Coding

**Quick judgement** A focused, privacy-preserving coding-style memory that reliably enforces *explicitly stated* preferences across sessions. It’s best suited for users with clear conventions who are willing to correct the assistant and confirm what should be remembered. **What it does well** - Remembers stack, style, and structural preferences in a compact local file (`~/coding/memory.md`). - Only learns from explicit corrections and confirmations, reducing unwanted or incorrect inferences. - Applies stored preferences at session start for consistent code output over time. - Supports queries like “show my coding preferences” and targeted forgetting of entries. **Key risks / limitations** - Requires user effort: you must actively correct and confirm preferences; it won’t infer from your codebase or silence. - Very compact entries (≤5 words) and a 100-line limit can make complex preferences hard to express or easy to misinterpret. - Relies on local filesystem access to `~/coding/`; non-standard environments (restricted FS, different home layout, multiple machines) may reduce effectiveness or cause desync. - Cannot auto-adapt to project-specific patterns or evolving team conventions without explicit updates. **Recommended scenarios** - You have strong, stable coding conventions (naming, testing layout, tooling decisions) and want them consistently obeyed. - You value privacy and don’t want the assistant scanning your repo or sending style data anywhere. - You’re okay with a “train by correction” workflow where you occasionally say “I prefer X over Y—please remember this.” - You want a lightweight, transparent style memory you can inspect and edit as simple markdown files.