Foreword

In the DSH “everything is a plugin” organizational structure, tools can be mounted to specific profiles as bundles, and combined with agent presets to form usable session capabilities. @amphilagus/dsh-literature is a DSH plugin maintained by amphilagus, positioned as a literature retrieval and targeted tracking plugin for DeepSeek Harness, complete with a corresponding “Literature Tracking Assistant” agent preset.

It targets not general search, but literature workflows: searching Crossref and arXiv by topic or researcher, writing to a global new library after manual screening and curation, and achieving expiration reminders and continuous tracking in selected preset sessions. Standard coding Agents do not load this toolset by default; only sessions with the “Literature Tracking Assistant” preset selected will enable it.

This plugin is listed on the skillhub.cn plugins page. This directory is an independent site with no official affiliation with DeepSeek / Hypi, and should not be regarded as an official app store.

What is it

@amphilagus/dsh-literature can be summarized into three things:

  1. Provide literature retrieval tools for agents.
  2. Provide data management for researchers and tracking schemes.
  3. Coordinate with the “Literature Tracking Assistant” preset to allow sessions to load runtime skills based on tasks.

Basic Information:

  • Plugin Name: @amphilagus/dsh-literature
  • Maintainer: amphilagus
  • License: MIT
  • Node Requirement: ^22.19.0 || >=24.0.0
  • package.json declaration: private: true
  • Companion Preset: “Literature Tracking Assistant”

It does not perform full-text scraping nor translation. The reminder mechanism relies on the current session process: it does not trigger while the session is closed, but will resend overdue tasks upon reopening.

Core Functions

Below is an introduction to the verified scope of the plugin’s capabilities.

Retrieval

  • Crossref remote search.
  • Local queries only hit the global new library where items have already been screened and curated.
  • DOI details are checked in the new library first; if not curated, it proceeds to Crossref.

Tracking

  • Search by research topic, with an optional ISSN whitelist.
  • Search by researcher, who must provide an ORCID.
  • Periodically search Crossref and arXiv.
  • Write to the global new library after manual grading.

Auxiliary Data

  • SCI journal directory, fields include journal name, ISSN, impact factor, and CAS subject categories.
  • Cross-session researcher profiles.
  • Local expiration reminders for the session.

Runtime Skills

The plugin registers three runtime skills:

  • literature-survey
  • literature-tracking-setup
  • literature-tracking-search

These skills are runtime skills registered by the plugin while it is enabled, not files that need to be placed separately in the skills directory.

Data Management

Data management covers:

  • Search cache.
  • Global new library.
  • Tracking schemes.
  • Researcher profiles.
  • Search logs.

Installation & Enablement

Prerequisites

Below assumes the profile name is web. If DSH_HOME is set, the relevant data directory will fall under $DSH_HOME.

Confirm the following before installing:

  1. The local Node meets ^22.19.0 || >=24.0.0.
  2. Plugin source code, dependencies, and MIT license have been reviewed.
  3. It has been confirmed that this plugin will run with the current dsh process permissions.
  4. Prepare a session where you will select the “Literature Tracking Assistant” preset, rather than a standard coding session.

Only plugin installed, no preset: All tools are disabled.
Only preset installed, no plugin: The @amphilagus/dsh-literature line in the preset fails to resolve, and the session cannot start.

Installing the Plugin

Build first, then link into the profile:

cd dsh-literature
pnpm install
pnpm run build
./scripts/install-profile.sh web

Or use the link installation command:

dsh plugin --profile web add "link:/absolute/path/to/dsh-literature"

This step mounts the plugin to the target profile. At this point, even without the companion preset, literature tools will still not appear in ordinary sessions.

Installing the schedule package

Literature tracking requires schedule capability. The installation steps require installing the schedule package. The methods provided in the documentation include:

dsh plugin --profile web add "link:/absolute/path/to/deepseek-harness/packages/schedule/schedule"

Or:

dsh plugin --profile web add @deepseek-ai/dsh-schedule

This step ensures that the profile’s node_modules contains the schedule dependency, allowing the preset to correctly parse the schedule tool.

Placing the preset

The preset directory name is the preset ID; in the documentation, it is literature-tracking-assistant:

mkdir -p ~/.dsh/.agent-presets
cp -R preset ~/.dsh/.agent-presets/literature-tracking-assistant

You should get:

~/.dsh/.agent-presets/literature-tracking-assistant/preset.yml
~/.dsh/.agent-presets/literature-tracking-assistant/agent.cordis.yml

preset.yml handles the display name “Literature Tracking Assistant” on the picker. If you want to change the persona, mailto, or whether to mount schedule, you edit the copied agent.cordis.yml, not the source file in the repository.

After modifying the preset file, new sessions will use the new configuration; running sessions remain with the version at startup. Existing sessions cannot switch presets mid-way; a new session must be opened and selected.

