What problem does it solve? Manually writing consistent commit messages and safely pushing verified code changes is error-prone: wrong branches get pushed, unverified code slips in without receipts, credentials leak into commits, and failed CI goes unnoticed. This Skill automates the collect-diff → generate-message → commit → push pipeline with six hard guardrails. ## Core Features & Use Cases - Guarded commit and push: Enforces six pre-commit gates including branch guard (never main/master), Chinese commit message prefix validation, baseline declaration registration, untracked-file whitelist, credential scanning, and receipt-to-commit-scope binding. - AI-generated Chinese commit messages: Collects the working tree diff (with degradation sampling for large diffs) and generates a message following the <中文type>(<scope>): <subject> format with six allowed types. - CI gate before push: Checks the GitHub Actions run conclusion of the previously pushed commit and blocks on failure, with an explicit GWP_SKIP_CI_CHECK=1 escape hatch. - Use Case: After cross-device edits are verified on a Raspberry Pi build, run this Skill to commit the code plus verification receipts in one batch and push to origin dev, with automatic remote SHA verification. ## Quick Start Ask the AI to commit and push the current dev branch changes using git-works-push after verification receipts are in place.