Skip to content

datarobot-oss/datarobot-agent-skills / skills/datarobot-agent-assist/agent-assist-build

datarobot-agent-assist-build

Flagged

Use when the user wants to design, build, code, or deploy an AI agent on DataRobot; mentions agent_spec.yaml, dress rehearsal, the DataRobot agent template, LangGraph, CrewAI, LlamaIndex, NAT, Base agents, MCP servers, backend APIs, custom frontends, or the DataRobot CLI.

references/llm-selection.md

View on GitHub
## LLM Selection

`list_llm_models.py` returns two kinds of entry, told apart by `source`:

- `gateway` — a model in the DataRobot LLM Gateway catalog. Selected by its `llm_default_model`.
- `deployed` — an existing DataRobot text-generation deployment. Selected by its `deployment_id`.

**Copy `llm_default_model` verbatim.** It is the only field that goes into `agent_spec.yaml` and `.env`. A `gateway` entry also carries an `id` (the catalog's `llmId`, e.g. `azure-openai-gpt-5`) and an `api_model` (e.g. `azure/gpt-5-2025-08-07`); neither works as a model name. The gateway rejects the `id` outright, and `api_model` is missing the `datarobot/` prefix that routes the request to DataRobot rather than straight to the provider. `setup_template.py` refuses an `id` rather than letting it reach `.env`. On a `deployed` entry, `id` is the deployment id instead, and it is what you want: see below.

### Recommending

- Prefer a `gpt-5`, `claude-4-5`, or `gemini-2.5` gateway model unless the user gives cost or other constraints.
- If none of those families appear, take the highest-capability gateway model by name: prefer `large`, `pro`, `opus`, `sonnet` over `mini`, `haiku`, `flash`.
- **If there is no `gateway` entry at all**, the LLM Gateway is disabled or empty on this instance. That is the normal shape of an on-prem install, not an error. Recommend a `deployed` entry instead, name it by its `name` (the deployment label), and tell the user that is what you are doing and why.
- Only show the full catalog when the user explicitly asks to browse.

### Recording a deployed LLM in `agent_spec.yaml`

Every deployment reports the same `llm_default_model`, the placeholder `datarobot/datarobot-deployed-llm`. It identifies the *source*, never one deployment. So a deployed choice is a pair, and both fields are required:

```yaml
model: "datarobot/datarobot-deployed-llm"
llm_deployment_id: "6a43eb5f10dbecadbebc5b2b"
```

Take `llm_deployment_id` from the entry's `deployment_id`. Without it nothing downstream can tell which deployment was chosen: `setup_template.py` refuses the pair outright, and the dress rehearsal would have to guess.

A gateway choice sets `model` only and leaves `llm_deployment_id` out.

### What the choice changes downstream

| Step | Gateway model | Deployed LLM |
|---|---|---|
| [Template setup](helper-scripts.md#setup_templatepy) | `--llm-model` | `--llm-model` **and** `--llm-deployment-id` |
| `.env` written | `LLM_DEFAULT_MODEL` | plus `LLM_DEPLOYMENT_ID`, `INFRA_ENABLE_LLM=deployed_llm.py`, `USE_DATAROBOT_LLM_GATEWAY=0` |
| [Dress rehearsal](dress-rehearsal.md) | LLM Gateway chat endpoint | the deployment's own chat endpoint |

On the deployed path the deployment id is the only thing that selects the model. `dr dotenv setup` rebuilds `.env` from the template's own prompt schema, which does not carry `LLM_DEFAULT_MODEL` for that path, so a real model name passed alongside an id is not persisted. Routing is unaffected.

### Caveats worth telling the user

- A deployment is offered when it is active and its target type is text generation. Nothing checks that it answers chat requests, and some text-generation deployments (guard models, for instance) do not. If the rehearsal reports the deployment as unavailable, that is the likely cause.
- Deployed entries carry no provider or context window, so those columns read `-`.
- The `Deployment ID` column appears in the table only when a deployed entry is present.
- Deployed entries require `dr` v0.2.79 or newer to appear in the CLI listing. On an older `dr` the script falls back to a direct API call, so they still list.
writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deploymentStars: 323kUpdated Claude CodeCodexVerifiedInstall
xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy.Stars: 119kUpdated Claude CodeCodexVerifiedInstall
childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release.Stars: 118kUpdated Claude CodeCodexVerifiedInstall