Daily RSS/Atom news digest for DSH: built-in channels plus a curated awesome-rsshub-routes catalog and custom feeds (validated on save), aggregated into one "今日值得读" Markdown digest each day and injected into the system prompt; Web GUI settings card and sidebar reader.
Install
# from npm (prebuilt)
dsh plugin --profile web add @hyzyn/dsh-rss
# from GitHub (first run asks for allowBuilds approval — follow the hint, retry)
dsh plugin --profile web add github:hyzyn/dsh-plugin-kit#path:/packages/rss
Any plugin you install runs third-party code with your own permissions — it can read your files, use your credentials, and reach the network, and tool approvals don’t sandbox it. GitHub-sourced plugins also run build scripts at install time — pnpm blocks those until you allow them, so an install can stop with ERR_PNPM_GIT_DEP_PREPARE_NOT_ALLOWED or ERR_PNPM_IGNORED_BUILDS; dsh prints the exact key to add under allowBuilds in your profile’s pnpm-workspace.yaml, and the install works on the next run. Allowing a build is a trust decision: only install sources you trust, and pin a commit (github:owner/repo#sha).
README
中文 | English
An RSS / news aggregation plugin for DSH: subscribe to multiple RSS / Atom feeds, get one automatically compiled “Today's Worth Reading” digest every day, and have the result injected into the systemPrompt so the model can cite it directly.
Features
- The digest is injected as a systemPrompt section: the model can cite the day's items directly, so asking "what's worth reading today" returns real content. The switch (
announceToAgent) is hot — saving takes effect (shown or withdrawn) without a restart; the section lists at most the first 8 items (indigest.itemsorder, category headings kept) and points to that day's digest file for the rest, so every request does not carry an over-long list. - Scheduled fetching, aggregation by category: generated daily; the modal lets you search, filter by category, and copy the Markdown.
- Feeds maintained inside the panel: built-in channels + custom sources, one-click add from the curated RSSHub catalog, and OPML import/export.
- A failed source degrades in place: when an individual source is down only that source is flagged and the rest still render; the optional AI summary falls back to the original excerpt and explains why.
Transition layer while refreshing
When the “Today's Worth Reading” modal and the settings card refresh, they show a hint in the same visual language as dsh-docker's "target switch transition": a 2px indeterminate shimmer progress bar at the top of the content area, and a blue pill (⟳ Refreshing feeds · current data generated at 14:29, full explanation in title); the bar sits on the lower edge of the category row and the pill rides it, centred — absolutely positioned, no layout shift. It only appears on slow refreshes (the existing 180ms debounce), so millisecond requests never flash; when data lands the body does an 8px slide-up + fade-in, respecting prefers-reduced-motion.
During refresh the body stays dimmed as before (opacity:.55) and the old content remains readable; stale-while-revalidate, the skeleton debounce, and the post-refresh focus/caret and scroll restoration are all retained. The modal's slow path deliberately does not re-render the whole tree, so the transition layer injects two absolutely-positioned nodes instead — miss that path and the overlay never appears.
Implementation notes (the accent-token trap, why absolute positioning is mandatory, the 6 regression assertions) are in ROADMAP § completed (Chinese).
Capabilities
- Built-in channel library (Ruanyifeng's blog, sspai, Solidot, Hacker News, Juejin, ITHome, 36Kr) — just tick the channels you want to show in the settings; 36Kr's official feed is blocked by anti-scraping, so the built-in address is a third-party RSSHub mirror;
- Custom channels are supported: enter your own RSS / Atom URL; on save it is really fetched for validation, and an address that returns no content is reported and not saved; supports OPML import / export (import subscriptions from any RSS reader or website) and bulk import by pasting a URL list (one address per line, or "name, address"), automatically skipping items already subscribed, and after importing you can save and validate them all with one click;
- Feed catalog (multiple sources): the catalog can come from several sources — the built-in curated awesome-rsshub-routes list (official RSS and RSSHub routes, snapshot + silent refresh every 12 hours) + any number of custom OPML catalogs (add OPML addresses in the “Feed catalog” section of the settings card; name / URL are both configurable); results from all sources are merged together, labeled "from xxx" and filterable by source, with search / category filtering / multi-select followed by bulk add (select all / clear / add selected with one click) and live "subscribed / selected" counters; a single failing catalog source does not affect reading the rest of the catalog;
- Parses RSS 2.0 and Atom with zero dependencies, deduplicates by link / id / title, and generates Markdown in reverse chronological order;
- Optional AI summaries: when enabled, the host LLM is called to generate a one-sentence Chinese summary for each item, shown in the digest / modal and appended to the systemPrompt; summaries are cached per entry for 30 days (up to 500 entries,
