Skip to content
For Self-Managed AI Platform users running v11.1, see the on-premise platform documentation
Workload API > Get started: Workload API > Workload API skill

Workload API skill

プレビュー

Workload APIはプレビュー機能で、デフォルトではオンになっています。

機能フラグ：Workload APIのエクスペリメント機能へのアクセスを有効にする

The Workload API skill (datarobot-workload-api) packages Workload API knowledge and operational workflows for use inside third-party coding agents. Instead of consulting reference documentation while you work, you describe what you want in natural language and the skill guides your coding agent through the correct API calls, debugging steps, and lifecycle decisions.

このスキルは、他のDataRobotスキルとともにDataRobot Agentic Skillsリポジトリにあります。 It covers four operational areas: deploying and scaling Workloads, diagnosing failures, observing running Workloads, and managing the artifact lifecycle from draft to production rollout.

スキルのインストール

Install all DataRobot skills (including the Workload API skill and the required datarobot-setup skill) with the universal skills installer:

npx ai-agent-skills install datarobot-oss/datarobot-agent-skills

デフォルトでは、インストーラーはマシン上のサポートされているすべてのコーディングエージェントにスキルをコピーします。 特定のエージェントをターゲットにするには、--agentフラグを追加します。

npx ai-agent-skills install datarobot-oss/datarobot-agent-skills --agent cursor
npx ai-agent-skills install datarobot-oss/datarobot-agent-skills --agent claude

エージェント固有のマーケットプレースまたは拡張機能カタログからインストールすることもできます。

エージェント インストール元
Claude Code claude.com/plugins/datarobot-agent-skills
Cursor cursor.com/marketplace/datarobot
Gemini CLI geminicli.com/extensions
Skillsリポジトリ github.com/datarobot-oss/datarobot-agent-skills

エージェント固有のインストールの詳細およびサポートされているエージェントの完全なリストについては、DataRobotのエージェントスキルを参照してください。

インストールの確認

インストール後、コーディングエージェントに利用可能なDataRobotスキルは何ですか？と尋ねます。

You should see datarobot-setup and datarobot-workload-api listed. The skill source and helper scripts are in skills/datarobot-workload-api/ in the repository.

環境設定

Before working with Workloads, run the datarobot-setup skill once per workspace. It checks Python and other dependencies, configures your DataRobot API token, and sets DATAROBOT_ENDPOINT and DATAROBOT_API_TOKEN for the session.

Trigger setup with either:

  • Run datarobot-setupを入力します。
  • またはスラッシュコマンド./datarobot-setupを使用します。

プロンプトに従って環境設定を完了します。 後のセッションで設定が再度実行される場合は、前提条件を最新に保つために実行を許可してください。

Use the skill in your coding agent

The skill activates automatically when you describe a Workload API task in natural language. You do not need to use a slash command or explicitly name the skill. The skill covers four areas:

Deploy, configure, and scale

Describe what you want to run and the skill handles artifact and Workload creation, environment variable injection, compute bundle selection, and autoscaling configuration.

Example prompts:

  • "Deploy ghcr.io/my-org/my-app:latest as a Workload on DataRobot listening on port 8000."
  • "Scale my agent-service Workload to 3 replicas and enable CPU autoscaling."
  • "Change the importance on prod-llm to critical."
  • "Inject my S3 credential into the data-pipeline Workload as AWS_ACCESS_KEY_ID."

Diagnose failures

When a Workload is stuck, errored, or crash-looping, the skill runs a structured diagnosis across Workload status, lifecycle events, proton state, and per-replica container conditions, then recommends a fix.

Example prompts:

  • "My rag-service Workload is stuck in launching. What's wrong?"
  • "Debug the CrashLoopBackOff on inference-server."
  • "Why is my Workload failing with ImagePullBackOff?"

Observe running Workloads

Pull logs, traces, metrics, and service statistics from a running Workload. The skill knows the OTel observability surface endpoints and the Workload API stats surface, and applies the correct unit conversions before presenting results.

Example prompts:

  • "Show me the last 100 error logs from llm-gateway."
  • "Which traces on agent-service had errors in the last hour?"
  • "What's the request error rate and p50 latency on prod-api?"

Manage the artifact lifecycle

Iterate draft artifacts, lock them for production, and roll out new container versions to running Workloads with zero downtime.

Example prompts:

  • "Lock the artifact behind staging-api and promote it to production."
  • "Roll out ghcr.io/my-org/my-app:v2 to prod-api using a rolling replacement."
  • "Clone the artifact on llm-service, update the image tag to :v3, and start a replacement."

Bundled helper scripts

The skill includes Python scripts in skills/datarobot-workload-api/scripts/ that your coding agent can run directly:

スクリプト 目的
wait_for_running.py <workload_id> Poll until running; exits non-zero on terminal failure or timeout.
diagnose_workload.py <workload_id> Run the full 5-step debug flow and print a structured diagnosis. Add --json for machine-readable output.
wait_for_build.py <artifact_id> <build_id> Poll a server-side image build; dumps the last 2 KB of logs on failure.
wait_for_replacement.py <workload_id> Poll a rolling replacement to completion.
check_limits.py Print the effective org-set concurrency and replica limits.

All scripts read DATAROBOT_ENDPOINT and DATAROBOT_API_TOKEN from the environment.

Skill 目的
datarobot-setup Install the DataRobot SDK, configure auth, and set environment variables. Required before using this skill.
datarobot-app-framework-cicd Declarative artifact and Workload management via Pulumi and CI/CD pipelines.
datarobot-external-agent-monitoring Instrument arbitrary agent code with OpenTelemetry and route telemetry to DataRobot.