Can I prevent a specific tool from being used if it recently failed?

Yes, the skill applies a failure pattern penalty if a tool has recently failed on tasks with similar tags, reducing its priority score automatically.

How does Agent Selection determine the best tool?

It uses a weighted decay formula based on historical success (outcome) and the age of the observation, blending recent data with a neutral baseline for confidence.

What tools are supported by this skill?

It is designed to rank any CLI tool like Gemini, Codex, or Claude, provided they are configured in your static priority list and have logged performance data.

Where is the performance data stored?

All task results, success rates, and tool reports are stored locally in a 'tribunal-stats.jsonl' file for transparent tracking and scoring.

Agent Selection Scorer

Name: Agent Selection Scorer
Author: jpeggdev

byjpeggdev

0•

Productividad y Flujo de Trabajo

Routes developer tasks to the most effective CLI tool using intelligent performance scoring and historical success data.

Agent Selection (part of the Tribunal plugin) optimizes AI-driven development by intelligently routing tasks to the most capable agent or CLI tool based on real-world performance metrics. Instead of relying on static ordering, this skill uses a weighted decay algorithm to prioritize tools with recent success and applies penalties for failure patterns matching the current task's tags. By maintaining a local history of outcomes, it ensures that Claude Code always selects the most reliable tool for implementation, refactoring, or debugging, resulting in higher success rates and less manual intervention.

Características Principales

01Weighted decay scoring algorithm to prioritize recent successes

02Failure pattern matching with tag-based context analysis

030 GitHub stars

04Customizable scoring parameters via tribunal.yaml configuration

05Structured JSONL performance logging for transparent tool auditing

06Automatic static fallback for tools with insufficient historical data

Casos de Uso

01Benchmarking CLI tool effectiveness for specific projects or domains

02Dynamic routing between LLM agents like Gemini and Codex based on task type

03Reducing agent failures through context-aware penalty application

Características Principales

01Weighted decay scoring algorithm to prioritize recent successes

02Failure pattern matching with tag-based context analysis

030 GitHub stars

04Customizable scoring parameters via tribunal.yaml configuration

05Structured JSONL performance logging for transparent tool auditing

06Automatic static fallback for tools with insufficient historical data

Casos de Uso

01Benchmarking CLI tool effectiveness for specific projects or domains

02Dynamic routing between LLM agents like Gemini and Codex based on task type

03Reducing agent failures through context-aware penalty application