# How to Install and Run the Full ARS Pipeline: Complete Setup Guide

> Install the full ARS pipeline using the Claude Code plugin or symlink. Follow this guide to launch the 10-stage academic research workflow and streamline your publication process.

- Repository: [Edward Cheng-I Wu/academic-research-skills](https://github.com/Imbad0202/academic-research-skills)
- Tags: getting-started
- Published: 2026-05-13

---

**Install the Academic Research Skills (ARS) suite via the Claude Code plugin marketplace or classic symlink method, then execute `/ars-full` to launch the 10-stage research-to-publication workflow.**

The Academic Research Skills (ARS) suite is an open-source Claude Code plugin maintained at `Imbad0202/academic-research-skills` that automates academic paper production through a rigorous, multi-agent pipeline. When you install and run the full ARS pipeline, you trigger a complete workflow from research question formulation through peer review to final PDF generation, all within Claude Code's environment.

## Prerequisites

Before you can install and run the full ARS pipeline, ensure your environment meets these requirements:

- **Claude Code (latest)** – Provides the runtime for Claude-powered skills
- **Anthropic API key** – Set as `ANTHROPIC_API_KEY` environment variable for all LLM calls
- **Pandoc** (optional) – Enables DOCX export functionality
- **Tectonic + Source Han Serif TC** (optional) – Required for LaTeX-to-PDF rendering in APA 7.0 format

The pipeline functions with Markdown output alone; PDF and DOCX capabilities depend only on these optional components.

## Installation Methods

There are two first-class installation paths for the ARS suite. Both install the same skill package containing four registered skills: `deep-research`, `academic-paper`, `academic-paper-reviewer`, and `academic-pipeline`.

### Plugin Installation (Recommended, v3.7.0+)

The plugin method uses the manifest defined in [`.claude-plugin/plugin.json`](https://github.com/Imbad0202/academic-research-skills/blob/main/.claude-plugin/plugin.json) to register all skills and slash-commands under `commands/` in a single step:

```text
/plugin marketplace add Imbad0202/academic-research-skills
/plugin install academic-research-skills

```

This approach automatically configures the `ars-full` and `ars-plan` commands without manual file manipulation.

### Classic Symlink Installation

For Claude Code versions that support manual skill loading, clone the repository and symlink the skill definitions:

```bash
git clone https://github.com/Imbad0202/academic-research-skills.git
ln -s $(pwd)/academic-research-skills/skills/* ~/.claude/skills/

```

All skill definitions live under the `skills/` directory. After symlinking, restart Claude Code or run `claude reload` to load the skills.

## Verify Your Installation

Confirm the skill is active by running the lightweight verification command:

```text
/ars-plan

```

You should see a Socratic dialogue that begins building a paper outline. If this initiates successfully, the skill is ready for the full pipeline execution.

## Run the Full ARS Pipeline

The slash-command that orchestrates the complete workflow is **`/ars-full`**. When invoked, it executes the following 10 stages:

1. **Stage 1: RESEARCH** – Socratic-guided research question formulation and methodology blueprint creation using the `deep-research` skill
2. **Stage 2: WRITE** – Outline approval, draft generation, citation handling, and figure verification via the `academic-paper` skill
3. **Stage 2.5: INTEGRITY** – Machine-checked 7-mode failure checklist using the `integrity_verification_agent`
4. **Stage 3: REVIEW** – Multi-agent peer review (EIC + 3 reviewers + Devil's Advocate) using the `academic-paper-reviewer` skill
5. **Stage 3→4: Revision Coaching** – Optional Socratic rounds to refine reviewer feedback
6. **Stage 4: REVISE** – Author-driven revision with delta report and score-trajectory tracking
7. **Stage 3′: RE-REVIEW** – Narrow re-review of the revised manuscript
8. **Stage 4′: RE-REVISE** – Final revision freeze
9. **Stage 4.5: FINAL INTEGRITY** – Zero-tolerance re-run of the 7-mode checklist
10. **Stage 5: FINALIZE** – Format selection (MD/DOCX/LaTeX/PDF) and AI-disclosure generation
11. **Stage 6: PROCESS SUMMARY** – Collaboration-depth report, AI self-reflection, and complete process record

Execute the pipeline with:

```text
/ars-full

```

All stages are **user-gated** with decision-heavy checkpoints except the two integrity gates (Stages 2.5 and 4.5), which are machine-verified and then acknowledged by the user.

