Walllmat/verdict
🧑⚖️ Evaluate AI skill and agent performance with configurable scores and dual-mode operation on Claude Code and Cowork platforms.
What's novel
🧑⚖️ Evaluate AI skill and agent performance with configurable scores and dual-mode operation on Claude Code and Cowork platforms.
Code Analysis
11 files read · 4 roundsA Claude Code plugin that evaluates the execution quality of other skills/agents using LLM-driven 7-dimension scoring with supplementary regex-based transcript heuristics, benchmark comparison, and visual scorecard reporting.
Strengths
Well-organized architecture with clear separation between LLM instructions (SKILL.md, agent definitions), automated heuristics (Python scripts), and lifecycle hooks (shell scripts). Comprehensive test suite with thorough edge case coverage and clean, well-documented code with proper type hints.
Weaknesses
The README is severely misleading, describing a GUI desktop application with download/install steps when the project is actually a Claude Code plugin. The Python scoring engine is shallow—relying on regex word matching against transcripts rather than meaningful code analysis, making the 'quality scoring' largely dependent on LLM judgment rather than algorithmic depth.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
7
Contributors
2
Files
37
Active weeks
2
Repository
Language
Python
Stars
1
Forks
0
License
MIT