# Changelog

All notable changes to this project are documented here. The format follows
[Keep a Changelog](https://keepachangelog.com/en/1.1.0/) and the project uses
[Semantic Versioning](https://semver.org/) (pre-1.0: a minor bump may break).

## [Unreleased]

## [0.2.13]

### Added

- Portable UI renderer multilinguality: supports `host.locale` and `renderer.setLocale(lang)` (e.g. `zh-CN`, `en-US`), translating declarative UI toolbars, options, and actions via built-in localization dictionaries instead of defaulting strictly to English.
- Detailed validation error diagnostics: surfaces specific schema validation errors (`treeValidation.errors`) on invalid UI definitions to help locate incompatible fields instead of generic `Unsupported or invalid UI definition`.

### Fixed

- Table caption unescaping: automatically decodes HTML entity strings (`>`, `<`, `&`, `"`) in table captions emitted by LLM outputs.

## [0.2.12]

### Added

- `Video`: a YouTube (`watch`, `youtu.be`, `shorts`), Bilibili (`BV…`, with `?p=`) or Vimeo page link plays in the platform's official embed player (YouTube via `youtube-nocookie.com`), which is loaded only after the user presses play; the player URL is rebuilt from a validated id, never copied from model input, and the frame is sandboxed without top navigation. A direct https `.mp4`/`.webm` file plays with native controls and `preload="metadata"`. Optional `title`, `poster`, `caption`, `credit`/`creditUrl`, `start` (seconds) and `aspectRatio`. An "Open on …" / "Open original video" link is always shown, and a file that fails to load is replaced by an explanation. The Host rejects non-https sources and platform links without a playable id. Follows `mediaAllowedOrigins` and the offline package's no-remote-media rule.

- `Image` and `Gallery` (1–24 photos): absolute https sources only (no `http:`, `data:`, relative or credentialed URLs, checked by the Host), loaded lazily with `referrerpolicy="no-referrer"`; required `alt` text, optional `caption` and `credit`/`creditUrl`. Tapping opens a modal full-size viewer (`<dialog>`, focus starts on Close, arrow keys and buttons step through, Escape closes). A photo that fails to load is replaced by its description, the reason and an "Open original" link; the rest of the gallery is unaffected. New live setting `mediaAllowedOrigins` restricts sources to listed origins; the offline package never loads remote media, even for replayed results.

- `Map`: places on a Web Mercator map with markers (WGS84 `lat`/`lon`, optional `group` colours with a toggleable legend and a `note` per place), routes as `[lat, lon]` point lists (straight segments, `dashed` for itinerary sketches; drawn the short way across the antimeridian), drag/keyboard panning, zoom buttons, `+`/`−`, double-click and Ctrl+wheel (the page scroll is never captured), a scale bar, and a place list with a details panel. View and selection survive remounts and reloads. Works offline with a built-in Natural Earth world outline; set `mapTileUrl`, `mapTileAttribution` and `mapTileMaxZoom` to use a raster tile basemap, which falls back to the outline when tiles cannot load. The Host rejects out-of-range or swapped coordinates, duplicate ids and oversized data (500 markers, 50 routes, 5,000 route points).

## [0.2.11]

### Changed

- Supports DeepSeek Harness 0.2.1-rc.3, 0.2.1-rc.4 (`next`) and 0.2.2-alpha.2 (`alpha`). The `@deepseek-ai/dsh` and `@deepseek-ai/dsh-llm` peer ranges now list the 0.2.2 prereleases, and the `@deepseek-ai/cordis` peer range the 4.0.6 prereleases, so ordinary npm co-installation agrees with DSH's own compatibility gate. Install smoke test, the full browser E2E (43 checks) and the DSH extension E2E pass on all three; no runtime changes were needed.

## [0.2.10]

### Changed
- Widget policy settings (`cdnAllowlist`, `allowNetwork`, `maxWidgetHeight`, `maxWidgetBytes`, `sendPromptMaxChars`, `openLinkAllowedOrigins`) are now live settings: they can be edited from the official settings forms, including DeepSeek Harness Remote clients, and apply to the next tool call without reloading the plugin.
- Settings descriptions are available in Simplified Chinese.

