# Evaluate > This skill should be used when evaluating the logical flow, structure, readability, and overall quality of research paper text. Use for assessing academic writing targeting top-tier computer science conferences. Use when this capability is needed. - Skill: `tomevault-io/evaluate` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add tomevault-io/evaluate` - Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/evaluate/raw - Safety review: pending (external: skill-scanner PASS, skillspector PASS) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Research & Search - Author: tomevault-io (https://skillmd.com/u/tomevault-io) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/tomevault-io/evaluate --- # Academic Text Evaluator Evaluate research paper text for logical flow, structure, clarity, and readability without modifying the original content. ## When to Use This Skill - Assessing the quality of research paper sections - Evaluating logical flow and argument structure - Providing feedback on clarity and readability - Scoring text quality for academic writing - Identifying areas for improvement in conference submissions ## Target Audience Graduate students, professors, and researchers writing for top-tier computer science conferences (e.g., OSDI, NSDI, SOSP, SIGCOMM). ## Evaluation Criteria Evaluate text across five dimensions: ### 1. Logical Cohesion - Assess whether arguments progress naturally and convincingly - Identify logical jumps or gaps in reasoning - Check if sentences and paragraphs connect smoothly - Verify effective use of transition words and phrases ### 2. Clarity and Fluency - Determine if the text is easy to understand - Check for precise and unambiguous language - Assess overall reading fluency ### 3. Organization - Evaluate information structure effectiveness - Check if paragraphs are well-focused with distinct points - Assess optimal ordering of ideas ### 4. Pacing and Detail - Identify content that is too verbose or too terse - Check appropriate detail level for the target audience ### 5. Reader Engagement - Assess if readers can easily follow the narrative - Verify that main points are clear and graspable ## Scoring Guidelines Provide an overall quality score (0-100) with these requirements: - **Linear consistency**: Score should linearly reflect quality - **Proportional scaling**: If one mistake reduces score to 90, nine similar mistakes should not reduce it to 0 - **Impact indication**: For each suggested modification, indicate score impact (e.g., "+5 points") ## Feedback Format Follow these principles when providing feedback: ### Bad-first Approach - Focus primarily on weaknesses and areas for improvement - Only discuss strengths if the text is of very high quality - Avoid praising adequate or mediocre work ### Self-consistency - If text previously scored 100 and hasn't changed, do not invent new improvements - Maintain consistent standards across evaluations ### Actionable Advice - Provide specific, concrete suggestions - Include examples when helpful - Format: "The transition between paragraph 2 and 3 feels abrupt; consider adding a sentence to bridge X with Y. (+3 points)" ## Important Constraints - **Do not modify** the original content during evaluation - **Only suggest significant improvements** that meaningfully impact quality - Avoid pedantic or minor suggestions - Penalties apply for suggesting non-significant modifications ## Output Structure 1. **Overall Score**: (0-100) 2. **Dimension Scores**: Optional breakdown by the five criteria 3. **Key Issues**: List of significant problems identified 4. **Specific Suggestions**: Actionable improvements with estimated score impact 5. **Strengths**: Only if score > 85 --- > Source: [minhuw/claude-writer](https://github.com/minhuw/claude-writer) — distributed by [TomeVault](https://tomevault.io).