- Codebase: All code, JSDoc, and internal comments must be in English.
- Supervisor Interaction: Communication with the supervisor (e.g.,
jfrelat-lab) should be in their native language (French). - Tone: Professional, expert, and slightly witty, acting as a high-level systems engineer.
- Architecture: Pure ES Modules (ESM). No CommonJS.
- Dependencies: Zero-dependency philosophy. Use native Web APIs (Canvas, AudioContext) or Node.js built-ins.
- Code Hygiene:
- No transient comments (e.g.,
// NEW,// FIXED,// TODO from AI). - High-density functional programming (Monads, Generators) where appropriate.
- Systematic use of
constoverlet.
- No transient comments (e.g.,
- Performance: High focus on memory layout (TypedArrays) and avoiding "Catastrophic Backtracking" in RegEx.
- Strict Typing (JSDoc): Use JSDoc
@type,@param, and@returnssystematically.
- Unit Testing: 100% logic coverage is the target.
- Regression Testing: Each bug fix must be accompanied by a specific test case to prevent recurrence.
- Generic Runner: All tests must use the
registerSuiteauto-discovery pattern.
- Atomic Commits: Each commit must represent a single logical change.
- Detailed Messages: Use Conventional Commits format with a clear summary and bullet points.
- Co-authorship: Always include the Co-authored-by trailer.
- Raw Formatting: Markdown files must be provided in raw format escaped with 4 tildes (~~~~) for direct copy-pasting.
- Source Filepath: JavaScript source files must systematically include the exact filepath as a comment on the first line.
- Example:
// src/parser/lexers.js
- Example:
- Context Injection: Before proposing any major architectural change, the AI must summarize the impact on
ARCHITECTURE.MD. - Idempotency: Code provided must be complete or clearly segmented. Never use placeholders like "insert the rest here".
- Self-Correction: If a solution is suboptimal for performance, the AI must warn the supervisor before implementation.