### Security

- Separated popup, top-navigation, form and download probes in the sandbox audit. A form check now requires an actual CSP violation and no request; download checks use browser events with a positive control. Cross-origin download links can still issue self-navigation requests even when saving the response is blocked.
- Widget sandbox boundary audit ([#19](https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/issues/19)), with the measured matrix in [docs/widget-csp-audit.md](docs/widget-csp-audit.md) and a regression suite in `npm run e2e:security` covering every resource type and request channel (`img`, `script`, `style`, `font`, `media`, `fetch`, XHR, WebSocket, `EventSource`, `sendBeacon`, `<a ping>`, `iframe`, `object`, form, download, popup, navigation).
- **Fixed:** host messages now use a document-bound `MessageChannel` established by the bootstrap before generated code. A navigated document cannot receive host RPC results, HTML or theme updates, even if it delays its `load` event indefinitely. A later iframe `load` still removes foreign UI; it is a best-effort visual signal, not the data boundary. Rebuilds rotate credentials and reset lifecycle in a layout effect; stale async replies cannot reach a new channel. Self-navigation itself and data in its initial URL remain possible.
- Documented that `allowNetwork=false` stops script-visible network access (`fetch`, XHR, WebSocket, `EventSource`, `sendBeacon`, `<a ping>`) but not resource-element requests to an **allowlisted** origin, and that this implementation emits origin-only CSP sources without path restrictions. The `allowNetwork` setting description says so, and an empty `cdnAllowlist` remains the way to block external resource requests, but cannot stop self-navigation.

## [0.2.9]

### Fixed

- `ui_render` and `ui_compose` catalog blocks: the tool schema and description now say that the top-level `root` field is required, and a call without it gets an error that says how to fix it. In real-model runs DeepSeek models often named the first component `"root"` but left out the `root` field, so the first call was refused and had to be retried ([#62](https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/pull/62)).
- Release publication verification: wait up to 10 minutes (30 attempts) for npm registry propagation to avoid false-alarm verification failures ([#56](https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/pull/56)).
- Evaluation routing: guard against missing chart properties and empty block lists, accept local OpenAI-compatible HTTP endpoints via `DEEPSEEK_EVAL_ENDPOINT`, and judge delivery on inline widgets instead of workspace tool calls ([#59](https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/pull/59), [#61](https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/pull/61)).

### Added

- [docs/presentation-routing.md](docs/presentation-routing.md) records the presentation-routing study for [#58](https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/issues/58): rule overlaps, four comparable projects, and a real-model A/B (DeepSeek V4.1 Flash, 3 runs × 39 cases) in which a rewritten decision-order guidance did not route better than the current one (114/117 vs. 111/117 routed correctly), so the current guidance is kept ([#64](https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/pull/64)).
- Real-model evaluation group F: paired presentation-routing cases for [#58](https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/issues/58). Each pair keeps the topic and changes only the task, so a prompt cannot score well by always or never drawing UI; the whole answer is judged (explanation, code, interaction, the user's own data), not just the tool name. Categories `no-ui`, `needs-ui` and `update` report the unnecessary-UI rate, the missed-UI rate and follow-up edits separately. API cases also record prompt and completion tokens ([#59](https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/pull/59)).
- Storefront screenshots manifest: `screenshots.json` declares the showcase screenshots and order for plugin markets like `dshmarket`.
- [CONTRIBUTING.md](CONTRIBUTING.md) guide covering development setup, the committed-bundle rule, test layout, required CI checks, translations, PR automation, and release integrity rules ([#57](https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/pull/57)).

## [0.2.8]

### Added

- `ui_compose`: one answer whose ordered blocks interleave markdown prose with the existing declarative catalog and the existing sandboxed widget, for analysis, teaching and sectioned reports where UI must sit between paragraphs ([#48](https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/pull/48)). Blocks reuse the existing validators, renderers and sandbox; each catalog block is patchable in place as `<composition uiId>#<block id>`, and block-level session identity keeps same-named components in different blocks independent. The finished composition is shown beneath the reply with a "Copy text" action, and the portable renderer accepts a `composition` artifact. New `enableCompose` setting (default on).

