Platform

Agent access

Metix AI skills

Install verified, task-focused skill packages when an agent should plan and execute multi-step data work rather than make one call.

Verified skill names

Skill
metix-platform-assistant
Use
Coordinate multi-step work across the people, job, and company datasets.
Skill
metix-people-search
Use
Search profiles by role, skills, employer, education, location or seniority, then read the records behind the ids.
Skill
metix-job-search
Use
Search active postings, then fetch the records behind the job IDs.
Skill
metix-company-search
Use
Search organizations, then fetch the records behind the company IDs.

One skill per dataset, plus a coordinator for work that crosses them. Each skill already knows the two-step flow, so it searches for IDs and then fetches records instead of trying to read facts out of a search response.

These four are the whole set

A skill name outside this list is either retired or was never published. If an agent references one, point it at the REST reference instead.

Install

GitHub is the single distribution point. Install the repository as one package:

npx skills add MetixAI-Official/metix-skills

export METIX_KEY="metix_xxxxxxxxxxxx"

The installer asks three things. The first is which of the four skills to install, and it starts with none of them ticked, so press space on each one before you press enter. The second is which agents to install for; the default already covers Codex, Cursor, Cline, Copilot, Gemini CLI and a dozen others through the shared .agents/skills directory they all read, and adds .claude/skills for Claude Code. The third is project or global. Choose project and the skills are committed with the repository; choose global and every project on the machine gets them.

The key is the same one everything else uses. There is no second variable and no bridging line to copy: set METIX_KEY in the shell you start the agent from, and the skills, the MCP server and every curl example on these pages all read it.

Codex blocks the network inside its sandbox

Codex runs shell commands with network access switched off by default, so the first curl a skill tries comes back could not resolve host, which reads like an outage rather than a local policy. It is not: either approve the command when Codex asks, or set network_access = true under [sandbox_workspace_write] in ~/.codex/config.toml. Registering the MCP server avoids the question entirely, because MCP calls do not go through the command sandbox.

Moving from the OpenJobs skills

The earlier package, installed as OpenJobsAI/openjobs-openclaw-skills, is retired. Its skills are still loaded by any agent that installed them, and they describe an API surface that has moved on, so an agent running both will sometimes follow the older one. Remove them before you rely on the four above.

See what is installed:

npx skills list        # skills in this project
npx skills list -g     # skills installed globally

Remove the retired ones. Add -g to either command to work on the global scope instead of this project:

npx skills remove openjobs-people-search openjobs-company-search   openjobs-jobs-search openjobs-people-match   openjobs-ai-talent-search openjobs-platform-assistant

Running npx skills remove with no names opens an interactive picker, which is easier when you are not sure which are present.

Or ask the agent to do it:

List every skill you have installed. Remove the ones whose names start with
"openjobs-", check both the project and the global scope, then list what is
left so I can confirm.

Your API key does not change

The skills, the MCP server, and every curl example read one variable, METIX_KEY. If you followed an older page that said MIRA_KEY, rename it and change nothing else. The key itself is the same key, and a key issued before the rename still begins with mira_ and still works.

Market-analysis prompt

Find companies hiring machine learning platform engineers in San Francisco,
rank them by active job count, then identify senior engineering profiles
connected to those markets.

That crosses all three datasets: a job search for demand, a company lookup for context, and a people search for the profiles behind it. Every underlying call keeps its documented request shape, limits, and Credit rules.