update-provider-models
by vercelupdate-provider-models helps update provider model IDs in vercel/ai. Use it to add new models or remove obsolete ones with exact-match search, ordering checks, and full codebase updates across related files.
This skill scores 78/100, making it a solid directory listing for users who maintain model catalogs inside the Vercel AI SDK. It is clearly scoped to adding and removing provider model IDs, includes trigger phrases and workflow guidance, and should help an agent execute with less guesswork than a generic prompt. Users should still expect a documentation-only workflow without bundled automation or install-specific setup help.
- Clear triggerability: the description explicitly targets "add model", "remove model", "new model ID", and similar requests.
- Operationally useful: the skill defines separate workflows for adding and removing model IDs and includes concrete rules like exact matching, preserving sort order, and sequential processing.
- Repository-grounded guidance: it references searching the codebase and affected provider package locations, which gives agents real execution direction beyond generic advice.
- No support files or scripts are included, so execution depends on the agent correctly following prose instructions rather than running packaged automation.
- There is no install command or quick-start setup guidance in SKILL.md, which limits immediate adoption clarity for directory users.
Overview of update-provider-models skill
What update-provider-models does
The update-provider-models skill is a focused maintenance workflow for adding new model IDs or removing obsolete ones in the Vercel AI SDK codebase. It is built for code editing tasks where a plain “update the models” prompt is too risky because model IDs often appear in multiple packages, typed lists, tests, and examples.
Who should use this skill
This update-provider-models skill is best for contributors working on provider support in vercel/ai, maintainers processing provider model changes from issues, and agents asked to make model-list edits without missing secondary update locations.
The real job-to-be-done
Users usually need more than a single string replacement. The real task is to identify all exact references for one model, update the right provider package and shared surfaces, preserve sort order, avoid substring mistakes, and complete one model change cleanly before moving to the next.
Why it is better than a generic edit prompt
The main value of update-provider-models for Code Editing is reduced guesswork. The skill gives a workflow with explicit constraints like exact-match searching, sequential handling of multi-model requests, and naming rules for examples. That matters because mistakes here are usually subtle: wrong provider surface, broken alphabetical order, or accidental edits to similarly named models.
Best-fit and poor-fit use cases
Use it when:
- adding a new model ID to an existing provider
- removing an obsolete model ID
- processing a provider issue with several model additions/removals
Do not expect it to design a new provider integration from scratch. This skill is about updating existing provider model inventories, not building a new adapter architecture.
How to Use update-provider-models skill
How to install update-provider-models
A practical update-provider-models install command is:
npx skills add vercel/ai --skill update-provider-models
If your environment already exposes installed skills, you can then invoke it by naming the skill and passing the concrete provider and model change request.
What inputs the skill needs
The skill works best when you provide:
- the provider name
- whether each model is being added or removed
- exact model IDs
- the source of truth, such as an issue, changelog, or provider docs
- whether this is a single change or a batch
Weak input:
- “Update OpenAI models”
Strong input:
- “Use
update-provider-modelsto addgpt-4.1-minito the existing provider model lists and examples invercel/ai. Check all exact matches only, preserve existing ordering, and show me every file changed.”
Turn a rough request into a strong prompt
A good update-provider-models usage prompt usually includes scope, validation, and output expectations:
Use update-provider-models for Code Editing in vercel/ai.
Provider: [provider name]
Changes:
- add [exact-model-id]
- remove [exact-model-id]
Requirements:
- search for exact matches only
- update all relevant provider and shared locations
- preserve current sort/order conventions
- if examples are needed, use kebab-case filenames
- process one model completely before starting the next
- summarize why each file changed
This format reduces the most common failure mode: partial edits that update one list but miss another package or test.
What to read first in the repository
Start with:
skills/update-provider-models/SKILL.md
Because this repository evidence points to a single-file skill, SKILL.md is not just an overview; it is the operating spec. Read the “Critical Rules” section first, then the add/remove workflows.
The most important workflow rules to keep in mind
The upstream skill emphasizes a few rules that materially change output quality:
- use exact matching because model IDs can be substrings of others
- respect existing sort order in arrays, unions, and tables
- use kebab-case filenames for examples, with dots replaced by hyphens
- handle multi-model requests sequentially, one full workflow at a time
- update both the primary provider package and related AI SDK surfaces when required
These are the details a generic prompt often misses.
