ConardLi/garden-skills

Selection

Select when

• You want installable Claude Code skills rather than a separate application.
• You need one collection spanning design, presentations, images, retrieval, and editorial output.
• You want structured checkpoints before substantial creative implementation.
• You need a recordable 16:9 web presentation driven by narration beats.
• You want opinionated frontend direction with many named style recipes.
• You need GPT Image prompts that can fall back when no image tool exists.
• You have a locally indexed Markdown, PDF, or Excel knowledge directory.
• You want source material converted into a themed offline-capable article.
• You value pinned release archives and checksum-based verification.
• You can provide the runtime tools and credentials required by the selected skill.

Boundaries

Avoid when

• You need a standalone GUI or hosted service.
• You need a general-purpose Claude model gateway or MCP integration layer.
• You want unopinionated design guidance that preserves every existing convention.
• You cannot run shell, Node, Python, or frontend build tooling required by the chosen workflow.
• You need image generation without an API credential or host image tool.
• You lack a structured local knowledge base or required PDF and Excel utilities.
• You need guaranteed fully automated delivery without user approval checkpoints.
• You require independently audited security or quality assurances.
• You need a native video renderer rather than a screen-recorded web stage.
• You need a generic web application from the article workflow.

Strengths

Capabilities

Risk profile

Risks and limitations

• Uncertain: the supplied evidence omits the actual SKILL.md files, marketplace manifest, scripts, CI configuration, and release artifacts, so their behavior was not independently inspected.
• Uncertain: repository activity timestamps are dated in 2026 relative to this analysis context, so recency and maturity signals cannot be independently validated.
• The root README alternates between five skills and comments or commands saying four, while the marketplace table lists only four and omits beautiful-article.
• The image documentation disagrees on whether the case library has 17 or 18 categories and reports 160+ or 161 cases.
• Local image generation requires Node, an enabled mode, an API key, and a compatible image endpoint; host and advisor modes have reduced execution capability.
• The knowledge retriever requires a user-supplied indexed corpus plus grep, windowed file reads, PDF tooling, and pandas for the documented formats.
• The article workflow depends on shell and frontend build tooling, installs the external Reacticle package, and may require Python or a headless browser for optional paths.
• The presentation workflow still requires a browser, screen recorder, and provider-specific credentials or tools for optional narration.
• Named design recipes and anti-cliché rules are intentionally opinionated and may conflict with established brand systems or project constraints.
• External APIs, npm packages, hosted assets, websites, and TTS providers can change or expose submitted content under their own policies.
• Repository-authored compatibility, production-readiness, and test claims were not independently verified from the supplied files.
• No independent security audit, automated compatibility report, or output-quality benchmark is present in the supplied evidence.

Release archives are documented with SHA-256 checksums, but bundled scripts can execute shell, Node, Python, npm, browser, TTS, and API operations. Image credentials may be read from process variables and local env files, including a home-directory gateway file. Inference: users should review skill code and pin releases before use in sensitive work; no independent audit is evidenced.