Skip to content

split

regit split [<commit-ish>] [flags]

Split one commit into 2–4 semantic commits using an LLM. The model groups the commit's hunks into coherent pieces and proposes a message for each; regit replays them as separate commits while preserving the combined content byte for byte. By default the proposal is only previewed — pass --execute to apply it.

Positional arguments

Argument Default Description
<commit-ish> HEAD The commit to split — hash, HEAD~n, or branch.

Flags

Flag Type Default Description
--llm VALUE — provider/model, e.g. ollama/qwen2.5-coder.
--agent VALUE — Agent CLI to drive instead of HTTP.
--agent-args VALUE — Replace default agent args; {schema} is injected.
--llm-base-url VALUE — Override the OpenAI-compatible endpoint.
--max-commits N 0 Clamp on resulting commits (0 = config/auto).
--prompt VALUE — Short guidance passed to the LLM.
--test-cmd VALUE — Validate each intermediate commit before applying.
--no-cache BOOL false Ignore any cached proposal.
--clear-cache BOOL false Delete the split cache and exit.
--repo VALUE . Path to the git repository.
--execute BOOL false Apply the split (default: dry-run preview).
--force BOOL false Allow rewriting already-pushed commits.
--sign BOOL false Re-sign rewritten commits with your configured signing key.
Examples

Correct:

# preview a split of HEAD (dry-run)
regit split HEAD --llm=ollama/qwen2.5-coder

# apply the split
regit split HEAD --llm=ollama/qwen2.5-coder --execute

Incorrect:

# BAD: --yes is a deprecated alias for --execute — use --execute instead
regit split HEAD --yes

Limitations

  • Requires an LLM (--llm or --agent).
  • Skips commits that are too small, single-hunk, pure-rename, or binary-only.