- `Choice` single-choice card with "Continue in this session" / "Start new session", and Button `sessionTarget: "new"` to start an independent Agent session in the same workspace, so a decision can be confirmed without typing.

### Changed

- CI runs the full macOS matrix only nightly, on demand and for releases; pull requests and pushes to `main` cover every other cell ([#49](https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/pull/49)).
- Repository automation: new issues are classified with existing labels, and pull requests receive a Chinese static review bound to the reviewed commit ([#50](https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/pull/50), [#51](https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/pull/51), [#54](https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/pull/54)). The bots run trusted default-branch code only, read PR content through the API, review in batches so no file is dropped, and never approve or merge.
- The 0.2.7 changelog described `ui_compose` before it shipped; it is released for the first time here.

### Fixed

- Composition state is restored from history only when the receipt and capability checks passed, so a rejected composition cannot become a patchable UI after a reload.
- Retrying a patched block reuses its original session instead of creating a second one, and widget runtime errors are surfaced in both the DSH client and the standalone renderer.
- The option picker no longer draws a full-row focus ring, and choosing an option focuses the action instead of the row.

## [0.2.7]

### Added

- `ui_compose`: one answer whose ordered blocks interleave markdown prose with the existing declarative catalog and the existing sandboxed widget, for analysis, teaching and sectioned reports where UI must sit between paragraphs. Blocks reuse the existing validators, renderers and sandbox; each catalog block is patchable in place as `<composition uiId>#<block id>`. New `enableCompose` setting (default on). The finished composition is shown beneath the reply with a "Copy text" action, and the portable renderer accepts a `composition` artifact. Guidance now routes prose-heavy answers here while keeping dashboards terse.

- `Choice` single-choice card with "Continue in this session" / "Start new session", and Button `sessionTarget: "new"` to start an independent Agent session in the same workspace ([#39](https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/pull/39)).
- `Table` data-grid options ([#17](https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/issues/17)): typed sorting (numbers, ISO 8601 dates, booleans; empty values last), `columnTypes`, `searchable`, `filterable` (per-column contains / min-max ranges, AND-combined), `pageSize` for inline rows, and `exportable` CSV/Markdown of the filtered, sorted rows. All local; nothing is sent to the model. Interaction state survives remounts and reloads.
- CSV export is RFC 4180 with a UTF-8 BOM and neutralises spreadsheet formula injection (`=`, `+`, `-`, `@`, TAB, CR); Markdown export escapes table syntax.
- `Scatter`: continuous x/y observations with axis labels and units, per-series legend toggles, nearest-point hover, tap and arrow-key inspection; points missing a coordinate are hidden and counted. Laid out at the card's real width so axis text stays readable on phones. Up to 2,000 points.
- `Heatmap`: category × category matrix with sequential or zero-centred diverging scale, optional fixed `min`/`max`, a colour bar, cell values when they fit, and keyboard/hover/tap cell inspection. `null` cells are drawn hatched as "no data", never as 0. Up to 60 × 60 cells.
- Code and structure viewers: `CodeDiff` (multi-file, before/after text diffed in the browser or a unified `patch`; line numbers, +/- counts, expandable unchanged context, large files collapsed), `FileTree` (slash paths with status badges and notes, keyboard expand/collapse) and `JsonView` (collapsible tree with `$.a[0].b` paths and copy path/value). All content is rendered as text; nothing is read from disk or executed. Size and depth limits are validated by the Host.
- `Chart`: keyboard focus with arrow/Home/End/Escape stepping through categories, tap-to-inspect on touch screens, and `exportable` data export. Bound `dataRef` tables and charts can export too. Charts are laid out at the card's real width, so axis labels stay legible on phones and fewer x labels are shown when space is short.
- Browser E2E checks 13a–13g for the new components (filters, CSV contents and formula neutralisation, scatter hover, heatmap keyboard, CodeDiff as text, no model calls); `E2E_DATA_ONLY=1` runs only them.
- `docs/component-roadmap.md`: what #17 shipped, and design proposals for `Map` and `Calendar`.
- Per-feature demo videos in the README (Choice, data grid, Scatter/Heatmap, CodeDiff/FileTree/JsonView), recorded live in the DSH desktop app with DeepSeek models through magpie. `npm run demo:features` re-records them: `scripts/demo/record-features.mjs` drives the desktop app over the DevTools protocol and `scripts/demo/camera.mjs` renders a camera that follows the cursor, holds between actions and fast-forwards model waits.