## Under the Hood: Pipeline Architecture

The full pipeline orchestrates multiple specialized agents across three core skills:

| Layer | Skill | Core Agents | Key Artifacts |
|-------|-------|-------------|---------------|
| **Research** | `deep-research` | 13 agents including `research_question_agent`, `bibliography_agent` | RQ brief, methodology blueprint, annotated bibliography |
| **Write** | `academic-paper` | 12 agents including `intake_agent`, `structure_architect_agent` | Outline, draft, bilingual abstract, figures |
| **Integrity** | `academic-pipeline` | `integrity_verification_agent` | Material Passport with `repro_lock`, claim-verification report |
| **Review** | `academic-paper-reviewer` | 7 agents (EIC, domain, methodology, etc.) | Full review package, editorial decision |
| **Finalize** | `academic-paper` | `formatter_agent` | Publication-ready outputs |

The complete stage matrix and data-flow diagrams are documented in [`docs/ARCHITECTURE.md`](https://github.com/Imbad0202/academic-research-skills/blob/main/docs/ARCHITECTURE.md), specifically in Section 3, which defines the single source of truth for the pipeline structure.

## Customizing Your Run

Modify pipeline behavior using these environment variables and commands:

- **Cross-model verification** – Set `ARS_CROSS_MODEL=1` to enable a secondary model double-check at integrity gates
- **Citation format conversion** – Issue `/ars-citation-check IEEE` after `/ars-full` to convert to IEEE style
- **Resume from checkpoint** – Enable `ARS_PASSPORT_RESET=1` and provide `resume_from_passport=<hash>` to resume from a saved Material Passport (documented in the *passport reset* protocol)

All optional flags are detailed in [`docs/SETUP.md`](https://github.com/Imbad0202/academic-research-skills/blob/main/docs/SETUP.md).

## Summary

- **Install** the ARS suite via plugin (`/plugin install academic-research-skills`) or symlink (`ln -s` to `~/.claude/skills/`)
- **Verify** installation by running `/ars-plan` to trigger the Socratic planning dialogue
- **Execute** the full pipeline with `/ars-full`, which implements the complete 10-stage workflow defined in [`commands/ars-full.md`](https://github.com/Imbad0202/academic-research-skills/blob/main/commands/ars-full.md)
- **Stages** progress from research question formulation through integrity checks, peer review, revision cycles, and final publication formatting
- **Output** formats include Markdown (default), DOCX (requires Pandoc), and PDF (requires Tectonic)

## Frequently Asked Questions

### What is the difference between the plugin and symlink installation methods?

The plugin method (v3.7.0+) uses `/plugin marketplace add` to automatically register all skills via [`.claude-plugin/plugin.json`](https://github.com/Imbad0202/academic-research-skills/blob/main/.claude-plugin/plugin.json) without manual file manipulation. The symlink method requires cloning the repository and manually linking the `skills/` directory to `~/.claude/skills/`, offering compatibility with older Claude Code versions. Both methods provide identical functionality once installed.

### Why does the pipeline stop and ask for confirmation at each stage?

The ARS pipeline implements **user-gated checkpoints** at every major transition (except the machine-verified integrity gates at Stages 2.5 and 4.5). This design ensures you maintain authorship control over research direction, methodology approval, outline confirmation, and revision decisions. Simply respond to the prompts when the system requests confirmation.

### Can I run the pipeline without installing Pandoc or Tectonic?

Yes. The pipeline generates publication-ready Markdown by default. **Pandoc** is only required if you need DOCX export, and **Tectonic** with Source Han Serif TC is only needed for APA 7.0 PDF rendering. You can select your output format at Stage 5 (FINALIZE); if you choose PDF without Tectonic installed, the system will notify you of the missing dependency.

### How do I resume a pipeline run if my session interrupts?

Enable the `ARS_PASSPORT_RESET=1` environment variable and provide the `resume_from_passport=<hash>` parameter, where the hash corresponds to your saved Material Passport's `repro_lock` identifier. This protocol, documented in [`docs/ARCHITECTURE.md`](https://github.com/Imbad0202/academic-research-skills/blob/main/docs/ARCHITECTURE.md), allows you to restart from any checkpoint without losing progress on your researchquestion, bibliography, or draft revisions.