Abstract Plugins Index
English | 中文 · Repository
An open-source curated index that lives inside DeepSeek Harness: the most absurd plugins in the DSH ecosystem, each with a published score breakdown, one editorial verdict, and native install.
It is not another plugin market — the host already has Discover and the market to answer "what exists and can it be installed". This answers a different question: which plugins are the most unhinged, and why.
What it looks like
After installing, the Plugins page gains an Abstract tab:
- A ranked list of entries showing theme, package, stars, and date added
- An expandable score breakdown: absurdity / emotional value / craft, each 0–100, weighted 50 / 30 / 20 — the number is reproducible
- One verdict per entry: why it counts as abstract, not a feature recap
- Native install: the Install button opens the host's own install flow; the CLI command is copyable too
- A published standard: scope, formula, and the excluded entries with reasons — in the tab footer and in this repository
- Provenance: facts come from awesome-dsh-plugin (CC0), and the tab shows the upstream commit
Install
After the npm release:
dsh plugin --profile <profile> add dsh-abstract-index
To develop locally, install this directory through the host's plugin manager (absolute path):
plugin_manager → action: install_bundle → target: /absolute/path/dsh-abstract-index
Then refresh the page and open Plugins → Abstract. On hosts without the plugins.tab slot (older versions) the tab does not appear and nothing fails.
Curation and scoring
Scope: every entry in the upstream
funcategory — 58 today — judged one by one.Inclusion: some entries are excluded (a plain music player, a skin, a vocabulary app…) and the reasons are shown. Inclusion is an opinion, open to argument.
Three axes (0–100 each):
Axis Meaning absurdity how far the premise sits from a coding tool emotion how much it changes the waiting and the mood craft visible product polish in the description Formula:
score = round(absurdity×0.5 + emotion×0.3 + craft×0.2)Scores are computed at build time from the three factors in
data/curation.json; they are never hand-written, so they stay reproducible.
Contributing
Scores, verdicts and inclusion decisions live in data/curation.json; the workflow and PR checklist are in CONTRIBUTING.md. To argue about a score, a verdict, or an exclusion, open an issue with a specific reason.
Where the data comes from
Two committed inputs:
| File | Contents | Owned by |
|---|---|---|
data/snapshot.json |
Upstream facts: URL, name, bilingual description, npm package, stars, added date, upstream commit | npm run sync (generated; do not edit) |
data/curation.json |
Editorial layer: inclusion, theme, three-factor scores, verdicts, exclusion reasons | humans / pull requests |
The build joins them into the browser artifact:
npm install # js-yaml, needed only by sync
npm run sync -- --upstream /path/to/awesome-dsh-plugin
npm run check # validate → build → node --check
client.js is a committed build artifact so the bundle can be installed from git; CI proves that rebuilding produces no diff.
Repository layout
dsh-abstract-index/
package.json # bundle manifest + dsh.client section
cordis.patch.yml # inserts the abstract-index row
index.js # Host half: no behaviour, inventory entry only
client.js # built artifact (do not edit)
src/client.js # the Abstract tab UI
data/snapshot.json # upstream facts (generated)
data/curation.json # editorial layer (human)
scripts/sync.mjs # snapshot generator
scripts/build-client.mjs# data + UI -> client.js
scripts/validate.mjs # data checks
scripts/lib/data.mjs # shared model: merge, scoring
locale/{zh,en}.json # display title/description for the plugin manager
icon.svg
Boundaries
- No network, no session reads, no tools, no billing; the catalog is static data baked into the artifact.
- It does not install for the host: the button calls the host's own install flow.
- It is not a security review. Read the repository README before installing anything.
- Entry text belongs to its authors; facts are CC0; this repository's code is MIT.
License
- Code: MIT, see LICENSE
- Data and editorial text: CC0-1.0, see LICENSE-DATA
No comments yet. Be the first to write one.