Stripe Connect Integration Advisor

The skill requires AskUserQuestion at key decision points and includes warnings for compatibility, fees, disputes, and negative balances. However, it permits broad tools such as Bash, Write, and Edit, while providing limited controls for payment/KYC data flows, sensitive-data handling, rollback, and least privilege. No malware, credential theft, or covert exfiltration is evident, so no red line is triggered; incomplete confirmation and data governance reduce the score.

The workflow is structured and includes compatibility checks, abnormal-combination warnings, and dispute transfer-reversal guidance. Static evidence contains no executed tests or reproduced key paths. There are also internal inconsistencies, including differing treatment of on_behalf_of and partially inconsistent guidance for custom onboarding with dashboard:none, so the score is conservatively limited and reduced.

Triggers, target users, and marketplace, SaaS, subscription, and multi-party payment scenarios are relatively clear, and the skill distinguishes direct, destination, and separate flows. It does not adequately define non-fit boundaries, Chinese-language interaction support, or mainland-China reachability; complex product cases are sometimes deferred broadly to documentation or sales, so points are deducted.

The skill has useful progressive structure, stepwise workflow, decision matrices, compatibility rules, examples, and limitation disclosures, and the supplied metadata states an MIT license. It lacks a clear version, changelog, named maintenance responsibility, or update path, and assumes reference files and external documentation without installation or dependency guidance.

The skill can produce a Connect configuration recommendation and covers account responsibilities, charge patterns, fee economics, and implementation concerns, giving the core output practical value. However, no statically verifiable representative outputs, test evidence, or comparative benefit over manual documentation review are provided; complex or potentially stale rules still require substantial human review, so the score is constrained by static calibration.

The source includes detailed decision tables, code examples, compatibility rules, and an auditable reasoning chain. Evidence is still primarily self-authored skill material, with no committed test suite, CI execution evidence, or independent third-party corroboration, so the score remains within the static limit.