How to handle add vs remove requests
For additions, treat the task as discovery plus insertion:
- identify the exact provider package and all model-list surfaces
- find insertion points that preserve ordering
- update any related examples or docs patterns if the repository convention expects them
For removals, treat the task as cleanup:
- confirm the model is obsolete and target the exact ID only
- remove references without touching similarly named models
- check whether examples, tests, or typed unions still mention it
How to process a batch safely
The skill explicitly prefers sequential processing. If an issue contains many model changes, complete the full workflow for one model before starting the next. This avoids mixed diffs and makes it easier to catch a missed file while the search context is still narrow.
Practical search strategy that avoids bad edits
When using update-provider-models, do not rely on naive substring search alone. Search exact strings, then inspect each hit to verify it refers to the same model and provider context. This matters for names like grok-3 versus grok-3-mini, where a fast replace can corrupt valid neighboring entries.
What good output looks like
A strong result from the update-provider-models guide should include:
- all relevant files updated, not just one obvious list
- correct ordering within each edited structure
- no accidental changes to similar model IDs
- a clear summary of additions/removals by file
- confidence checks or notes where ambiguity remained
When this skill saves the most time
This skill is most valuable when:
- the same model ID appears across multiple typed surfaces
- you are editing under time pressure from a provider update issue
- you are delegating the task to an agent that needs explicit guardrails
- the provider has many similarly named model variants
update-provider-models skill FAQ
Is update-provider-models only for vercel/ai?
It is written for the vercel/ai repository structure and conventions. You can adapt the workflow elsewhere, but the highest-confidence use case is updating model IDs inside that codebase.
Is this better than asking an agent to “search and replace”?
Yes, if correctness matters. update-provider-models skill is designed around repository-specific risks: exact-match verification, list ordering, cross-surface updates, and sequential processing. A generic prompt may edit faster, but it is more likely to miss dependent files or damage similarly named entries.
Is update-provider-models beginner-friendly?
Yes, if you can already navigate a repository and review diffs. The workflow is narrow and concrete. Beginners still need to verify provider scope and inspect search results carefully, especially for near-duplicate model names.
Can I use it for a brand-new provider?
Usually no. This is not the right first tool for creating a provider package from scratch. It is meant for updating model IDs for an existing provider integration.
What if I only have a provider issue and no exact file list?
That is a normal fit. The skill assumes you will discover affected locations through search. Just provide exact model IDs and whether each one is being added or removed.
When should I not use update-provider-models?
Skip it when the request is really about:
- designing provider architecture
- changing model capabilities or behavior logic unrelated to IDs
- broad docs rewrites without code changes
- speculative updates where the model names are not confirmed
How to Improve update-provider-models skill
Give exact model IDs and a source of truth
The fastest way to improve update-provider-models results is to provide exact model strings and the provider announcement, docs page, or issue link they came from. That lets the agent distinguish confirmed IDs from guesses.
State the provider package explicitly
If you already know the affected provider, say so. “Add model-x” is weaker than “Add model-x to the Anthropic provider surfaces in vercel/ai.” This reduces false positives and unnecessary searching.
Ask for a per-file change summary
One of the best output-quality upgrades is requesting a short rationale per file changed. That makes review easier and quickly exposes incomplete work, especially when one expected surface was missed.
Force exact-match validation in the prompt
Include an instruction like:
- “Verify every search hit is an exact model match, not a substring variant.”
This directly targets the repository’s biggest correctness risk.
Require ordering checks after edits
Because the skill depends on preserving existing sort conventions, explicitly ask the agent to re-check list order after insertion or deletion. This catches low-visibility defects that often slip through diff review.
Handle multi-model updates one by one
For batch issues, tell the agent to finish one model completely before moving to the next. This keeps the search context precise and makes rollback or review much simpler.
Ask for missed-surface verification
A strong follow-up instruction is:
- “After edits, do one final search for the exact model ID to confirm no outdated references remain and no unintended references were changed.”
That final sweep materially improves trust in the output.
Common failure modes to watch for
Typical problems with update-provider-models usage are:
- editing substring matches by mistake
- adding a model in one list but not all required surfaces
- breaking alphabetical or repository-specific order
- naming example files incorrectly when dots are present
- mixing multiple model updates into one confusing pass
How to iterate after the first draft
If the first output looks incomplete, do not re-prompt vaguely. Ask for a targeted second pass:
- “Re-run update-provider-models for
model-idand verify all exact matches across provider package, shared types, tests, and examples. List any files considered but intentionally unchanged.”
That kind of iteration is much stronger than “double-check this.”
A practical review checklist
Before merging, confirm:
- the provider is correct
- each model ID change is exact
- ordering still matches local conventions
- examples or tests follow filename rules if added
- no obsolete references remain after removals
That review pattern is the most reliable way to get high-confidence results from update-provider-models for Code Editing.