek-TUI/compare/v0.8.32...v0.8.33
39844112
[0.8.32]: https://github.com/Hmbown/DeepSeek-TUI/compare/v0.8.31...v0.8.32
39854113
[0.8.31]: https://github.com/Hmbown/DeepSeek-TUI/compare/v0.8.30...v0.8.31

CONTRIBUTING.md

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -140,7 +140,7 @@ crates/
140140
```
141141

142142
See [docs/ARCHITECTURE.md](docs/ARCHITECTURE.md) for the live data flow across
143-
these crates and [DEPENDENCY_GRAPH.md](DEPENDENCY_GRAPH.md) for build ordering.
143+
these crates, including the bottom-up build order.
144144

145145
## Submitting Changes
146146

Cargo.lock

Lines changed: 14 additions & 14 deletions
Some generated files are not rendered by default. Learn more about customizing how changed files appear on GitHub.

Cargo.toml

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -19,7 +19,7 @@ default-members = ["crates/cli", "crates/app-server", "crates/tui"]
1919
resolver = "2"
2020

2121
[workspace.package]
22-
version = "0.8.33"
22+
version = "0.8.35"
2323
edition = "2024"
2424
# Rust 1.88 stabilized `let_chains` in `if`/`while` conditions, which the
2525
# codebase relies on extensively. Cargo enforces this so users on older

DEPENDENCY_GRAPH.md

Lines changed: 0 additions & 65 deletions
This file was deleted.

0 commit comments

Comments
 (0)