What problem does it solve? Running the right set of local checks before a commit is error-prone: developers must remember which lint, type-check, test, and doc-validation commands apply to their changed files, and which commands have side effects requiring confirmation. This Skill automates that orchestration for the mj-agent repository. ## Core Features & Use Cases - Scope Detection: Maps git diff output to change domains (agent, llm, sql, docs, infra, skills) and selects the matching command set. - Two-Level Verification: Auto-runs Level A read-only checks (ruff, mypy, pytest unit/eval, compileall, wikilink and frontmatter checks) and asks for HITL confirmation before Level B side-effecting checks (integration/smoke/contract tests, langgraph dev Studio probe, docker compose lifecycle). - Safety Guardrails: Never runs Level C destructive operations such as docker compose down -v, database writes, or secret changes. - Use Case: After modifying src/mj_agent/tools/sql/guardrail.py, ask for local verification and receive a Verify Report with ruff, mypy, and unit test results plus a confirmation prompt for integration tests. ## Quick Start Ask the assistant to run local verification on my current uncommitted changes before I commit them.
