method-evolution

Detect recurring rubric violations and propose a single method change with measured loss.

2|Updated Apr 20, 2026

Search Tags:#evolution #measurement #experimentation #loss #outer-loop #method-evolution

Ready to run

npx skills add https://github.com/veegee82/loss-driven-development --skill method-evolution

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: method-evolution
Source: https://github.com/veegee82/loss-driven-development/tree/main/skills/method-evolution
Command: npx skills add https://github.com/veegee82/loss-driven-development --skill method-evolution

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Evolves an agent's method when the same rubric violation recurs across multiple tasks and domains, ensuring issues originate from the method rather than the tasks.

Core Features & Use Cases

Automatically detects recurring patterns across tasks and domains, signaling a need for method evolution.
Proposes a single, concrete change to the method artifact (e.g., a rubric tweak, a new constraint, or a clearer prompt) and documents the rationale.
Measures impact with a task-suite loss metric (Δloss) before and after the change, guiding auditable rollbacks if needed.

Quick Start

Identify a recurring pattern observed across three tasks, propose one concrete method change, and run the task suite to measure Δloss across the suite to guide the evolution.

Frequently Asked Questions about method-evolution

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I fix a recurring rubric violation that appears across multiple agent tasks?▼

To fix a recurring rubric violation, you evolve the agent's method by identifying the shared pattern, proposing a single concrete method change, and measuring the task-suite loss metric before and after the update. This ensures the issue stems from the method, not the tasks.

What is agent method evolution and when should I apply it?▼

Agent method evolution is the process of updating an agent's core method artifact when the same rubric violation recurs across multiple tasks and domains. You should apply it when you observe a recurring pattern across at least five varied tasks, ensuring the problem originates from the method itself.

How do I measure the impact of a method change across a task suite?▼

You measure the impact of a method change by calculating the delta loss (Δloss) across the task suite. Run the suite with the method_before and method_after configurations to compare the measurable loss metric, which guides auditable rollbacks if the change does not improve performance.

Does method evolution require a minimum number of tasks to work?▼

Yes, method evolution requires a task suite with at least five varied tasks. This minimum scale ensures the recurring rubric violations are detected across multiple domains, confirming the issue originates from the method rather than individual task anomalies.

What is the best way to document an agent method evolution commit?▼

The best way to document an agent method evolution commit is to name the observed violation pattern, propose a single concrete method change such as a rubric tweak or clearer prompt, and record the rationale alongside the measured Δloss to ensure a fully auditable evolution history.

Why does my agent keep failing the same rubric criteria on different tasks?▼

Your agent keeps failing the same rubric criteria on different tasks because the underlying method artifact lacks the necessary constraints or clarity. When a violation recurs across multiple domains, it signals a method-level issue that requires a concrete method evolution rather than individual task fixes.