micro-commit

Overview

Splits uncommitted changes into small, logical commits — one per feature, fix, or layer. Each commit should be independently meaningful and reviewable.

When to Use

• Working tree has changes across multiple files with different purposes
• About to open a PR and want a clean, readable history
• Mixed changes: feature + refactor + config in one dirty working tree

Not needed when: all changes belong to a single logical unit — just commit normally.

Instructions

When delegation is available and permitted by the host environment, delegate all Git operations (status checks, staging, committing) to one sub-agent using the host's supported delegation tool. If delegation is unavailable or prohibited, execute the same workflow directly in the main session.

Model Selection

When the host agent supports model selection for the sub-agent, prefer a low-cost, fast model from those available in that environment, provided it can reliably review diffs, group changes, and execute Git operations. Use the host's documented model identifiers and selection mechanism; do not assume a particular provider or invent model names or parameters. Respect explicit user and project model settings. If model selection is unavailable or no suitable alternative is known, omit the model override and use the host's default or inherited model. For direct execution, keep the current model.