Autousers
auto_awesome

The Autousers Agent Skill

schedule3 mincalendar_todayLast updated 2026-05-01

What the Autousers Skill is, how it complements MCP, and the open standard it follows.

What is an Agent Skill?

An Agent Skill is a folder containing a SKILL.md file plus optional reference materials and templates. The SKILL.md format is a lightweight, vendor-neutral open standard that gives an AI agent specialized knowledge and workflows for a domain. Anthropic released Skills as an open spec in 2025 at agentskills.io, and the format is now supported by a growing list of agent products.

open_in_new
Open standard
The Autousers Skill follows the agentskills.io spec. The same SKILL.md works in any agentskills.io-compliant client — see https://agentskills.io/clients for the up-to-date list.

What the Autousers Skill adds

The Autousers MCP server gives the agent ~40 tools for driving evaluations, autousers, calibration, templates, and analytics. The Skill is the operating manual: it teaches the agent when and how to chain those tools, what the dryRun-first cost-preview rule is, what kappa thresholds matter, and how to format results consistently.

  • Five hard rules including dryRun-first cost preview, drafts-by-default, never-pick-personas-for-the-user, ID resolution shortcuts, and required teamId on create
  • Five workflow files covering eval creation, autouser dispatch, calibration, results analysis, and template management
  • A glossary of Autousers domain terms (SxS, SSE, kappa, dimension, factor, anchor, BYOK)
  • Output templates for evaluation summaries and calibration reports
  • Reference for the smart-resolver shortcuts (latest, recent, running, pick, position numbers, fuzzy match)

Skill vs MCP — complementary, not standalone

The Skill and the MCP are complementary. The MCP is the wiring (tools the agent can call); the Skill is the operating manual (when and how to chain them). On its own, the Skill is just useful prose the agent has read — it can describe Autousers concepts but can't take action without the MCP tools available.

  • MCP without Skill — works, but the agent must figure out tool sequencing on its own each time
  • Skill without MCP — the agent learns the concepts but cannot drive Autousers; suitable only if the host has the Autousers REST API wired in some other way
  • Skill + MCP (recommended) — the agent has the tools and the manual, so workflows run consistently and cost-safely

Where it works

Anywhere the SKILL.md spec is supported. Pick the install guide that matches your environment:

  • Claude Code → Install the Skill in Claude Code
  • Claude.ai or the Claude API → Install the Skill in Claude.ai or the Claude API
  • Cursor, Codex, Gemini, and other agentskills.io clients → Install the Skill in other clients
infoThe Autousers Skill is a pure-markdown package — no executable scripts. It works identically across every agentskills.io-compliant runtime.
Was this article helpful?