Extension Facts — AI skill for Claude Code

Extension Performance Facts

A skill's description tells you what it's for, not whether it works. Say a skill claims to make an agent better at writing docs or reviewing code. How would you know? "It works great for me" isn't evidence, and neither is a slick demo, a starred repo, or one good-looking output. And even if the skill does improve results, it might use twice the tokens to do it. That's a tradeoff only you can judge, for your own workload.

Extension Performance Facts is a nutritional label for agent extensions: skills, MCP servers, instructions, custom agents, and plugins that bundle them. It shows whether an extension makes the agent better (lift) or worse (drag) compared with the same agent without it. Next to that it shows what the extension costs in tokens and dollars, the exact agent setup it was measured on, and links to the eval behind every number.

Each extension version gets a machine-readable facts document. The 76-column text label is rendered from it.

┌──────────────────────────────────────────────────────────────────────────┐
│ EXTENSION PERFORMANCE FACTS                                EPF-1 · REV A │
├──────────────────────────────────────────────────────────────────────────┤
│ EXTENSION  gitload 1.2.0 · skill                                         │
│ SCOPE      Download files, folders or whole repos from GitHub URLs.      │
├──────────────────────────────────────────────────────────────────────────┤
│ TASKS 11   PROFILES 2   RUNS/ROW 5   EVAL 2026-10-05                     │
╞══════════════════════════════════════════════════════════════════════════╡
│ CONFIGURATION          SCORE¹  RANGE     Δ²   DRAG│LIFT       TOK³  USD³ │
├──────────────────────────────────────────────────────────────────────────┤
│ T9  PRIVATE REPO FOLDER WITH GH CLI SIGNED IN                 6 CRITERIA │
│     Copilot SDK 1.0.14 · gpt-5.6-terra · default · macOS 27              │
│     baseline              60%  50–100     —       │            19k 0.047