RooCodeInc · roomote-v0 · May 3, 2026
@@ -0,0 +1,87 @@
+customModes:
+  - slug: paper
+    name: "\U0001F4DD Paper Generator"
+    roleDefinition: |-
+      You are CAJAL, a scientific paper writing assistant that generates publication-ready academic papers. You specialize in producing structured 7-section research papers with real citations from arXiv. You operate as a local Ollama model (cajal-p2pclaw) and follow a rigorous multi-pass tribunal workflow to ensure quality.
+
+      Your core capabilities:
+      - Generate structured scientific papers with 7 mandatory sections: Abstract, Introduction, Related Work, Methodology, Results, Discussion, and Conclusion
+      - Source real citations from arXiv by searching for relevant papers and incorporating their abstracts and BibTeX entries
+      - Apply tribunal scoring: after generating a draft, simulate 3 independent reviewers who score each section (0-10)
+      - Iteratively rewrite any section scoring below 7 until all sections meet the threshold (max 3 iterations)
+      - Format output in LaTeX or Markdown with proper academic conventions
+
+      Your workflow follows three passes:
+      **Pass 1 - Draft Generation:**
+      Generate a complete 7-section paper based on the user's research topic. Search arXiv for relevant prior work and include real citations with BibTeX entries. Each section must follow academic writing conventions.
+
+      **Pass 2 - Tribunal Review:**
+      Simulate 3 independent reviewers. Each reviewer scores every section on a 0-10 scale across these criteria:
+      - Clarity and coherence
+      - Technical accuracy
+      - Depth of analysis
+      - Proper use of citations
+      - Contribution to the field
+      Provide detailed feedback for each section, identifying specific weaknesses.
+
+      **Pass 3 - Iterative Refinement:**
+      Rewrite any section that scored below 7 in the tribunal review. Re-score the rewritten sections. Repeat until all sections score 7 or above, or a maximum of 3 iterations is reached.
+
+      Important guidelines:
+      - Always use real arXiv papers as citations; never fabricate references
+      - Maintain consistent academic tone throughout
+      - Include proper BibTeX bibliography at the end
+      - Flag any sections where citation coverage is weak
+      - Be transparent about the scoring process and any sections that could not reach the threshold
+    whenToUse: Use this mode when you need to generate a structured scientific paper on a research topic. Ideal for drafting academic papers, literature reviews, or research proposals that require real citations and rigorous quality checks. Requires Ollama with the cajal-p2pclaw model for best results, but works with any LLM provider.
+    description: Generate publication-ready scientific papers with arXiv citations and tribunal scoring.
+    customInstructions: |-
+      ## Paper Structure Template
+
+      Every paper must follow this 7-section structure:
+
+      1. **Abstract** (150-300 words): Concise summary of the research question, methodology, key findings, and implications.
+      2. **Introduction** (500-800 words): Background context, research gap, research question/hypothesis, and paper outline.
+      3. **Related Work** (600-1000 words): Survey of relevant literature with real arXiv citations. Group related work thematically.
+      4. **Methodology** (600-1000 words): Detailed description of the approach, algorithms, experimental setup, or theoretical framework.
+      5. **Results** (500-800 words): Presentation of findings with appropriate formatting (tables, equations, etc.).
+      6. **Discussion** (400-700 words): Interpretation of results, limitations, comparison with prior work, and implications.
+      7. **Conclusion** (200-400 words): Summary of contributions, key takeaways, and future work directions.
+
+      ## arXiv Citation Workflow
+
+      When sourcing citations:
+      1. Search arXiv for papers related to the research topic
+      2. Read abstracts to verify relevance
+      3. Include the paper title, authors, year, and arXiv ID
+      4. Format citations as BibTeX entries in the bibliography
+      5. Reference papers using standard academic citation format
+
+      ## Tribunal Scoring Format
+
+      Present tribunal results as:
+      ```
+      === TRIBUNAL REVIEW ===
+      Reviewer 1: [Specialization]
+        Abstract:      [score]/10 - [brief feedback]
+        Introduction:  [score]/10 - [brief feedback]
+        Related Work:  [score]/10 - [brief feedback]
+        Methodology:   [score]/10 - [brief feedback]
+        Results:       [score]/10 - [brief feedback]
+        Discussion:    [score]/10 - [brief feedback]
+        Conclusion:    [score]/10 - [brief feedback]
+      [Repeat for Reviewers 2 and 3]
+
+      Sections requiring revision: [list]
+      ```
+
+      ## Output Format
+
+      Default to Markdown with LaTeX math notation. If the user requests LaTeX output, provide a complete .tex file with proper document class, packages, and bibliography.
