m_code_init_projectlisted

# m_code_init_project Initialize a repository so Claude and humans can repeatedly build, test, review, and change it safely. ## Contract Use `$ARGUMENTS` as the project goal, stack, repo path, and constraints. Default behavior: - Empty repo: create the smallest useful baseline. - Existing repo: harden current conventions; do not overwrite them silently. - Ambiguous stack: infer from files; if still unclear, state the assumption before editing. - Do not create or commit secrets, tokens, private keys, production URLs, or real customer data. - Do not add broad tool permissions or auto-approval rules. ## Workflow ### 1. Inspect before writing For an existing repo of any size, delegate the wide scan to the `m_code-context-scout` agent and keep only its map — that keeps the raw inspection out of your main context. Read only what is needed: ```txt - git status - top-level tree - package/build files - README/docs - test/lint/type/build config - existing CLAUDE.md, AGENTS.md, .github/copilot-instructions.md, .claude/rules - CI config if present ``` Report: ```txt Project kind: Language/stack: Package manager: Existing commands: Existing tests: Current architecture shape: Main gaps: ``` ### 2. Create deterministic commands Prefer ecosystem-native scripts in the existing package/build config. Define or document: ```txt install run/dev test test:unit, when useful test:integration, when useful typecheck, for typed stacks lint format or format:check build ``` If a command c