Restart and Verification

The web profile only takes effect after a restart:

dsh --profile web

Verification order:

  1. Create a new session and select “Literature Tracking Assistant”.
  2. In the session, you should see literature retrieval, tracking, and schedule-related tools, such as literature_search, tracking_*, researcher_profile_*, schedule_create.
  3. The skill list should contain literature-survey, literature-tracking-setup, literature-tracking-search.
  4. Open another standard mode session: literature tools and schedule tools should not appear.

Typical Usage

The following explains the three paths. You can simply state the requirements directly in a “Literature Tracking Assistant” session, and the agent will load the corresponding skills based on the task.

Search for recent articles on a topic

This path only performs retrieval and curation; it does not create tracking schemes or schedules.

  1. Load literature-survey.
  2. Call literature_search with topic keywords and recentDays.
    - Last week: 7
    - Last month: 30
  3. After screening, use tracking_curate to write to the global new library.

Search for a researcher

This path also only performs retrieval and curation.

  1. Load literature-survey.
  2. If ORCID is available, call literature_search directly with orcid.
  3. If only a name is provided, first call researcher_profile_disambiguate to confirm identity before searching.
  4. You can similarly use recentDays to limit the window.
  5. After screening, write to the global new library.

Establish continuous tracking

This path creates a tracking scheme and creates a schedule.

  1. Load literature-tracking-setup.
  2. Confirm the tracking direction: topic or person.
  3. If it is person, first disambiguate and create a profile.
  4. If the topic is topic, you can configure an ISSN whitelist.
  5. Call tracking_plan_add.
  6. Call schedule_create.

Note: The period for schedule_create is every_seconds, with a minimum of 300 seconds, not a calendar-based “every Monday at 9 AM”.

Execute tracking on expiry

When the expiration reminder is triggered or manually requested:

  1. Load literature-tracking-search.
  2. Perform dual-source search.
  3. Manual grading.
  4. Write to the global new library.
  5. Report after calling tracking_log_complete.

Reminders are only triggered while the current session process is alive. They do not trigger while closed, but will resend overdue tasks upon reopening.

Data, Configuration & Permissions

Data Location

The data paths in the documentation are:

$DSH_HOME/data/literature

Which contains:

$DSH_HOME/data/literature/literature.db
$DSH_HOME/data/literature/sci_journals.db

literature.db is used for search cache, global new library, tracking schemes, researcher profiles, and search logs. sci_journals.db is used for the SCI journal directory.

Sandbox Permissions

DSH with sandboxPolicy.grant registers $DSH_HOME/data/literature as an additional writable root, allowing bash/fs to access backup and export.

For older DSH without grant, the plugin can still load, but the shell may not be able to write. The literature tools themselves do not use sandbox.

Common Configuration

Plugin configuration should be written to the preset’s remount, not the host’s shell config. Edit the file:

~/.dsh/.agent-presets/literature-tracking-assistant/agent.cordis.yml

Example provided in the documentation:

- id: literature-search
  name: '@amphilagus/dsh-literature'
  config:
    enabled: true
    mailto: you@example.com
    # dbPath: ~/custom/literature.db
    # cacheRemote: true

Common configuration items:

  • enabled: Main switch. Must be true in this preset.
  • mailto: Crossref polite pool email.
  • dbPath: SQLite path, overriding search cache, new library, and tracking schemes.
  • cacheRemote: Whether to write remote hits to the search cache.
  • sciJournalsPath: SCI directory path.

After modifying the preset file, new sessions will use the new configuration; running sessions remain with the version at startup.

Applicable Scenarios & Notes

Suitable for the following situations:

  • Need to perform topic literature retrieval within a DSH session.
  • Need to track researchers by ORCID.
  • Need to consolidate manually filtered results into a global new library.
  • Need session-level expiration reminders and continuous tracking.

Not suitable or requires attention:

  1. When full-text scraping or translation is required, this plugin does not provide these capabilities.
  2. Standard coding Agents do not load this toolset by default; the “Literature Tracking Assistant” preset must be selected.
  3. Skills are registered at runtime; do not copy literature-survey, literature-tracking-setup, literature-tracking-search into the skills directory.
  4. The schedule period is every_seconds with a minimum of 300 seconds, not a specific calendar time point.
  5. The plugin runs with the current dsh process permissions; source code, dependencies, and license should be reviewed before installation.

Conclusion

The value of @amphilagus/dsh-literature is quite clear: combining Crossref, arXiv, local SQLite data, researcher profiles, and schedule reminders into a DSH plugin, and coordinating with the “Literature Tracking Assistant” preset to form an executable literature workflow.

Directory page: https://www.skillhub.cn/plugins/amphilagus/dsh-literature

Source code: https://github.com/amphilagus/dsh-literature