Code

**Quick judgment:** A solid, conservative coding-workflow skill focused on planning, guidance, and verification rather than automation. Best suited for users who want structured, test-driven development assistance and are comfortable executing code and tests themselves. **What it does well** - Enforces a clear workflow: request → plan → execute (by user) → verify → deliver. - Promotes smaller, testable changes and explicit verification of functions and UI changes. - Uses a tightly scoped local memory (`~/code/memory.md`) and only stores preferences when the user explicitly asks, which is good for privacy and predictability. - No code execution, no network access, and no self-modification of skill files, which reduces side-effect risk and surface area for failure. **Key limitations / risks** - Provides guidance only; it never executes code or tests. Users must correctly follow instructions and run commands themselves, which can introduce human error. - Memory is limited to explicit user preferences; it does not learn project context automatically, so users may need to repeat project-specific details. - The structured workflow can add overhead for users who want quick, one-off snippets rather than a full plan/verify cycle. - No integration with external tools or services (CI, issue trackers, code hosts), which may limit utility in highly automated pipelines. **Recommended scenarios** - You want a “safety-first” coding assistant that won’t run or deploy anything on its own. - You’re working on code where test coverage and verification are important, and you’re willing to follow a disciplined workflow. - Educational or mentoring contexts where clean, stepwise software development practices are a priority. - Environments with strict security or compliance requirements that prohibit automatic code execution and networked tools.