qdd: A Research Workflow Orchestration Plugin for DeepSeek Harness
billychen123/qdd
QDD turns long-horizon research into an auditable question-driven discovery loop, with six human workflows and auto mode, bringing research orchestration to DeepSeek Harness.
Turn DeepSeek Harness into an auditable research loop where agents and humans share a readable research state, optimizing for the next question rather than just the next task.
dsh plugin --profile web add github:billychen123/qdd- Category
- Workflow & Automation
- Platform
- DSH-Plugin
- Author
- billychen123
- Distribution
- Plugin
qdd Key Features
qdd Repository
qdd Repo Summary
QDD (Question-Driven Discovery) is a research-workflow orchestration plugin (DSH plugin) for DeepSeek Harness, maintained by BillyChen123 under the Apache-2.0 license. It turns long-horizon biomedical discovery into an auditable loop of questions, evidence, artifacts, and next-study decisions. The plugin addresses the pain point of maintaining coherence in multi-step scientific projects when partial results, failed hypotheses, or data limitations arise—humans and agents share a readable research state instead of scattered chats and scripts. Core capabilities include question governance (recording refinement, confirmation, pivot, or dissolution), agent-ready memory (contracts, study files, task records, and evolution history), domain skill injection (34 local skills routed by role and task), and public-data grounding (CELLxGENE, GEO, PubMed, etc. become auditable local artifacts).
What core features does the QDD DSH plugin support?
QDD provides five research-loop workflows plus a project-level conclude workflow. Start establishes the project contract (theme, scope, data assumptions, runtime environment); Propose turns the current frontier into a bounded study with a judgeable question, falsifiable expectation, and task graph; Explore stress-tests the study design before execution, refining boundaries and deciding whether public data is needed; Apply executes study tasks, injecting task-local domain skills and running code inside the project; subsequent workflows handle validation and conclusion, turning negative results into pivots, validations, or robustness studies.
How do I install the QDD DSH plugin?
QDD requires Node.js 20.19.0 or higher and recommends an Anthropic-compatible model configuration for Auto Mode. Install via the DSH plugin system:
dsh plugin --profile web add github:billychen123/qdd
After installation, initialize a research project and start Auto Mode:
mkdir my-qdd-project
cd my-qdd-project
qdd init .
qdd auto --max-turns unlimited
How does the QDD DSH plugin integrate with DeepSeek Harness?
As a DSH plugin, QDD registers with DeepSeek Harness by exporting apply(ctx), leveraging the framework's models, tools, and memory capabilities. It emphasizes a readable research state shared by humans and agents, domain skills injected at the right time rather than dumped into every prompt, and public-data searches recorded as reusable evidence. This design lets DeepSeek Harness agents optimize for the next question rather than just the next task in long-horizon research.
What are the compatibility and licensing details of the QDD DSH plugin?
QDD is maintained by BillyChen123 under the Apache-2.0 license. The current version is v0.1.0-rc.1, a public submission-candidate line that accepts reproducibility fixes and documentation improvements. The project was created on 2026-05-27 and last updated on 2026-08-19, with 5 stars, 1 fork, and no open issues. It requires Node.js 20.19.0+ and recommends an Anthropic-compatible model for Auto Mode.