Skill Evaluation Graph banner
MaxLaurieHutchinson MaxLaurieHutchinson

Skill Evaluation Graph

Development community

Description

Description Evidence-driven evaluation, benchmarking, and verified repair for Agent Skills across Codex, Claude Code, Gemini CLI, and Antigravity.

Installation

This entry records only its repository, not the path inside it, so there is no exact command to give. Open the source below and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

README

SEG: Skill Evaluation Graph

SEG: Skill Evaluation Graph

[![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](LICENSE) [![Architecture Evidence](https://github.com/MaxLaurieHutchinson/skill-evaluation-graph/actions/workflows/architecture.yml/badge.svg)](https://github.com/MaxLaurieHutchinson/skill-evaluation-graph/actions/workflows/architecture.yml)

**Evidence-driven evaluation for Agent Skills.**

SEG combines deterministic static analysis, an executable Evaluation Graph, fail-closed gates, behavioural Control vs. Treatment trials, and bounded Repair Verification.

Why SEG exists

A valid `SKILL.md` is not the same thing as a reliable skill.

SEG evaluates several independent concerns without allowing one good score to hide a critical failure:

  • Specification Conformance — Agent Skills and configured host manifests.
  • Evaluation Integrity — required Evaluators must complete successfully.
  • Safety & Privacy — destructive commands and common workstation-path leaks.
  • Link Integrity — relative references and package assets.
  • Static Quality — deterministic heuristics for routing, structure, context efficiency, and maintainability.
  • Behavioural Reliability — optional live Control vs. Treatment trials through a Harness Adapter.

**Gates own acceptance. Scores are supporting evidence.**

Quick start

Run a deterministic evaluation:

python plugins/skill-evaluation-graph/scripts/audit_skill.py 

Run the Evaluation Graph and bounded repair loop. This is read-only by default:

python plugins/skill-evaluation-graph/scripts/run_loop.py  --target-score 95

Preview and then explicitly apply a verified Repair Candidate:

python plugins/skill-evaluation-graph/scripts/run_loop.py  --apply

Run static behavioural policy analysis:

pyth