poplarity/dsh-science-workbench
A reproducible science workbench plugin for the DeepSeek Harness: agent-driven cells, inline figures with feed
dsh-science-workbench is a reproducible science workbench plugin for DeepSeek Harness blending Jupyter cells and inline figures, Claude Science-style agent-driven execution, and Nextflow/nf-core-style provenance. Nine agent-facing tools (bio_init_project, bio_run_cell, bio_rerun_cell, bio_add_feedback, bio_get_project, bio_list_projects, bio_set_projects_dir, bio_delete_cell, bio_mark_cell) plus a three-panel Analysis workbench tab with inline figure preview, cell search, native directory picker, and mark-as-final badges. Every figure and artifact is traceable and replayable: a plain-text manifest.json is the single source of truth for cells, artifacts, provenance, and feedback history. It also bundles two publication-grade figure skills adapted from Claude Science (Apache-2.0): figure-style and figure-composer.
Install
dsh plugin --profile web add dsh-science-workbenchnpm package dsh-science-workbench 0.2.0 (registry-verified 2026-08-24; README badge matches). Install: dsh plugin --profile web add dsh-science-workbench, then restart dsh web. The package declares dsh.bundle.patch, so dsh plugin automatically adds it to dsh.profile.bundles. Local development: dsh plugin --profile web add file:/path/to/dsh-science-workbench. Dual-face plugin (Host + Client): the bio_* tools become globally available and the Analysis workbench tab appears; the plugin shows up under Settings → Plugins.
Compatibility
DeepSeek Harness (dual-face Host + Client plugin). Cross-platform: Host shell speaks bash on macOS/Linux and PowerShell on Windows; Python resolves to python on Windows and python3 on POSIX. Reproducible by construction: self-contained scripts, fresh subprocess per cell, environment.lock, SHA-256 input/output hashes, fixed seed; each project is git init-ed on creation and auto-committed at every step (never pushed).
Details
- Repo: poplarity/dsh-science-workbench
- Category: Web UI Enhancements
- Stars: 7
- Version: npm package dsh-science-workbench 0.2.0 (registry-verified 2026-08-24)
- Last push: 2026-08-26
- First seen: 2026-08-14
Recent updates
The current English README documents: the core promise (traceable, replayable artifacts), features, the nine tools, the Analysis workbench tab, the bundled figure skills with ATTRIBUTIONS.md, install (npm + local), and quick start with the manual tool flow.
FAQ
- How does provenance work?
- Each project keeps a plain-text manifest.json as the single source of truth: cells, artifacts, provenance (producing cell, SHA-256 output hash, params, seed, derived-from, created time), and feedback history. Projects are git init-ed on creation and auto-committed at every step (never pushed).
- Can I ask the agent to redraw a figure?
- Yes — attach structured feedback to a figure with bio_add_feedback, and bio_rerun_cell regenerates a derived version (v1 → v2 → v3) with lineage recorded in the ledger.
- Does it work on Windows?
- Yes — the Host shell layer speaks bash on macOS/Linux and PowerShell on Windows; Python resolves to python on Windows and python3 on POSIX.
Alternatives
biociao/dsh-science · poplarity/dsh-science-workbench · shuguang1994/project-blueprint