Evaluate completed work with two-axis scoring and devil's advocate reasoning.
npx skills add https://github.com/Patasse97/claude-skills --skill self-eval-patasse97
Please help me install this Agent Skill. Skill: self-eval Source: https://github.com/Patasse97/claude-skills/tree/main/engineering/self-eval Command: npx skills add https://github.com/Patasse97/claude-skills --skill self-eval-patasse97
Self-eval provides honest, calibrated work evaluations after completing tasks or reviews by applying a two-axis scoring system and mandatory devil's advocate reasoning, addressing the common bias of inflated self-assessments.
After completing work in a Claude Code session, run /self-eval with context about what was evaluated to generate the assessment.
High-intent search queries and answers about installing and using this skill.
Calibrated work evaluations apply a two-axis scoring system and mandatory devil's advocate reasoning to independently rate task ambition and execution quality, preventing inflated self-assessments.
Devil's advocate reasoning for AI quality assurance requires arguing for both higher and lower scores before finalizing an assessment, ensuring honest reviews and mitigating self-evaluation bias.
Prevent score inflation by persisting evaluation results to a dedicated JSONL history file, which analyzes past scores to detect clustering and stabilize assessments over time.
Self-eval requires no external tools or dependencies to run, operating entirely within the session to apply matrix-based composite scoring and write results to local history files.
The best way to score task ambition and execution quality independently is using a two-axis scoring system, where a predefined matrix derives the final composite score for consistent judgments.
Avoid automated self-evaluations when you lack session history for anti-inflation detection, as the tool relies on reading past score files to flag clustering and calibrate current assessments.
sgcarstrendsAutomate detection of security vulnerabilities, performance issues, and code quality defects in code reviews.
devCharuzuScore AI work quality on ambition and execution axes with devil's advocate reasoning.
iZiTTMarvinArchive completed code development sessions with commit hashes and task status.
4lermanEvaluate AI work quality with two-axis scoring and devil's advocate reasoning.
