jakubkrehel/skills

Selection

Select when

• You want a Claude Code-native plugin focused specifically on product-interface quality.
• You need one consolidated review spanning accessibility, layout, copy, typography, color and polish.
• You need interface-focused review of uncommitted work, a branch or a pull request.
• You want design alternatives built into the real page for human comparison.
• You want to expose one component to realistic hostile content and state combinations.
• You prefer prescriptive values and concrete verification checks over broad design advice.
• You need skills that preserve the target project's existing styling system and conventions.
• You want user invocation required before a skill writes throwaway exploration or test-harness code.

Boundaries

Avoid when

• You need general code correctness, security, test or backend review.
• You need automated screenshot comparison or a hosted visual-regression platform.
• You cannot allow an agent to create temporary UI routes or harness files.
• You want an agent to autonomously select a subjective design winner.
• You need evidence of comprehensive automated testing for every rule and reference.
• Your product intentionally conflicts with the collection's opinionated design prescriptions.
• You need guaranteed browser verification in an environment without an available preview.
• You need a standalone application rather than guidance loaded into an agent host.

Strengths

Capabilities

Risk profile

Risks and limitations

• Uncertainty: only selected first-party files were supplied, so better-accessibility, better-typography, explain-interface and supporting references were not directly inspected.
• Uncertainty: no comprehensive inventory confirms that every referenced supporting file or per-skill OpenAI policy file is present and consistent.
• The repository explicitly has no build, lint or test tooling, so content integrity is not shown as automatically enforced.
• No supplied evidence demonstrates the skills' review accuracy, visual outcomes or compatibility across host versions.
• Several workflows depend on project access, framework conventions and human visual judgment.
• Rendered verification may be skipped or marked Not verified when a browser or cheap preview is unavailable.
• Variant and break deliberately write temporary code, which requires cleanup discipline and repository-write permission.
• The rules are intentionally prescriptive and may conflict with an established product design language.
• The plugin installs every repository skill together; selective installation is documented only through the separate skills CLI path.
• External host behavior, browser APIs, libraries and plugin update semantics may change beyond this repository's control.
• Security coverage is explicitly out of scope for interface-review.
• Privacy behavior is uncertain because no telemetry policy or exhaustive external-data-flow documentation was supplied.

The supplied evidence describes a documentation-only repository with no build, lint or test tooling. Reviews are read-only by default, and change review forbids checkout or working-tree mutation; however, variant and break intentionally create temporary project files when explicitly invoked. Security review is outside interface-review's scope. Users should inspect skill text and resulting edits before use because the skills can influence code changes and browser-facing workflows.