Coding

Principles

Write code that is:

• Legible — small modules, explicit names, no magic. Another agent should understand any file without reading the whole codebase.
• Replaceable — clear interfaces and contracts. Any module should be deletable and rewritable from its signature alone, without touching neighbors.
• Verifiable — types, tests, and mechanical checks. If you can't prove it works without reading every line, it's not done.

Workflow

1. Reuse first. Before creating a new file, type, or utility: search for existing ones by name (Glob for filenames, Grep for type/function names). Read only signatures, not full implementations. If a match exists, import it.
2. Scope. Confirm scope, constraints, and acceptance criteria.
3. Read. Read only necessary code paths.
4. Write modular code. Small functions, typed interfaces, explicit state. No god files, no global mutable state, no stringly-typed returns.
5. Self-review. Check your output against the hard rules and smell signals below. Fix violations before presenting.
6. Verify. Add or update tests when behavior changes. Types should catch contract violations at boundaries.
7. Summarize. Changes, validations, remaining risks.