Skip to content

Unique-AG/ai / .claude/skills/model-deployment

model-deployment

Verified

Deploy 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.

references/TOOLKIT-REGISTRY.md

View on GitHub
# Track C — AI toolkit details

## LanguageModelInfo code example

In `unique_toolkit/unique_toolkit/language_model/infos.py`:

### 1. Enum entry

Convention:
- Azure models: `AZURE_GPT_54_2026_0305 = "AZURE_GPT_54_2026_0305"`
- LiteLLM models: `OPENAI_GPT_54 = "litellm:openai-gpt-5-4"`

### 2. `LanguageModelInfo.from_name()` case

```python
case LanguageModelName.OPENAI_GPT_54:
    return LanguageModelInfo(
        name=model_name,
        provider=LanguageModelProvider.LITELLM,
        version="2026-03-05",
        token_limits=LanguageModelTokenLimits(
            token_limit_input=200_000,
            token_limit_output=100_000,
        ),
        capabilities=[
            ModelCapabilities.FUNCTION_CALLING,
            ModelCapabilities.STREAMING,
        ],
        temperature_bounds=TemperatureBounds(
            min_temperature=0.0, max_temperature=1.0
        ),
        supported_reasoning_efforts=[
            "low", "medium", "high", "xhigh",
        ],
        default_options={"reasoning_effort": "medium"},
    )
```

## Field reference

| Field | Description |
|-------|-------------|
| `token_limits` | Input and output token limits — cite the model card |
| `capabilities` | `FUNCTION_CALLING`, `STREAMING`, `VISION`, etc. |
| `temperature_bounds` | Min/max temperature — reasoning models typically use `(0.0, 1.0)`. Never use `(1.0, 1.0)` — this was a past mistake that locked temperature to 1.0 regardless of reasoning state. |
| `supported_reasoning_efforts` | `None` = unknown (pass-through), `[]` = no reasoning, `["low", "medium", "high"]` = validated list |
| `default_options` | Default `reasoning_effort` and other defaults. **Critical: use the string `"none"` — never Python `None`.** See [LESSONS-LEARNED.md](LESSONS-LEARNED.md). |

## LANGUAGE_MODEL_INFOS env override (early exposure)

Before a toolkit release is published, you can expose a model immediately by setting the `LANGUAGE_MODEL_INFOS` environment variable on the relevant service. Useful for smoke-testing a model in QA before the toolkit PR is merged.

### Format

JSON dict where each key is a model identifier and each value is a `LanguageModelInfo`-compatible dict:

```bash
LANGUAGE_MODEL_INFOS='{"MY_NEW_MODEL": {"name": "litellm:my-new-model", "provider": "LITELLM", "version": "2026-01-01", "capabilities": ["function_calling", "streaming"], "token_limits": {"token_limit_input": 128000, "token_limit_output": 16384}}}'
```

### Where to set it

Set the env var in the service's app configuration or Helm values overlay for the target environment. The toolkit parses this at startup and merges it with the hardcoded model registry.

### Caveats

- This is a **temporary** measure — always follow up with a proper toolkit PR.
- Models loaded via env have `supported_reasoning_efforts=None` (unknown / pass-through) unless explicitly set in the JSON.
- If `default_options.reasoning_effort` is set but `supported_reasoning_efforts` is empty, the model's default effort will not be validated.
writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deploymentStars: 322kUpdated 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