Deep reading & summarization for DSH: books/papers/videos/web into structured Obsidian notes via plugin parsers, MapReduce parallel deep-read subagents, JSON Schema outputs and an idempotent cache.
Install
# from npm (prebuilt)
dsh plugin --profile web add deep-read-summarize
# from GitHub (first run asks for allowBuilds approval — follow the hint, retry)
dsh plugin --profile web add github:PensiveFei/deep-read-summarize
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
Turn a paper, book, video, or web page into a Markdown reading note you can keep, edit, and check against the source.
This third-party plugin runs in DeepSeek Harness (DSH). It attempts to retrieve the source text, reads long material in chunks, and combines the results into a structured note with YAML frontmatter. You can use the note in Obsidian, Logseq, or another Markdown tool.
See real notes · Get started · Quality and limitations · Common options
AI-generated notes can omit or misinterpret information. Use them as a starting point for reading, and check important claims and quotations against the source.
See the notes
- 📄 Paper: Emotional Intelligence of Large Language Models
- 📄 Paper: Chain-of-Thought Prompting
- 🎬 Video: Andrej Karpathy's introduction to LLMs
These are historical notes supplied by the maintainer, in Chinese. The paper notes have records of early project runs; how the video note was generated has not been confirmed. Their model, project version, and runtime were not recorded, so they are not performance claims for the current release. Some quotations and timestamps have been spot-checked, and edits to the public copies are documented in the examples guide.

What it does
| You provide | The workflow attempts to retrieve | You receive |
|---|---|---|
| A book: PDF, EPUB, MOBI | Text and chapters | A reading note organized by topic |
| A paper: arXiv, PDF, HTML | Paper text and structure | Research question, method, findings, and limitations |
| A video link | Subtitles first; local transcription when needed and available | A breakdown based on the text obtained |
| A web page | Page text | Key points, concepts, and source information |
Extraction depends on source accessibility, available tools, the model, and host permissions. Scanned PDFs generally need OCR first; the workflow does not provide automatic OCR. Complex formulas, diagrams, and poor-quality captions or audio may also need manual handling.
Notes typically include an overview, structural breakdown, glossary, key quotations, critical analysis, and applications. Their contents are model-generated and may differ between inputs and configurations.
Get started
1. Install the plugin
Install and start DSH Desktop once to initialize its desktop profile. Quit the application completely, then use the DSH command bundled with Desktop to install the plugin:
dsh plugin --profile desktop add deep-read-summarize
Reopen DSH Desktop when installation finishes. The bundled command may not be on your system PATH; use resources/runtime/cli/bin/dsh inside the installation directory (dsh.cmd on Windows). Interfaces and commands can change between host versions; see the official DSH Desktop instructions. Do not run npm install or npm update directly inside a DSH profile directory.
Deep reading also requires a working model and suitable permissions in DSH. Some sources need additional tools: PDF extraction may need pdftotext, YouTube captions may need yt-dlp, and local transcription of a video without subtitles may require uv and a transcription model.
2. Request a reading note
In a DSH conversation, for example:
Please read this paper in depth and save a Markdown reading note:
https://arxiv.org/abs/2307.09042
You can also provide a local file, web page, or video link. If needed, specify an output location, sections to retain, or whether you want timestamps.
For a shorter note, ask for “a concise reading note with key points, terminology, and quotations.” The main agent can select concise mode, which defaults to a minimum count of 500 and omits sections 5–7 while retaining source review. Ask explicitly if you also want critical analysis or the other optional sections. The minimum is not a length cap; the model determines the actual output length.
Hosts with plugin-tool support register deep_read_summarize; older hosts use the bundled skill and workflow entry point. The tool returns the note and a suggested path; the main agent must write the file. Check that the file was saved and whether the result reports gaps or quality issues.
When an identifiable note is extracted from a wrapper, the optional draftContext field records the extraction and retains any surrounding text. That text is not part of the note reviewed against the source.
