efaimo

Version updated for https://github.com/efaimo-ai/efaimo to version v0.1.0.

  • This action is used across all versions by 0 repositories.

Action Type

This is a Composite action.

Go to the GitHub Marketplace to find the latest changes.

Action Summary

The efaimo action is a tool designed to audit and validate Agent Skills and MCP servers. It checks the context budget, trigger quality, linting errors, and migration diffs against the 2026-07-28 spec. Additionally, it provides a way to measure whether a skill improves task completion using A/B testing with LLM judges. The tool is useful for developers to ensure that their agent’s skills are well-formed, efficient, and effective.

What’s Changed

First release. One CLI that audits what your agent loads: MCP servers and Agent Skills.

  • efaimo weigh: context-window cost of MCP tool definitions (stdio, remote, or a whole client config) and of Agent Skills, in three serializations, with an optional --anthropic exact Claude count. Per-tool numbers plus a block-framing line reconcile exactly with the total. CI budget gates (--out/--diff/--max-tokens/--allow-increase) and a badge.
  • efaimo check --mcp: a quality grade (descriptions, schemas, annotations, cost) plus a separate ungraded 2026-07-28 migration diff, each item naming its SEP. Speaks both the legacy handshake and bare stateless requests, so 2026-07-28 servers audit fine.
  • efaimo check --skill: lints Agent Skills against the agentskills.io spec, per-skill grades over a folder, plus a reproducible Skills Quality Index (pinned-commit corpus).
  • efaimo test (experimental): with/without-skill A/B trials with an LLM judge; dry-run by default, opt-in --live.
  • efaimo mcp: a read-only MCP server exposing the skill checks to an agent.

Install: npx efaimo or npm i -D efaimo. Node 22+. Apache-2.0.

Full changelog: https://github.com/efaimo-ai/efaimo/blob/main/CHANGELOG.md