Agent skill

If your team builds with Claude Code, Codex or Cursor, install this skill and the agent gets the whole API: the endpoints, the credential rules, the failure modes and the traps that are not obvious from a spec.

It is not a copy of these docs. It is written for a machine that is about to write an integration, so it leads with the rule, then the call, then what breaks.

Two skills, one per product

They are separate because the APIs are. Install the one that matches what you are building; install both only if you integrate both.

Install

# Evaluating candidates
curl -sL https://docs.experthire.io/experthire-hiring-api-skill.zip -o /tmp/eh.zip
unzip -o /tmp/eh.zip -d ~/.claude/skills/

# Helping people prepare
curl -sL https://docs.experthire.io/experthire-prep-api-skill.zip -o /tmp/eh.zip
unzip -o /tmp/eh.zip -d ~/.claude/skills/

That gives every project on the machine the skill. For one repository, unzip into .claude/skills/ at its root and commit it, so the whole team gets the same answers.

Codex and Cursor read agent instructions from their own locations. The files are plain markdown with YAML frontmatter, so put SKILL.md and any references/ wherever your tool looks.

Verify it landed:

ls ~/.claude/skills/experthire-hiring-api
# SKILL.md

Both tell the agent to start in sandbox

Each skill makes the environment an explicit decision before any code is written: default to the sandbox key, use live only when the user has asked for a real result. Sandbox costs no credits, contacts nobody, and never reaches a recruiter's pipeline, so an agent building an integration cannot leave real activity behind.

The base URL is stated too, along with the fact that it does not change between environments: the key decides that, not the host.

What is in it

Preparation

How to use it

Ask for what you want, in your own words:

Add Expert Hire interviews to our ATS. When a candidate reaches the
screen stage, create an interview and email them the link. Store the
score on the application when it comes back.

The agent loads the skill, reads only the references it needs, and writes the client. It should ask you which environment to target and where the key lives before it writes anything.

What it will refuse to do

It is written to say no to a few things, on purpose:

• Put a secret key anywhere a browser can reach it.
• Poll for completion instead of registering a webhook.
• Retry a create call without an Idempotency-Key. The header is optional, so a blind retry after a timeout can create a second interview and spend a credit.
• Book a coding interview round when the requirement is graded execution. The room does render an editor and the code is stored and read by the scorer, but nothing runs it against test cases. It books a coding test instead.

If your agent does any of those, it did not load the skill.

Keeping it current

The bundle is rebuilt whenever these docs deploy. Re-run the install command to pick up changes. The skill states its own limits, so an agent reading a stale copy fails by refusing to build something that now exists rather than by inventing something that never did.

Hiring