### Changed

- The guidance points the model at the new components (Choice for picking between alternatives, data-grid options for explorable tables, Scatter, Heatmap, CodeDiff, FileTree, JsonView) instead of prose, Tabs or custom widget JavaScript. With DeepSeek-V4.1-Flash in the DSH desktop app, a price/rating question produced a Scatter and a function comparison a CodeDiff on the first try.
- Interaction state that guards a side effect (starting a new session) is written durably: if browser storage is unavailable the action fails before creating anything, so a retry after reload cannot start the task twice.
- The README was reorganized: stale sections removed, an Extending section and a documentation index added, and the feature tour videos embedded.

### Fixed

- CI: Windows browser jobs no longer fail intermittently. Every real-browser script launches Chrome through one helper that uses a fresh profile directory per attempt and retries ("The browser is already running for …"), and first navigations get 60 s with one retry instead of 10 s.

## [0.2.6]

### Added

- Catalog API v1 for trusted third-party components: namespaced manifests, strict shared props schemas, version checks, React renderers, disable/uninstall lifecycle, and safe fallback in streamed, patched and portable UIs. See `docs/component-extensions.md`.
- Public schema-inferred TypeScript declarations and a runnable GitDiffViewer extension template.
- README npm download badges.

### Fixed

- Extension validation and offline package verification now agree on what an extension package must contain.

### Changed

- Bundled React and React DOM 19.3.1; dev tooling bumps (puppeteer-core 25.14.0, vitest, pnpm/action-setup 6.1.0). CDN-pinned widget libraries (KaTeX, Mermaid, Three.js, D3, Chart.js, canvas-confetti) are excluded from Dependabot so online and offline versions stay in lockstep.

## [0.2.5]

### Added

- Controlled data binding and action bridge for trusted hosts ([#15](https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/issues/15)): Table/Chart `dataRef` share one host-registered dataset with local search, filter, sort and paging; action Buttons and `Bridge.invokeAction` call allowlisted, schema-checked host actions only after a native click, with host authorization and write confirmation. Exported as `deepseek-harness-intelligent-ui/controlled`. See `docs/controlled-data-actions.md`.
- Optional offline browser-resource tarball (`npm run build:offline`) for Math/KaTeX fonts, Mermaid, Three.js/OrbitControls, D3, Chart.js and canvas-confetti ([#18](https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/issues/18)). Exact locked versions, upstream licenses, checksums, size budgets, Host capability guidance and network-isolated Chromium regressions. Default online installation stays compatible.
- Explicit static fallback status for unavailable Math/Mermaid libraries.

### Fixed

- Action Buttons no longer fail when DSH Web is served over plain HTTP (e.g. a LAN address), where `crypto.randomUUID` is unavailable.
- A widget rebuilt while an action was running no longer loses its newly staged action request.
- CI: the plain-answer E2E check waits for the answer itself; the release workflow waits up to 5 minutes for registry, tarball and attestation propagation.

### Compatibility

- Tested with DSH 0.2.1-rc.2 (`next`) and 0.2.1-alpha.5 (`alpha`) on cordis 4.0.5; CI adds a non-gating `next` column.

