Cost
Function usage rose unexpectedly
Check usage and route metrics before attributing the bill change to a particular endpoint or cache setting.
01 / The skill · Vercel · Cost and performance audit
Vercel Optimize: investigate what the metrics point to
A Vercel cost and performance Skill that collects production signals before reading code, then narrows investigation to routes and files backed by evidence.
Publisher, source, and installation
Published by Vercel Labs · Type: Metrics-led audit Skill with scripts
Independent guide. Vercel Labs has not endorsed Floatboat.
Upstream installation
npx skills add vercel-labs/agent-skills --skill vercel-optimizeInstalls upstream instructions and bundled resources; Vercel CLI access and Observability Plus may still be required.
A realistic request
The source prescribes an observability-first audit, not a repository-wide optimization sweep. Its scripts collect and merge Vercel signals, gate investigations, and produce ranked recommendations; route-level advice depends on access and observability prerequisites.
Ask the agent
“Investigate last month's Vercel function-cost increase. Confirm the project scope, use available metrics, and rank only evidence-backed opportunities.”
A useful result
A useful audit names measured routes, supporting signals, bounded code locations, prerequisites, and uncertainty; it does not invent a savings figure.
When Vercel Optimize fits
Use it for a deployed Vercel project with a specific cost or performance question and accessible production signals.
Routes
One page seems expensive or slow
Let measured route signals identify which source files and settings deserve investigation.
Cache
A cache change is being proposed
Confirm the current hit pattern and framework behavior before recommending ISR, middleware, or response-cache changes.
Where evidence may be missing
- A precise savings claim when metrics, usage, or project scope cannot be verified.
- An unrestricted code sweep of an unlinked or unsupported project.
How a metrics-first audit proceeds
Project scope and signals come first; code reading follows only candidates selected by the gate.
- 01
Resolve project scope
Identify the linked project and owner before collecting metrics or usage.
- 02
Collect and gate signals
Use the bundled collection and gate workflow to select measurable candidates.
- 03
Inspect and rank
Read only the candidate's route-local files, then present verified, version-aware recommendations.
02 / Practical guide
Evidence before optimization advice
Its gate narrows work to measurable candidates. A missing metric or ambiguous project scope is a limitation to report, not a cue to invent savings.
Why does Vercel Optimize start with metrics?
The source explicitly delays source-code inspection until collected signals and a deterministic gate point to a candidate. That helps distinguish high-cost routes from merely suspicious code patterns.
What access is needed for route-level recommendations?
The source lists a linked Vercel project, supported CLI and Node versions, correct account scope, and Observability Plus for route-level metric-backed advice. Without those, the audit must state the limitation.
What you can take away
- A signal-backed candidate list
- Ranked cost or performance recommendations
- Explicit blockers and data limitations
What to check first
- Is the Vercel project and owner unambiguous?
- Are route metrics available for the desired claim?
- Does the framework fall within supported scope?
Explore More Agent Skills
Read another Skill guide and check its original source before using it.
Frequently Asked Questions
Does Vercel Optimize scan the entire repository first?
No. Its documented doctrine collects production signals first and gates code investigation to metric-backed candidates.
Can it estimate savings without Observability Plus?
Route-level metric-backed recommendations require the stated observability access. Missing data should be reported as a limitation, not replaced with invented precision.
Which frameworks does it address?
The source lists strong support for Next.js, support for SvelteKit and Nuxt, limited Astro support, and a restricted path for others. Check the current source before running an audit.
Is it only a prose checklist?
No. The Skill bundles scripts and references for signal collection, gating, and report generation. Those still depend on authorized Vercel access.
03 / Put it to work · Prioritize the work
Turn verified bottlenecks into bounded tasks
Bring the metric, affected route, and proposed change into Floatboat; keep the original Vercel audit source and access requirements visible.