{
"input": "https://arxiv.org/abs/2307.09042",
"type": "paper",
"options": {
"maxChunks": 4,
"outputDir": "./output"
}
}
type can be auto, book, paper, video, or web. Specifying it can skip classification for ambiguous inputs. The tool-call interface depends on the DSH host; asking in conversation is also supported.
Quality and limitations
The workflow normalizes chunk ranges, checks format and length, and has an independent review agent provide source and quotation evidence. The plugin tool also reads the acquired file in the host and checks its actual line count and the verbatim quotations extracted from section 4. An unreadable or changed source, or a quotation that does not match its line range, cannot pass verification. Results include qualityPassed, qualityChecks, qualityEvidence, and citationVerification.
sourceStatus is complete only when the acquired text exactly matches the user's local plain-text input; partial means a known incomplete source, and unknown means completeness cannot be established, including web pages, extracted PDFs, subtitles, and transcripts. The compatibility boolean sourceComplete is true only for complete; false covers both other states. The source status is retained in the note. Other checks may pass when the status is unknown, but that does not prove the original source was fully obtained.
citationVerification.status is verified, mismatch, unverified, or not_applicable. The legacy raw workflow entry cannot read files in the host; when citations require checking it reports unverified and fails quality. The literal check covers only quotations extracted from section 4 and their stated line ranges, not the correctness of other claims or paraphrases.
When quality checks fail, the saved note includes a notice, failed chunks, and up to five issues. The completion status in the default Obsidian and generic templates is changed to “pending review” when it can be identified safely. The complete issue list remains in qualityIssues.
These checks have limits:
qualityPasseddoes not prove factual accuracy. Semantic coverage and reasoning still depend on the review agent; a pass withsourceStatus: unknowndoes not prove acquisition completeness. Check important claims, numbers, and quotations yourself.- Retrieving the full text does not guarantee understanding it. Complex tables, formulas, images, or extraction errors can affect the result; incomplete sources or failed chunks are reported as gaps.
- Captions and transcripts can differ from audio. Automatic captions and speech recognition can omit words or mishear numbers and negations; checking the note against a transcript does not prove the transcript is accurate.
- Video timestamps cannot be inferred reliably. Locations depend on whether the obtained subtitles or transcript contain time information.
Common options
| Option | Default | Purpose |
|---|---|---|
lengthMode |
standard |
Set to concise for the preset: minWords: 500 and fastMode: true by default |
minWords |
2500 |
Minimum length; the check counts Chinese characters and words in other text |
fastMode |
false |
Omit the critical analysis, personal applications, and further-reading sections |
maxChunks |
4 |
Chunk-count limit, capped at 12; this is not the concurrency setting |
readConcurrency |
3 |
Maximum concurrent chunk reads; higher concurrency is not always faster |
maxRetries |
0 |
Repair attempts after a failed quality check; by default, a successfully drafted note receives one review and no repair |
outputDir |
./output |
Suggested note directory; the main agent performs the write |
requireCitations |
true |
Require verifiable source information for key quotations |
includeTimestamps |
false |
Request timestamps for video notes |
transcribe |
true |
Attempt local transcription when a video has no usable subtitles |
whisperModel / language |
small / zh |
Choose a transcription model and language; set en for English when appropriate |
device / cpuThreads |
cpu / 0 |
Select the transcription device and CPU thread policy |
transcriptCache |
true |
Reuse validated transcripts when the audio and configuration match |
parallelQuality |
false |
Opt in to parallel source review; the original review path remains the default |
readCache |
false |
Optional cross-run chunk cache; supported by the plugin tool only |
For a direct call, use options: { "lengthMode": "concise" }. Explicit minWords and fastMode values in that call override the preset; for example, "fastMode": false retains all sections. Concise mode aims near the minimum but does not impose a maximum length.
To enable readCache, also set readCacheIdentity to include the provider, pinned model version, and output-affecting settings; change it whenever those change. The caller supplies this identity, and the host cannot verify it automatically. Cache entries are local plaintext and may contain source excerpts; cache hits still receive independent source review.
Other options include frontmatter (obsidian, generic, or minimal) and tempDir. See the advanced reference for detailed options, supported formats, and transcription and cache behavior.
Time, cost, and privacy
- Time depends on content length, retrieval, chunk count, model, and host concurrency. A video without subtitles needs extra transcription; the first run may also download a transcription model.
- Cost depends on your model and input/output usage; the project cannot promise fixed time or cost.
- Privacy: speech transcription runs locally, but deep reading uses your configured DSH model. With a cloud model, source text, subtitles, notes, or related context may be sent to its provider; check its data-handling terms first.
- Local files: temporary source text, transcripts, and optional caches may remain on your device. Check directory permissions and cleanup before processing sensitive material.
Troubleshooting
- Plugin unavailable: check the DSH profile, restart Desktop after installation, and check host-version compatibility.
- Fetch failure or missing sections: check source accessibility and required tools; inspect the returned stage, source status, and failed chunks.
- Only a video description is available: a complete subtitle or transcript has not been confirmed. Check subtitle and transcription availability, or provide a transcript.
- Quality check failed: inspect
qualityIssuesandqualityEvidenceand review the source as needed. Do not treat an unverified note as a complete account.
For further help, open a GitHub issue with your operating system, DSH and plugin versions, input type, and error details. Remove secrets, private links, and personal data first.
Development and contributing
Work in a separate project directory; do not run npm inside a DSH profile:
git clone https://github.com/PensiveFei/deep-read-summarize.git
cd deep-read-summarize
npm ci
npm test
npm run test:node
npm run lint
npm run validate
The project uses parsers for different input types and supports custom parsers. See the repository, CHANGELOG, and CONTRIBUTING.
The reading-quality benchmark provides 16 short public-use cases, annotated reference points, and an offline scorer. Semantic quality requires human review; unreviewed results are not scores.
The relevant host contracts were previously verified with DSH 0.2.0-rc.2. DSH interfaces can change; verify other versions against the actual host and compatibility information.
License and disclaimer
Project code is available under the MIT License. Papers, video, and short quotations cited in examples remain subject to their respective rights and permissions.
deep-read-summarize is an independent third-party project. It is not affiliated with, endorsed by, or sponsored by DeepSeek or DeepSeek Harness.
Links
More in this category
loopx-project/loopx#dsh-loopx-plugin★ 6722
LoopX, a provider-neutral, local-first state kernel and control plane for long-horizon agents: keeps Goal, Todo, gate, evidence, quota, recovery, and handoff state above DeepSeek Harness, while the plugin bootstraps the CLI and skills, admits bounded same-session continuation, and adds a loopback GoalBar for the exact bound loop.
Q00/ouroboros#integrations/dsh-plugin★ 6381
Config-only bundle that mounts Ouroboros through the DSH MCP client, exposing 36 interview, Seed, execution, evaluation, and evolution workflow tools in DSH.
chuspeeism/dashi-taskboard#deepseek-harness★ 3507
Embeds the active installed Codex Taskboard runtime in the DeepSeek Harness sidebar, using its launcher runtime descriptor instead of a fixed port.
NanmiCoder/dsh-agent-teams★ 3026
AgentTeams multi-agent teams.
EthanYoQ/AI-Novel-Writer#dsh-ai-novel-writer★ 2711
Installs a dedicated AI novel-writing preset and workbench: revisioned local project assets, a compact side drawer, and native approval-gated single-file changes.
shanliuling/dsh-image-gen★ 1626
Native conversational image generation for DeepSeek Harness: ask the agent to create an image, and it handles generation and keeps the result directly in the conversation.
Community comments
Comments are public GitHub Discussions. Loading them connects to GitHub and Giscus; a GitHub account is required to post.