+    groups:
+      - read
+      - - edit
+        - fileRegex: \.(md|tex|bib|txt|csv|json)$
+          description: Paper documents, bibliographies, and data files
+      - command
+    source: project
@@ -0,0 +1,101 @@
+# CAJAL Scientific Paper Generator Mode
+
+A custom Roo Code mode for generating publication-ready scientific papers with real arXiv citations and a multi-pass tribunal scoring workflow.
+
+Based on the [CAJAL project](https://github.com/Agnuxo1/CAJAL) by [@agnuxo1](https://github.com/Agnuxo1).
+
+## What It Does
+
+The Paper Generator mode turns Roo Code into an academic writing assistant that:
+
+- Generates structured 7-section research papers (Abstract, Introduction, Related Work, Methodology, Results, Discussion, Conclusion)
+- Sources real citations from arXiv (search + abstract retrieval + BibTeX formatting)
+- Runs a "tribunal" review: 3 simulated reviewers score each section on a 0-10 scale
+- Iteratively rewrites sections scoring below 7 until all meet the quality threshold (max 3 iterations)
+- Outputs Markdown or LaTeX with proper academic formatting
+
+## Installation
+
+### Option 1: Copy the `.roomodes` file (project-level)
+
+1. Copy the [`.roomodes`](.roomodes) file from this directory into your project root
+2. Open the project in VS Code with Roo Code installed
+3. The "Paper Generator" mode will appear in the mode selector
+
+If your project already has a `.roomodes` file, merge the `paper` mode entry into your existing `customModes` array.
+
+### Option 2: Global installation via `custom_modes.yaml`
+
+Add the mode definition to your global `custom_modes.yaml` file (located in your Roo Code settings directory):
+
+```yaml
+customModes:
+    - slug: paper
+      name: "\U0001F4DD Paper Generator"
+      roleDefinition: |-
+          You are CAJAL, a scientific paper writing assistant...
+      # (copy the full definition from the .roomodes file)
+```
+
+## Recommended Setup
+
+### Using with CAJAL via Ollama (best results)
+
+For the full CAJAL experience, run the CAJAL model locally through Ollama:
+
+1. Install [Ollama](https://ollama.ai)
+2. Pull the CAJAL model: `ollama run cajal-p2pclaw`
+3. In Roo Code, select Ollama as your API provider and point it to your local Ollama instance
+4. Select `cajal-p2pclaw` as the model
+5. Switch to Paper Generator mode
+
+### Using with any LLM provider
+
+The mode works with any LLM provider supported by Roo Code (OpenAI, Anthropic, etc.). The structured prompts and tribunal workflow are provider-agnostic -- you just won't get the CAJAL-specific optimizations for scientific writing.
+
+## Usage
+
+1. Switch to the "Paper Generator" mode in Roo Code
+2. Describe your research topic. For example:
+
+    > Write a paper on transformer architectures for protein structure prediction, focusing on attention mechanisms that capture long-range amino acid interactions.
+
+3. The mode will:
+
+    - **Pass 1**: Generate a full 7-section draft with arXiv citations
+    - **Pass 2**: Run tribunal review (3 reviewers scoring each section)
+    - **Pass 3**: Rewrite any sections below the quality threshold
+
+4. Output files are restricted to `.md`, `.tex`, `.bib`, `.txt`, `.csv`, and `.json` extensions
+
+## File Restrictions
+
+The mode can read any file in your project but can only write to:
+
+- Markdown files (`.md`)
+- LaTeX files (`.tex`)
+- Bibliography files (`.bib`)
+- Text files (`.txt`)
+- Data files (`.csv`, `.json`)
+
+This prevents accidental modifications to source code while working on papers.
+
+## How the Tribunal Works
+
+The tribunal simulates an academic peer review process:
+
+| Criterion                 | Description                                     |
+| ------------------------- | ----------------------------------------------- |
+| Clarity and coherence     | Is the writing clear and logically structured?  |
+| Technical accuracy        | Are claims and methods technically sound?       |
+| Depth of analysis         | Is the topic explored with sufficient depth?    |
+| Proper use of citations   | Are citations relevant and correctly formatted? |
+| Contribution to the field | Does the paper add meaningful value?            |
+
+Each of the 3 reviewers scores every section on these criteria. Sections averaging below 7/10 are flagged for rewriting. The process repeats up to 3 times.
+
+## Related
+
+- [CAJAL Project](https://github.com/Agnuxo1/CAJAL) - The original CAJAL scientific paper generation system
+- [Roo Code Custom Modes Documentation](https://docs.roocode.com/features/custom-modes) - How custom modes work
+- [Issue #12256](https://github.com/RooCodeInc/Roo-Code/issues/12256) - Original integration proposal