What does this skill do, and when should you use it?
This Agent Skill from Stripe’s ai repository helps platform businesses choose a Stripe Connect integration shape. It targets marketplaces, multi-vendor stores, gig platforms, subscription platforms, and SaaS products that route money between parties. It gathers business context, scans project signals, validates configuration compatibility, and produces a recommendation plan. It also identifies relevant embedded components and implementation priorities.
Asks sequential questions about the business, parties, money flow, account management, risk, and fee structure; scans the project for existing Connect plans and patterns involving accounts, transfers, payouts, charges, fees, and webhooks; maps findings to dashboard, responsibility, fee-ownership, and charge-pattern recommendations; checks blocked and cautionary combinations; recommends embedded components for onboarding, notifications, account management, payments, disputes, payouts, and reporting; generates a recommendation plan and can write it to connect-recommend-plan.md at the project root.
- A marketplace needs to pay sellers or service providers.
- A multi-vendor store needs split payments or revenue distribution.
- A gig, booking, invoicing, or similar platform routes money between parties.
- A SaaS product has independent sellers who own customer relationships and needs Connect onboarding and payment guidance.
- A team needs to evaluate merchant onboarding, KYC, connected-account dashboards, or payment responsibility settings.
How do you install this skill?
- The skill sends user-provided company URLs/descriptions to external services (WebFetch/WebSearch); data-sensitive users should be aware of this flow before use.
- All core content depends on Stripe overseas services and English documentation; reachability of WebFetch/WebSearch and some Stripe pages from mainland-China networks is unverified and may hinder Chinese users.
- BLOCKED/CAUTION rules in the compatibility matrix are author-compiled; static review cannot confirm they match Stripe's current GA policy — re-check against official docs before implementing.
- The skill files carry no version number or changelog; compare repository revision diffs when updating.
- Publisher is unverified in the FollowSkills registry; identity is treated as unknown, which is neither a deduction nor a trust signal.
- Shell / CLI
- Network access
- Local filesystem
This skill is one of 24 skills in stripe/ai. The README documents this manual installation command: npx skills add https://docs.stripe.com. It also documents official Stripe plugin installation for Claude Code, Codex, and Cursor. Manually installed skills do not auto-update; the README says to run npx skills update -y. No standalone installation command for connect-recommend is documented.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/stripe/ai.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/providers/grok/plugin/skills/connect-recommend" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- We are building a marketplace connecting customers with freelancers and need to pay the providers.
In an Agent with the skill installed, describe the business or provide a company URL, for example: “We are building a marketplace connecting customers with freelancers and need to pay the providers.” The skill then asks questions, scans the project, and creates a Connect recommendation plan. The source does not document a more specific host-level trigger format.
What are this skill's strengths and limitations?
- Covers business discovery, project scanning, configuration recommendations, compatibility checks, and plan generation.
- Targets concrete marketplace, SaaS, and multi-party payment decisions.
- Addresses margin economics, negative-balance liability, disputes, onboarding, and embedded components.
- The repository is licensed under MIT.
- The skill explicitly relies on AskUserQuestion, Task, and reference files, so hosts without equivalent support may require adaptation.
- The source provides no test suite, version compatibility matrix, or independent performance evidence.
- The source does not document a standalone installation flow for this individual skill.
- Recommendations depend on project scans and user answers; the source does not define exact failure behavior when those inputs are unavailable.
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Stripe Connect Integration Advisor this page | 57 · Use with care | ★ 1.9k | 12d ago | MIT |
| Stripe Integration Best Practices | 53 · Use with care | ★ 1.9k | 12d ago | MIT |
| Bambu Labs Print Handoff Skill | 60 · Recommended | ★ 20k | 14d ago | MIT |
| NeMo AutoModel Model Onboarding ✓ NVIDIA · Official | 60 · Recommended | ★ 3.6k | 11d ago | Apache-2.0 |
| Learning Opportunities | 59 · Recommended | ★ 2.5k | 2mo ago | CC-BY-4.0 |
The README documents official plugin installation for Claude Code, Codex, and Cursor, as well as manual installation. It does not provide a feature comparison with another Connect recommendation tool.
How did FollowSkills review this skill?
Pure advisory recommendation flow: only reads local reference files, scans the codebase, and fetches a user-provided URL. No destructive operations, no credential exposure; auto-actions are limited to low-cost steps (plan generation, scan, reading references), every decision point requires numbered user confirmation, and the fee-economics check is mandatory. Deductions: WebFetch/WebSearch send user business data externally without explicit data-flow disclosure; no rollback concept (low impact here); publisher unverified, attribution limited to the repository itself.
SKILL.md and five reference files are internally consistent: decision matrix, compatibility matrix and template cross-reference coherently; BLOCKED/CAUTION/SALES-GATED rules align across files; failure paths (research unavailable, URL 403, conflicting signals) have clear degradation and messaging. Deductions: static review cannot execute anything; no test suite or CI evidence covering key paths; actual WebFetch-failure behavior depends on the host agent, so determinism is unproven — capped at the static anchor.
Trigger conditions are very specific (Connect configuration, split payments, marketplaces, KYC), with explicit non-fit exclusions (non-payment-routing SaaS, not-connect businesses); boundaries (on_behalf_of, OAuth, international expansion) are declared out of scope with redirect paths. Deductions: content depends entirely on Stripe overseas services and English docs, no Chinese-language support; mainland-China network reachability is undeclared, and company-researcher depends on WebFetch/WebSearch availability that is uncertain in the target environment.
Well-layered docs (SKILL.md main flow plus references with progressive disclosure), mature terminology rules, concise output and decision-point design, MIT license via the repository, and version awareness of v1/v2 API. Deductions: no skill-level version number, changelog, known-limitations list, or FAQ; maintenance responsibility is only inferable at the repository level, with no explicit skill-level update path.
Output usability is well designed: structured recommendation plan, optional write to connect-recommend-plan.md, clear code-side vs Dashboard responsibility split — clear marginal value over manually reading Stripe docs. Deductions: static review cannot verify actual correctness of recommendations; correctness rests on compatibility assertions in the references (e.g., blocked rules for Express + stripe losses) that are not independently corroborated, so the anchor cap applies.
Key claims link to official Stripe docs (docs.stripe.com, stripe.com/pricing); facts vs inference are separated via HIGH/MEDIUM/LOW confidence levels and required source lists. Deductions: no execution evidence or test reproducibility; whether parts of the compatibility matrix match Stripe's current GA policy cannot be verified statically; third-party corroboration is thin.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →