Unique-AG/ai / .claude/skills/model-deployment
model-deployment
VerifiedDeploy a new LLM model via LiteLLM (GitOps) and/or Azure (Terraform + node backend), add it to the ai toolkit, create the required pull requests, and verify end-to-end. Use when rolling out a model to any environment, adding a model to the toolkit, troubleshooting "model not available" errors, or finding pricing/token limits/model ids for any provider.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 10.3 KB |
| references/AZURE-NODE-CHAT.md | markdown | 4.4 KB |
| references/LESSONS-LEARNED.md | markdown | 4.7 KB |
| references/LITELLM-CONFIG.md | markdown | 2.5 KB |
| references/MODEL-SELECTABILITY.md | markdown | 2.7 KB |
| references/TOOLKIT-REGISTRY.md | markdown | 2.9 KB |
references/AZURE-NODE-CHAT.md
View on GitHub# Track B — Azure + node-chat details
## Terraform (infrastructure repo)
Terraform lives in the **infrastructure** repo. Look for existing `azurerm_cognitive_deployment` resources for the same Azure OpenAI account and copy the pattern:
1. Add or update `azurerm_cognitive_deployment` resource(s) for the new model — one per deployment (e.g. base + chat variants). Key properties: `model.name`, `model.version`, `sku.capacity`.
2. Update the Key Vault secret holding endpoint definitions (commonly `azure-openai-endpoint-definitions`) so node-chat can discover the new deployment at startup.
3. If the model requires a new API key or endpoint, add a corresponding Key Vault secret reference.
**Tip:** search the infrastructure repo for an existing model deployment (e.g. `gpt-5.4`) to find the exact Terraform module and variable pattern to replicate.
## Azure quota (`sku_capacity`)
`sku_capacity` is the deployment's share of a live Azure quota bucket. The bucket is **subscription + region + SKU** (`GlobalStandard` or `DataZoneStandard`). Two rows in the same subscription and the same region draw from one limit, even when they live in different tenant files. unique dev and unique qa/lz are that case: both Sweden Central, subscription `fb8bce09`.
Set each row to `min(requested capacity, remaining quota)`. If several new deployments share a bucket, their capacities must **sum** to no more than the limit. Split the way the sibling model already does when the bucket is shared (GPT-6.1 Sol is 7500/7500 on dev and qa against a 15000 Global Standard limit).
1. Region is the `oai_deployments_<region>` key the row sits under.
2. Subscription id is that workspace's Terraform `subscription_id` (Atlantis workspace / tfvars).
3. Read the live limit before writing the number. `currentValue` is what is already deployed; remaining is `limit - currentValue`. A new model is often `currentValue = 0`, so the row can use the full limit.
```bash
az cognitiveservices usage list --location <region> \
--subscription <subscription-id> \
--query "[?contains(name.value,'<model-name>')].{quota:name.value,used:currentValue,limit:limit}"
```
GPT-6.1 Terra (checked 2026-11-03) asked for 15000 Global Standard and 5000 Data Zone Standard. Live limits forced these clamps: us01 Global Standard 10000; skyline, sparkles, teapot, ch01, and tree/evx Data Zone Standard 3333; dev and qa 7500 each because they share one 15000 bucket. Rows already inside the limit stayed at the request. If a subscription is not available from the repo, say so in the PR and do not invent its capacity.
## Node-chat registration — code examples
Three files in `next/services/node-chat/src/openai/`:
### 1. `language-model.enum.ts`
Add to `LanguageModel` enum and `LanguageModelNameKey`:
```typescript
// In LanguageModel enum:
AZURE_GPT_55_2026_0901 = 'AZURE_GPT_55_2026_0901',
// In LanguageModelNameKey:
AZURE_GPT_55_2026_0901: 'gpt-5.5-2026-09-01', // EU, US ✅
```
### 2. `azure-sdk-openai.service.factory.ts`
Add a `case` in the factory `switch` and add the model to `UNIQUEAI_SUPPORTED_MODELS`:
```typescript
// In UNIQUEAI_SUPPORTED_MODELS array:
LanguageModelNameKey.AZURE_GPT_55_2026_0901,
// In the factory switch:
case LanguageModel.AZURE_GPT_55_2026_0901:
return new AzureSdkOpenAIService(
enableV1,
this.getConfigOfModel({
modelKey: 'gpt-5.5-2026-09-01',
maxToken: 1000000,
apiVersion: '2024-10-21',
}),
this.azureSdkOpenAICommonService.getTokenCount,
this.authTokenService,
openAIClientOptions,
);
```
### 3. `azure-sdk-openai.service.ts`
If the model needs special parameter handling (e.g. reasoning params), update the `getDeploymentName()` switch to map the `modelKey` to the Azure deployment name.
## Troubleshooting
| Symptom | Likely cause |
|---------|-------------|
| `reasoning_effort does not support null` | `default_options` in `LanguageModelInfo` uses Python `None` instead of string `"none"`. See [LESSONS-LEARNED.md](LESSONS-LEARNED.md). |
| Works in plain chat but fails in agentic space | Bug in the Python toolkit's parameter construction — check `responses_api.py` reasoning handling and the toolkit version pinned in `assistants-core`. |
| Works in agentic space but fails in plain chat | Bug in node-chat TypeScript Azure SDK service — check `azure-sdk-openai.service.ts` option spreading. |
| Model works on master but not on release branch | Cherry-pick was incomplete — check that `assistants-core` toolkit version was bumped alongside node-chat changes. |
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deployment | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| 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: 198k | Updated | Claude CodeCodex | Verified | Install |
| 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: 119k | Updated | Claude CodeCodex | Verified | Install |
| 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: 118k | Updated | Claude CodeCodex | Verified | Install |