Qiongli

English

简体中文

English

简体中文

Appearance

QiongliContract-bound research workflows for AI coding agents.

Install once, then use Codex, Claude Code, or Gemini to run academic workflows with explicit task IDs, quality gates, literature diagnostics, role handoffs, and auditable artifacts.

Start With Quickstart
Choose A Workflow
Install
Qiongli logo

Install Without Guesswork

Choose native plugin, bootstrap partial/full, npm, or pipx based on what runtime surface you actually need.

Research Routes

Start from systematic review, empirical, qualitative, RCT, theory, or code-first methods workflows.

Evidence Contracts

Keep claims, citations, search logs, diagnostics, methods, code, and review status traceable under standard artifact paths.

Multi-Agent Control

Use solo, duo, and triad execution modes with explicit handoffs, disagreement records, and verification outcomes.

Choose Your Entry Point ​

You want to...Start hereWhy
Try Qiongli in one clientInstallNative plugin / extension paths keep setup small.
Know what to type after installUsing Agent SkillsCodex, Claude Code, Gemini, and shell expose Qiongli differently.
Install global workflows for several clientsQuickstartBootstrap partial installs workflow assets without requiring Python.
Run validators, doctor, or orchestrated tasksMulti-Agent Runtimefull runtime explains Python, model CLIs, auth, broker/direct Gemini modes, and verification.
Pick a paper workflowTask RecipesMaps real research goals to paper types, stages, Task IDs, and expected outputs.
Automate installs or upgradesCLI ReferenceCovers qiongli, ql, npm/npx, pipx, compatibility aliases, and JSON checks.

What The Current System Covers ​

Qiongli ships a single portable workflow package, qiongli-workflow, with staged research skills and a shared task contract. The current documentation is organized around what a researcher or project owner needs to do:

  • Frame the work: refine questions, identify gaps, map theory, choose venues, and define contribution claims.
  • Build the literature base: plan provider-aware searches, materialize search bundles, run diagnostics, deduplicate results, screen papers, extract evidence, and snowball citations.
  • Design and execute the study: specify variables, datasets, robustness checks, preregistration, ethics artifacts, and data management.
  • Write and audit the manuscript: structure sections, maintain claim-evidence integrity, generate figures/tables, evaluate limitations, and prepare submission/rebuttal materials.
  • Handle research code: use the Stage-I specification -> planning -> execution -> review path for code-first or methods-heavy work.
  • Coordinate models: assign controller, primary, reviewer, and verifier roles across Codex, Claude Code, and Gemini while preserving handoffs and verification status.

Documentation Map ​

  • Guide: operational path for installation, usage, upgrades, troubleshooting, and runtime choices.
  • Using Agent Skills: client-by-client invocation rules, including Codex /skills and $qiongli.
  • Task Recipes: scenario-driven routes for paper types and common research goals.
  • Examples: concrete playbooks for systematic review, empirical, qualitative, methods, and theory papers.
  • Reference: CLI behavior, skill catalog, and operator-facing conventions.
  • Architecture: how contracts, skills, roles, pipelines, bridges, and package surfaces fit together.
  • Advanced: subject packaging, extension, MCP providers, Zotero, rigorous literature search, and plugin-first distribution.
  • Maintainer: release policy, naming policy, and implementation guidance for contributors.

Runtime Boundary ​

Asset installation and workflow discovery are intentionally lighter than full orchestration. You can install qiongli-workflow without Python, but doctor, validators, tests, and model-orchestrated task execution require Python 3.12+, the relevant model CLIs, and matching authentication.

Qiongli documentation

MIT License