## [0.2.4]

### Added

- `ui_patch`: change a UI shown earlier in the conversation by component id (`setProps`, `replaceNode`, `insertChild`, `removeNode`) instead of re-sending the whole tree ([#14](https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/issues/14)). `ui_render` results now report the UI's `uiId` and version. Patches are atomic, revalidated as a whole tree, version-checked (`baseVersion`; stale patches are refused with the current version), and limited to UIs of the same conversation. The Host rebuilds versions by replaying the session history, so it keeps no state.
- Untouched components keep their interaction state across versions (table sort, tab, form input and submitted lock, Plot sliders). Streaming, refused, failed or stopped patches never replace the previous version; superseded copies collapse to a pointer; a reload restores the latest version.
- Browser E2E: a cross-turn patch scenario against real DSH Web (refusal and retry, in-place bar → line, state kept, reload).
- Real-model quality baseline ([#13](https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/issues/13)): `npm run eval:real` runs three separately reported layers: the scripted-model browser E2E, a real-model **API** stage, and a new real-model **browser** stage in real DSH Web + Chrome. It exits non-zero on any failure.
  - API stage: the model sees the plugin's real three tools (now including `ui_patch`) and system guidance. The tools' own `execute()` produces the tool results, so `ui_patch` edits are scored on the UI the user ends up with. Each case records tool choice, first-attempt schema validity versus recovery, and per-criterion checks of the rendered components and data. New cases: two successive chart edits (D3); D2 accepts an in-place patch.
  - Browser stage: plain text, multi-series chart (legend, Q1–Q4, hover), sortable table, form submit → follow-up using the submitted values, Plot sliders, Three.js (non-blank WebGL canvas, drag rotates), stop → regenerate, two in-place edits, Fast 3G with cache off, CDN unreachable, and phone width. Every case runs in a new conversation and records console errors.
  - Timings are recorded per stage: API time per model call, and in the browser first visible preview, complete generation and browser-side execution, each with p50/p95. `--repeat N` gives per-case success rates.
  - The report (`docs/real-model-eval-report.json` + `.md`) records date, model (requested, API-reported and fingerprint), prompt hash, assertion version, environment and failure reasons. It is redacted, and the run refuses to write a report that still contains the key.
  - A manually triggered `Real-model eval` workflow uses the `DEEPSEEK_API_KEY` secret.
- `scripts/eval/edit-demo.mjs` (`npm run eval:demo`) turns the browser-stage recording into a designed demo video, narrated by the DeepSeek whale girl in 26 poses from [Sutera-Diffusus/dsh-whale-musume](https://github.com/Sutera-Diffusus/dsh-whale-musume) (MIT). She switches poses with crossfades to fit each beat (hello, thinking at her laptop, an umbrella for the slow network, tears for a failure, confetti for a success), with motion on top. Each chapter is marked ✓/✗ and the narration follows the real outcome. Every frame is rendered deterministically.
  - The recorder marks non-content spans (new conversations, screenshots, viewport switches) to cut, and holds each case's result on screen for a few seconds.
  - Model waits are sped up; the recording's real-time stopwatch and a speed-factor sticker keep that visible.
- `scripts/eval/gallery.mjs` puts the README screenshots on one uniform 1200×750 canvas.
- Portable renderer for remote clients ([#22](https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/issues/22), experimental):
  - `renderer/renderer.js` is a prebuilt, host-neutral WebView bundle (`DshIntelligentUiRenderer.mount(element, host)`) that renders accepted `ui_render`, `ui_patch` and `ui_render_widget` artifacts without DSH Web. It ships in the npm package and is checked in CI and before release.
  - `deepseek-harness-intelligent-ui/portable` exports `projectNativeUiEvents`, which turns native Session events into artifacts.
  - Widget errors inside the sandbox are reported to the remote client.
  - See [docs/portable-renderer.md](https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/blob/main/docs/portable-renderer.md).

### Fixed

- Browser E2E: the dark-theme check polls until the host card and the widget frame have both switched, instead of measuring after a fixed 1.2 s. The nightly `dsh latest · macos-latest` job had measured too early.

### Changed

- The guidance prefers `ui_patch` for changes to an existing `ui_render` UI; widgets are still re-rendered in full.
- `Form` fields added by a later version start at their defaults; a `Table` sort on a removed column is ignored.
- The 2026-10-10 real-model report (15/15, including the B3 false positive) moved to `docs/history/` as a labelled historical snapshot. A test now pins that the historical B3 answer fails.

## [0.2.3]

### Changed

- Verified against DeepSeek Harness 0.2.1-alpha.4 (install smoke test and the 24-check browser E2E pass with no code changes). It is now in `TESTED_DSH_VERSIONS`.
- Redesigned both READMEs with animated SVG illustrations (hero, agent routing, widget streaming, security model), a highlights grid, badges, a compatibility table and collapsible sections. `npm run docs:art` regenerates the illustrations.

