zuoyunlai/lunheng-article-pipeline-dsh
Lunheng turns multi-agent long-form writing into a nine-role pipeline with human checkpoints, packaged as a DeepSeek Harness bundle.
Lunheng is a long-form writing pipeline rather than a text generator: it splits a deliverable (academic papers, industry analysis, business commentary, long articles) into nine independent roles (T1–T9) across six phases, orchestrated with DSH subagent calls, with human checkpoints to confirm the outline before drafting and review the final draft. The README lists the artefacts it produces — graded literature, data and case cards, an analysis outline mapping claims to evidence, successive drafts with AI-trace cleanup, review reports and final deliverables — and says the bundle adds four read-only tools in DSH (lunheng_m_gate, lunheng_char_count, lunheng_handoff_check and one more). The lunheng-commands skill exposes 11 /lunheng-* slash commands for drafting, resuming, citing, auditing and rolling back.
Install
dsh plugin --profile web add lunheng-article-pipelineCopied verbatim from the README's install line. The README notes the bundle registers two skills (the nine-role lunheng-article-pipeline and the lunheng-commands slash-command wrapper), and that copying a plain skill directory is not supported because a directory carries no dsh.bundle declaration. Removal is dsh plugin --profile <profile> remove lunheng-article-pipeline.
Compatibility
The README's Requirements table lists Node ^22.19.0 || >=24.0.0 (the DSH runtime floor, per package.json engines) and pnpm, which is required because dsh plugin delegates to pnpm. The README's 'When not to use it' section is explicit that Lunheng collects and integrates published evidence only: it cannot do first-hand data collection, statistical analysis (SPSS/R/Python), raw chart-data scraping, original image/video generation, or code execution beyond a whitelisted set.
Details
- Repo: zuoyunlai/lunheng-article-pipeline-dsh
- Category: Workflow & Automation
- Stars: 9
- Version: GitHub repo zuoyunlai/lunheng-article-pipeline-dsh (MIT, per the README's License section); npm package lunheng-article-pipeline at 18.80.0 (registry check 2026-10-07); the README states version v18.80.1 for the DSH bundle.
FAQ
- Can Lunheng collect its own data?
- No. The README says it collects and integrates published evidence only; first-hand data collection, statistical analysis, OCR and original media generation are out of scope, so supply that material first.
- Does it run unattended?
- Not fully — the README describes human checkpoints: you confirm the outline before drafting and review the final draft, and the pipeline deliberately states no fixed hour range.
Alternatives
cloader/dsh-taskboard · btspoony/dsh-advisor · apheli0os/deepseek-harness-orchestrate