Git Workflow

Git Workflow

Commits — Conventional Format

<type>(<scope>): <description>

[optional body]
[optional footer]

Rules: imperative mood ("add" not "added"), scope = feature area, <72 chars, body explains WHY.

Branch Naming

Format: <type>/<ticket-or-desc> (kebab-case)

feat/task-crud-api
fix/auth-token-expiry
refactor/storage-layer
chore/update-deps

Commit Hygiene

• One logical change per commit
• Never commit broken tests
• No debug code (console.log, TODO hacks)
• No secrets (use .gitignore + env vars)

PR Size

• Ideal: <400 lines. Acceptable: 400-800. Too large: >800 (split).

Merge Strategy

• Feature → main: squash merge
• Release: merge commit
• Hotfix: cherry-pick

Monorepo Strategies

When working in monorepos with multiple packages/services:

• Scope commits to package: feat(api): add task endpoint — scope = package name
• Changelogs per package: each package has its own CHANGELOG.md
• Selective CI: trigger builds only for affected packages (path filters)
• Shared deps: pin at workspace root, override per-package only when needed
• Atomic changes: cross-package changes in single commit when they must deploy together

Semantic Release Automation

For automated versioning and changelog generation:

1. Conventional commits → version bumps: feat → minor, fix → patch, BREAKING CHANGE → major
2. Tooling: semantic-release (Node.js), release-please (GitHub), go-semantic-release (Go)
3. CI integration: run after merge to main, generates tag + changelog + release
4. Pre-release channels: alpha, beta, rc for pre-release testing

Worktree Best Practices

When orchestrator uses git worktrees for parallel agent execution:

Single-Instance Dispatch

• Directory: .wt/<agent-name> — consistent prefix for cleanup
• Branch: wt/<agent-name>-<timestamp> — unique per dispatch

Multi-Instance (Scoped) Dispatch

• Directory: .wt/<agent-name>-<scope> — scope qualifier from scope card
• Branch: wt/<agent-name>-<scope>-<timestamp> — unique per scoped dispatch
• Examples: .wt/backend-engineer-auth, .wt/backend-engineer-tasks, .wt/scout-infra

Merge Ordering

• Dependency order first: merge nodes that downstream nodes depend on before their dependents
• Smallest diff first: for independent nodes, merge smaller changes first (less conflict surface)
• Integration sub-tasks last: @<agent>[integration] branches always merge after all feature branches

General Rules

• Cleanup: always remove worktree + branch after merge
• Conflict resolution: merge sequentially, resolve conflicts before next merge
• Never commit directly to main from worktree — always squash merge
• Quality gate between merges: run Code Completion Mandate checks after each merge
• Full merge protocol: see parallel-dispatch-merge skill

Checklist

• Branch named with type prefix
• Conventional commit format
• No debug code or secrets
• Tests pass before commit
• PR <400 lines (or justified)
• Messages explain WHY
• Lock files committed with dependency changes
• CHANGELOG updated for user-facing changes

Related

• Code Completion Mandate GEMINI.md § Code Completion Mandate
• Testing Strategy GEMINI.md § Testing Strategy
• Security Mandate GEMINI.md § Security Mandate
• Dependency Management Principles @.gemini/skills/dependency-management-principles/SKILL.md