## [0.2.2]

### Fixed

- npm co-installation with the supported DSH 0.2.0-rc and 0.2.1-alpha lines now uses explicit prerelease peer ranges.
- README installation examples use the 0.2.2 release version, explain the optional `next` tag, and link to versioned documentation and media outside the tarball.

### Changed

- Release reruns verify registry integrity, exact package contents and provenance before reusing a version. GitHub Release assets come from that registry tarball and conflicting assets are refused.
- Release preflight requires the tagged commit to belong to main. Actions use immutable commit SHAs, and publishing can opt into npm Trusted Publishing after account-side setup.

## [0.2.1]

### Changed

- Rebuilt the client bundle with esbuild 0.28 and idiomorph 0.8 (widget DOM morphing). The catalog sources also type-check against React 19 types.
- CI and release workflows use `actions/checkout@v7`, `actions/setup-node@v7`, `actions/upload-artifact@v7` and `pnpm/action-setup@v6`. Browser E2E runs in isolated workspaces and cleans up its processes reliably.

### Fixed

- Real-model evaluation now reads `DEEPSEEK_API_KEY` from the environment, validates requested rendered components and multi-turn data, acknowledges all tool calls, and exits nonzero on failures. Partial runs no longer combine historical results under a new timestamp. API errors and missing messages fail clearly; requests have a timeout.
- Marked the legacy evaluation report's missing-Chart false positive and clarified the limits of API-only evaluation in both READMEs.

## 0.2.0 (development milestone)

### Changed

- **Renamed the npm package to `deepseek-harness-intelligent-ui`** (`dsh-intelligent-ui` belongs to an unrelated project on npm). The bundle row id stays `intelligent-ui`, so existing `cordis.patch.yml` overrides keep working. Remove the old `file:` install with `dsh plugin --profile web remove dsh-intelligent-ui` before installing this one.
- **Supports DeepSeek Harness `>=0.1.7-rc.1 <0.3.0`** instead of only `0.2.0-rc.1`, which DSH refused to install on `0.2.0-rc.2` and later. The peers are marked optional because DSH provides them, so pnpm no longer warns about "missing peers".

### Added

- Live streaming previews on DSH 0.2.1+. DSH 0.2.1 removed the `useToolCallArgumentsPartial` hook that tool views used to receive the streamed arguments. The views now fall back to the call's growing `args` view, so catalog UIs and widgets still render progressively and still survive a stopped generation.
- One-command install from npm: `dsh plugin --profile web add deepseek-harness-intelligent-ui`.
- `npx deepseek-harness-intelligent-ui install | uninstall | doctor`: checks Node, dsh (and that its version is supported), pnpm and the profile, then runs DSH's own plugin manager.
- Cross-platform E2E (Windows, Linux, macOS). It also waits for DSH 0.1.x's delayed first-run notice.
  - Chrome is detected per platform, process cleanup works on POSIX, and `--spec` installs a packed tarball or a registry version.
- GitHub Actions:
  - CI on 3 OSes × Node 22.13/24.
  - A nightly compatibility matrix across DSH versions (install smoke test plus browser E2E).
  - A tag-driven release that publishes to npm with provenance and creates the GitHub Release.

## 0.1.0 (development milestone)

### Added

- Initial release:
  - The declarative component catalog (`ui_render`) and sandboxed streaming HTML widgets (`ui_render_widget`).
  - The `/ui` command and the guidance system-prompt section.

[Unreleased]: https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/compare/v0.2.9...HEAD
[0.2.9]: https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/compare/v0.2.8...v0.2.9
[0.2.8]: https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/compare/v0.2.7...v0.2.8
[0.2.7]: https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/compare/v0.2.6...v0.2.7
[0.2.6]: https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/compare/v0.2.5...v0.2.6
[0.2.5]: https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/compare/v0.2.4...v0.2.5
[0.2.4]: https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/compare/v0.2.3...v0.2.4
[0.2.3]: https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/compare/v0.2.2...v0.2.3
[0.2.2]: https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/compare/v0.2.1...v0.2.2
[0.2.1]: https://github.com/dreamfarer-space/deepseek-harness-intelligent-ui/releases/tag/v0.2.1