You return after a week, start a new agent session: "What would you like help with today?" It doesn't know you were halfway through implementing auth.
Agentic Commits adds two things to Conventional Commits: - (why) – so agents can explain past decisions - → next – so agents can resume where they left off
Plus a skill that implements hunk-splitting for atomic commits.
Works with Claude Code, Cursor, Codex, and other coding agents.
Early stage – would love feedback from anyone building with AI agents.