---
schema_version: "newruntime-agent-readable-v0.1"
type: "raw_signal"
id: "tg-2707"
slug: "agent-skills-need-behavioral-evals"
title: "Agent skills need behavioral evals, not prose review"
description: "Benchmarks show that expert-authored skills can help while self-generated skills can underperform a no-skill baseline."
observed_at: "2026-07-16"
why_it_matters: "A skill is executable behavior packaged as files, so its quality has to be measured across tasks, models, and harnesses instead of judged by how polished the instructions look."
novelty: "structural"
verification_level: "source-inspected"
signal_type: "field-report"
evidence_kind: "mixed"
status: "published"
telegram_message_id: 2707
telegram_url: "https://t.me/qwgai/2707"
topics: ["agent-skills","evals","context-engineering"]
entities: ["SkillsBench","Claude"]
related_patterns: []
source_urls: ["https://arxiv.org/abs/2602.12670","https://arxiv.org/abs/2603.29919","https://arxiv.org/abs/2605.24050","https://github.com/benchflow-ai/skillsbench","https://platform.claude.com/docs/en/agents-and-tools/agent-skills/best-practices"]
import_batch: "telegram-2026-07-17-v1"
routes: {"html":"https://newruntime.com/signals/agent-skills-need-behavioral-evals/","markdown":"https://newruntime.com/signals/agent-skills-need-behavioral-evals.md","json":"https://newruntime.com/signals/agent-skills-need-behavioral-evals.json"}
source_format: "telegram-export-normalized-json"
---

# Agent skills need behavioral evals, not prose review

## Observation

Benchmarks show that expert-authored skills can help while self-generated skills can underperform a no-skill baseline.

## Why it matters

A skill is executable behavior packaged as files, so its quality has to be measured across tasks, models, and harnesses instead of judged by how polished the instructions look.

## Provenance

Normalized from QWG AI Telegram message 2707. The original Russian-language record remains available at https://t.me/qwgai/2707.