ai-cache.json), and repeated entries are not requested again; a single failure (timeout / non-stop finish / empty output) falls back to the original excerpt only and does not affect other items, while a missing route skips AI summaries for that whole run and records the reason in the digest; - Generates the day's digest automatically at
08:00by default; if the day's digest does not exist at plugin startup it is also generated to fill the gap; - Injects the day's digest into
systemPrompt, so the model can cite it directly when the user asks about “Today's Worth Reading”; - Provides a Web GUI card “RSS / News Aggregation” for managing built-in channel toggles / custom channels / news categories / aggregation settings, with the day's digest refreshed automatically after saving; custom channels support filtering and counting by name / URL, with an inline warning for duplicate URLs; the top of the card previews today's digest directly (item count / source count / generation time / failure warnings) and offers one-click view list, refresh, and copy Markdown;
- The UI uses "block-scoped update" rendering: typing, catalog search, and adding/removing channels rebuild only the affected blocks, so focus / caret / scroll are not lost; unsaved changes are flagged with a badge;
- Provides a “Today's Worth Reading” shortcut in the sidebar below "New Session" (near Task Board / SSH); clicking it opens a modal for reading the news directly; the modal supports search by title / summary / source, category filtering, a "visible / total" counter, and copy Markdown, plus Esc to close and focus trapping;
- The digest is displayed grouped by "category" (items without a category go under "Uncategorized"), and each item is labeled with its source and date; each source's "View more" links straight to its website;
- Manage built-in channel toggles, custom channels (RSS/Atom URL, category, item limit), and news categories on the UI's "Settings" page; a channel's category is chosen from the "News categories" list, and categories in use are merged into that list automatically on save;
- The generated Markdown is stored at
~/.dsh/rss-digest/YYYY-MM-DD.md, together withlatest.jsonfor external readers.
Screenshots
“RSS / News Aggregation” card (built-in channel toggles + custom channels + news categories):
Where the settings surface lives changes with the DSH version, but it is always the same form — the per-generation landing spots and the four registration contracts live in conventions.md § client settings surface.
Sidebar “Today's Worth Reading” modal (grouped by category, sources with "View more" linking straight to their websites, manual refresh at the bottom):
Asking the model about the day's news cites the day's digest directly (no need to open the modal):
Configuration example
Pass Config in DSH's plugin configuration:
{
sources: [
{ name: 'Ruanyifeng blog', url: 'https://www.ruanyifeng.com/blog/atom.xml', category: 'Tech' },
{ name: 'sspai', url: 'https://sspai.com/feed', category: 'Productivity' },
],
maxItemsPerSource: 5,
maxTotalItems: 30,
dailyTime: '08:00',
digestDir: '~/.dsh/rss-digest',
}
When
sourcesis omitted, all built-in channels are enabled by default; whensourcesis passed, all built-in channels are disabled and only that list is used.
Common configuration options:
| Field | Description | Default |
|---|---|---|
enabled |
Whether to enable the plugin | true |
announceToAgent |
Whether to inject the digest into systemPrompt (hot: shown or withdrawn on save, no restart) | true |
includeCatalog |
Whether to expose the feed catalog (/api/dsh-rss/catalog) | true |
sources |
Custom feed list; when passed it replaces all built-in channels | 7 built-in channels |
maxItemsPerSource |
Maximum number of items taken per source | 5 |
maxTotalItems |
Maximum number of items aggregated per day | 30 |
dailyTime |
Daily auto-generation time (HH:mm) | 08:00 |
autoGenerateOnMount |
Whether to generate the day's digest on startup if it does not exist | true |
digestDir |
Output directory | ~/.dsh/rss-digest |
requestTimeoutMs |
Single request timeout | 10000 |
AI summaries (optional)
AI summaries are off by default, and are configured only through the ai field of the editable store ~/.dsh/rss.json (editable in the "AI summaries" section of the settings card; no new Config field is added):
{
"ai": {
"enabled": true,
"provider": "deepseek",
"model": "deepseek-chat",
"maxItems": 20,
"concurrency": 3,
"timeoutMs": 20000
}
}
| Field | Description | Default |
|---|---|---|
enabled |
Whether to enable AI summaries | false |
provider / model |
Model route, must be filled in as a pair; if both are left empty it follows the host default model (agentDefaultModel / agent-default-model in settings; if neither can be resolved, AI summaries are skipped for that run) |
Left empty |
maxItems |
Maximum number of items summarized per digest (the first N items in the list), clamped to 1..50 | 20 |
concurrency |
Number of concurrent requests, clamped to 1..6 | 3 |
timeoutMs |
Per-request timeout in milliseconds, clamped to 5000..60000 | 20000 |
If only one of provider / model is filled in, the whole ai config is ignored and a warning is returned in the save response. Summary results are written to ai-cache.json in the digest directory ({ version: 1, entries: { <sha1(link|id|title)>: { text, model, at } } }); a hit that has not expired (30 days) is reused as-is, and beyond 500 entries the oldest are evicted by write time.
