AI skills

Configure and customize the AI expert panel behavior for generating responses.

Overview

Rebuttr uses OpenCode skills to define AI expert behaviors. Skills are configuration files that tell the AI how to analyze reviewer comments and generate appropriate responses. Each expert in the panel is defined by a skill file.

Default Skills

Rebuttr installs the following skills during setup:

D Domain Expert

~/.config/opencode/skill/rebuttr-domain-expert.md

Analyzes comments for scientific accuracy, methodology concerns, and technical correctness. Provides domain-specific insights for crafting accurate responses.

W Writing Specialist

~/.config/opencode/skill/rebuttr-writing-specialist.md

Focuses on clarity, academic tone, and effective communication. Ensures responses are well-structured and professionally written.

R Reviewer Perspective

~/.config/opencode/skill/rebuttr-reviewer-perspective.md

Considers what reviewers and editors expect to see in a response. Helps anticipate follow-up questions and concerns.

S Response Strategist

~/.config/opencode/skill/rebuttr-strategist.md

Advises on diplomatic approaches, tone, and strategy for addressing sensitive or critical feedback.

Skill File Structure

Skills are Markdown files with a specific structure:

~/.config/opencode/skill/rebuttr-domain-expert.md
# Domain Expert You are a domain expert helping researchers respond to peer review. ## Your Role Analyze reviewer comments for: - Scientific accuracy concerns - Methodology questions - Data interpretation issues - Literature gaps ## Response Guidelines 1. Identify the core scientific question 2. Assess if the concern is valid 3. Suggest specific evidence or analysis to address it 4. Recommend manuscript changes if needed ## Tone - Objective and evidence-based - Acknowledge valid criticism - Defend work when appropriate with evidence

Creating Custom Skills

Create your own skills to customize AI behavior for your field or preferences:

Step 1: Create the Skill File

Terminal
# Create a new skill file $ touch ~/.config/opencode/skill/rebuttr-my-expert.md

Step 2: Define the Skill

~/.config/opencode/skill/rebuttr-my-expert.md
# Bioinformatics Expert You are a bioinformatics expert specializing in genomics and computational biology. ## Your Expertise - Sequence analysis and alignment - Statistical methods for genomics - Pipeline development - Reproducibility in computational research ## When Reviewing Comments 1. Check if computational methods are correctly described 2. Verify statistical approaches are appropriate 3. Suggest additional analyses if helpful 4. Recommend ways to improve reproducibility ## Response Style - Technical but accessible - Include specific tool/method recommendations - Reference relevant benchmarks or standards

Step 3: Register the Skill

Add your skill to the Rebuttr configuration:

~/.config/rebuttr/config.json
{ "ai": { "expertPanel": [ "rebuttr-domain-expert", "rebuttr-writing-specialist", "rebuttr-reviewer-perspective", "rebuttr-my-expert" ] } }

Skill Configuration Options

Configure how skills are used in ~/.config/rebuttr/config.json:

Option Description Default
ai.expertPanel Array of skill names to use All default skills
ai.expertPanelSize Maximum experts per comment 4
ai.skillDirectory Custom skill directory path ~/.config/opencode/skill/
ai.parallelExperts Run experts in parallel true

Field-Specific Skills

Consider creating skills tailored to your research area:

Pro Tip

Share useful custom skills with your research group or lab. A shared skill directory can ensure consistent response quality across team members.

Disabling Skills

To disable specific skills or use fewer experts:

~/.config/rebuttr/config.json
{ "ai": { "expertPanel": [ "rebuttr-domain-expert", "rebuttr-writing-specialist" ], "expertPanelSize": 2 } }

Debugging Skills

To see how skills are being used:

Terminal
# Enable verbose mode to see expert outputs $ rebuttr generate --paper "My Paper" --comment 1 --verbose # List available skills $ ls ~/.config/opencode/skill/ | grep rebuttr
Skill Syntax

Skills must be valid Markdown files. Syntax errors may cause AI generation to fail. Test new skills on a single comment before using them broadly.

Next Steps