What problem does it solve? Committing code in the mj-agent repository requires strict adherence to a commit message convention (type(scope): summary), a closed 12-scope allowlist, and a branch-type compatibility matrix. Manual commits often violate these rules, causing rework at push time, and risk accidentally staging secrets like .env files or private keys. ## Core Features & Use Cases - Six-step pre-commit workflow: Verifies working location, reviews changes, composes the commit message, enforces type/branch discipline, evaluates split necessity, and executes the commit. - Secret and file exclusion guardrails: Hard-blocks staging of .env, decrypted secrets, and private keys (*.pem, *.key, *.p12), and warns on files over 10 MB. - Scope inference and split guidance: Derives the correct scope from modified file paths using the 12-scope closed allowlist and recommends splitting commits that span unrelated scopes or mix feat with refactor. - Use Case: You finished a bugfix on branch bugfix/qcm-yaml-encoding and ask to commit. The skill checks the branch, stages only the intended files, validates that type fix is allowed on a bugfix branch, and produces a compliant message like fix(sql): handle Chinese qcm names in yaml parsing. ## Quick Start Ask the assistant to stage my current changes and create a compliant commit for this branch in mj-agent.