Two easy pitfalls (both measured on real runs):
maxTokensis hard-coded in the plugin (currently 4096), and reasoning models need room for their reasoning budget: reasoning tokens and the final answer share this cap, so too small a value makes the model burn the budget on reasoning before it writes any prose, the finish reason becomesmax-tokens, and that item's summary is marked failed. Measured (commandcode/deepseek/deepseek-v4.1-flash, 20 items): 200 → 7 succeed, 1024 → 12, 4096 → all 20 succeed (48 s for the whole run, ~7 s each, still insidetimeoutMs). If a heavier reasoning model hits the cap again, the failure reason names the currentmaxTokensvalue.- Failure reasons are surfaced: on partial failure,
aiSummary.failuresgives the distribution of reasons (up to 3 kinds, with counts), and the Markdown “fetch failed” section states them too (e.g.原因:终止原因 max-tokens(…)×8). Previously there was only a bare “N failed”, which made it impossible to tell a timeout from rate limiting from a provider error — or to judge whether the plugin was at fault at all.
Development
pnpm --filter @hyzyn/dsh-rss build
pnpm --filter @hyzyn/dsh-rss typecheck
Installing into DSH
dsh plugin --profile web add link:$(pwd)
Or from the repository root:
dsh plugin --profile web add link:$(pwd)/packages/rss
Files
| File | Description |
|---|---|
package.json |
dsh.bundle.patch points to cordis.patch.yml; main / exports["."] point to lib/index.js |
cordis.patch.yml |
bundle patch: insert: { id: rss-digest, name: '@hyzyn/dsh-rss' } inserts this plugin's row into the profile roster |
src/index.ts |
plugin host half: RSS/Atom parsing, fetching, digest generation, AI summaries, scheduling, HTTP API, systemPrompt injection |
client.js |
browser half: the “RSS / News Aggregation” card |
test/ai-summary.test.ts |
regression tests for AI summaries (message construction / sanitization, cache TTL·LRU, route resolution, store allowlist, rendering, LLM termination branches, generation fallback paths) |
tsconfig.json |
extends the root tsconfig.base.json; tsc emits lib/ |
Links
More in this category
Tencent/WeKnora#dsh-weknora★ 37401
Four read-only tools over a WeKnora knowledge base: list knowledge bases, hybrid passage search, reassemble one document's chunks in order, and WeKnora's own cited RAG or ReAct-agent answer with a resumable session id.
DayuanJiang/next-ai-draw-io#dsh-plugin★ 36932
Draws draw.io diagrams in DeepSeek Harness through the next-ai-drawio MCP server: the link in the reply opens a live preview in the right sidebar, you can edit the diagram by hand, and each AI change is a version you can restore. Node.js 18+; npx downloads the MCP server on first use.
superdesigndev/treg★ 7929
Tool catalog for agents: search ~2,600 external endpoints (SEO and SERP, backlinks, social, people and company enrichment, ad libraries, scraping) by the task you want done, read each one's parameters and per-call price, then call it with the credential injected server-side. Ships the skill plus an MCP row that stays disabled until TREG_TOKEN is set.
T-Auto/dsh-ops★ 5798
Provides bash, PowerShell 7, and high-performance Rust tools for DSH on Windows to reduce token usage.
sixtysevenlf/dsh-blender-plugin★ 2081
Lets a DSH agent drive a local Blender over a direct TCP channel with 15 tools: viewport frames, custom-angle renders, Python execution, headless and job offload, transaction rollback, and a planning layer of 228 ops across 34 families.
yjl9903/AnimeGarden#dsh★ 1400
Anime Garden for DSH: search anime torrent resources from 動漫花園, 蜜柑计划, 萌番组 and ANi inside DSH — via chat, a sidebar panel with fansub and type filters, a seasonal calendar, and an in-app floating player.
Community comments
Comments are public GitHub Discussions. Loading them connects to GitHub and Giscus; a GitHub account is required to post.