Self-Eval: Honest Work Evaluation
Self-Eval is a Claude Code skill that produces honest, calibrated evaluations of AI work using a two-axis scoring system (task ambition and execution quality) combined via a fixed matrix. It forces devil's advocate reasoning, detects score inflation across sessions, and persists scores to a history file.
✨ What it does
- Two-axis scoring: rates task ambition (Low/Medium/High) and execution quality (Poor/Adequate/Strong) independently.
- Mandatory devil's advocate: argues for both higher and lower scores before finalizing.
- Score persistence: appends scores to .self-eval-scores.jsonl in the working directory.
- Anti-inflation detection: flags clustering when 4+ of the last 5 scores are identical.
- Matrix-locked scoring: composite score comes from a fixed matrix, not direct selection.
🎯 When to use it
- After completing a task or work session to get an unbiased assessment of the work quality.
- After a code review to evaluate the quality of the reviewed changes.
- When you suspect the AI's self-assessment may be inflated and want a more rigorous check.
🚀 How to use
Trigger: /self-eval
Trigger with the slash command /self-eval in a Claude Code session, optionally followed by a description of the work to evaluate. If no arguments are given, it reviews the conversation history to identify what was accomplished. The skill then scores the work on two axes, runs the devil's advocate check, and appends the result to the score history file.
Example prompts:
/self-eval
/self-eval the authentication refactor we just completed
📄 Output: A structured self-evaluation report with task summary, axis ratings, devil's advocate reasoning, and a final score, plus an appended JSON line to .self-eval-scores.jsonl.
📦 Add this skill to Claude Code
# 1. Get the skills repo
git clone --depth 1 https://github.com/alirezarezvani/claude-skills /tmp/claude-skills
# 2. Copy this skill into your project (or ~/.claude/skills for all projects)
mkdir -p .claude/skills
cp -r /tmp/claude-skills/engineering/skills/self-eval .claude/skills/self-evalSkill source: engineering/skills/self-eval/SKILL.md
❓ FAQ
How does the skill prevent score inflation?
It uses a two-axis scoring system with a fixed matrix that caps low ambition at 2/5, and it checks past scores for clustering (4+ of last 5 identical) to flag anchoring.
What happens if I don't provide any arguments to /self-eval?
It reviews the full conversation history to identify what was accomplished in the session and summarizes the work in one sentence before scoring.
Can I override the composite score from the matrix?
No, the matrix result is final. The devil's advocate step can cause you to re-rate an axis, but you cannot directly override the matrix result.
🤖 Overview, features, install steps and FAQ were generated from the project's SKILL.md on Sep 4, 2026. Always check the original source before running commands.