Orion exposes a lot of MCP tools, and several names overlap (record_message vs record_interaction_note vs ingest_paste, send_message vs mark_draft_sent, update_candidacy_stage vs update_candidate_interest). Picking the wrong one writes bad data or double-sends. The Orion MCP usage skill encodes the recruiting workflows — which tool to use, in what order, and the safety invariants — as a progressive-disclosure skill so an agent gets it right.

Download the skill

orion-mcp-usage-skill-1.0.0.zip — SKILL.md + per-domain reference docs

​
Install

The simplest way is to let the agent install it for you. Any agent that supports Skills (Codex, Claude Code, or a local agent) already knows where a skill belongs and will place it following the Skills convention — you don’t unzip or move files by hand.
1

Download the zip

Download orion-mcp-usage-skill-1.0.0.zip above.
2

Open a Skills-capable agent

Codex, Claude Code, or any local agent that supports the Skills convention.
3

Hand the zip to the agent

Drag orion-mcp-usage-skill-1.0.0.zip into the chat (or give the agent its file path), then send a message like:
Help me install this Orion MCP usage skill. Before installing, check whether a duplicate Orion skill already exists.
4

Done

The agent unpacks it into its skills directory and loads it. From then on it consults the skill automatically whenever you work on Orion recruiting tasks.
If the agent can’t read zip files directly, unzip it yourself and give the agent the extracted orion-mcp-usage/ folder (or its path) — it will finish the install from there.

​
What’s inside

The orion-mcp-usage/ folder is a standard skill:
  • SKILL.md — the mental model, the core recruiting loop, a tool-selection table for the overlapping tools, and the safety invariants (confirm before sending, search before creating, never invent ids).
  • references/ — deeper guidance loaded only when a flow needs it (progressive disclosure): connections.md (Gmail auth), cases.md, candidates.md, communication.md, interviews.md.
The agent reads SKILL.md first, then opens only the reference for the flow it’s in — so it stays token-efficient.

​
FAQ

No. The skill is pure guidance on how to use the tools. Your Orion access is the separate MCP connection — see Connect to Orion MCP. Install the skill once; connect the token once.
Any agent that supports the Skills convention (Codex, Claude Code, and more). MCP itself does not auto-serve skills, so you install the skill into the agent once; MCP just provides the tools it operates.
Download the latest zip and hand it to the agent again with the same prompt — it will replace the existing orion-mcp-usage skill.

​
Why a skill instead of more tools

The tool surface is intentionally kept lean. Scenarios (“log a call”, “advance and mark interested”, “connect my Gmail”) live in this guidance layer as tool sequences rather than as one bespoke tool per scenario — so the toolset stays small while the number of workflows the agent can handle stays open-ended.