Vibe Code Your Administration

Manage Confident AI from Cursor, Claude Code, or Codex with natural-language prompts.

Overview

The confident-client Agent Skill teaches your coding agent — Cursor, Claude Code, or Codex — how to drive the Admin SDK for you. Instead of writing SDK code by hand, you describe what you want (“create a project owned by alice@example.com”) and the agent writes and runs the correct call.

A skill is just a SKILL.md file plus reference docs. It ships inside the confident-client repository — the same repo as the Python and TypeScript SDKs — so the guidance always matches the code.

This is the full walkthrough behind the Vibe Code Your Administration section of the Admin SDK quickstart. It covers account administration — organizations, projects, members, roles, governance policies, and API keys — and authenticates with an Organization API Key (CONFIDENT_ORG_API_KEY), not the project key used for tracing and evaluations.

Here’s the flow you’ll set up:

1

Install the Admin SDK

The skill drives the confidentai Admin SDK — that’s the package the agent actually runs. Install it in the environment your agent executes commands in, matching the language your project uses:

$pip install confidentai
2

Install the skill

The skill lives in the confident-client repo. Install it via the Claude Code plugin, or with the Skills CLI for any other agent.

Run these four commands in Claude Code:

$/plugin marketplace add confident-ai/confident-client
$/plugin install confident-client@confident-ai-plugins
$/reload-plugins
$/plugins

The /plugins command should list confident-client under your installed plugins.

3

Set your Organization API Key

Every management call needs an Organization API Key (e.g. confident_us_org_...). Retrieve yours, then export it in the terminal your agent runs commands in:

$export CONFIDENT_ORG_API_KEY="confident_us_org_..."

This is not the same as CONFIDENT_API_KEY, which is a project-scoped key for tracing and evals. Keep both out of source control. The agent needs CONFIDENT_ORG_API_KEY in its shell session to actually create anything.

4

Create your first project

Open your agent and describe the project you want. For example:

Prompt
Create a Confident AI project called "Customer Support Bot" and make alice@example.com the owner.

The skill will:

  • Confirm which SDK language to use (Python or TypeScript) if your project isn’t clearly one or the other.
  • Ask whether to assign an owner (the email must belong to an existing organization member).
  • Create the project and return its id plus a one-time project API key — the full secret is only shown at creation.

Because the project key is only returned once, copy it somewhere safe as soon as the agent prints it. You’ll use it in the last step.

Done ✅. You just created a project without writing a line of SDK code.

5

Invite members and assign roles

Onboard your team the same way — with prompts:

Invite members
Invite alice@example.com and bob@example.com to my organization.

The skill asks which role to grant invitees before sending (inviting requires a paid plan).

Define and assign a role
Create an "Analyst" role with read-only access and assign it to bob@example.com.

Roles are built from policies, which are built from permissions — the skill composes them in the right order for you. You can manage governance too:

Governance
Assign our governance policy to all of my production projects.
6

Use your project API key

The project API key from step 3 is what your application uses for tracing and evaluations. Paste it into the environment where your app runs:

$export CONFIDENT_API_KEY="confident_us_proj_..."

That’s the project-scoped key (CONFIDENT_API_KEY) — distinct from the CONFIDENT_ORG_API_KEY you used for administration. With it set, deepeval and the tracing SDKs send data to your new project.

Done ✅. Your project is live and ready to receive traces and evals.

Next Steps

You installed the SDK and skill, created a project, onboarded members, and wired up a project key — all through prompts. To go deeper: