Aspirina commit
Create a git commit following project conventions. Use when: commit, commit this, make a commit, commit changes, git commit, save changes, commit my work, stage and commit, detailed commit.
install
source · Clone the upstream repo
git clone https://github.com/leandronsp/aspirina
Claude Code · Install into ~/.claude/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/leandronsp/aspirina "$T" && mkdir -p ~/.claude/skills && cp -r "$T/.claude/skills/commit" ~/.claude/skills/leandronsp-aspirina-commit && rm -rf "$T"
manifest:
.claude/skills/commit/skill.mdsource content
Git Commit
Modes
Quick (default)
Single-line commit message.
<type>: <short description>
Detailed (/commit detailed
or /commit -d
)
/commit detailed/commit -dMulti-paragraph commit for significant changes. Review the conversation and staged diff, then write:
<type>: <short summary> <paragraph explaining what changed and why> <paragraph on technical approach, trade-offs, or notable decisions>
Use detailed mode for: milestone features, non-obvious fixes, architectural changes, anything where "why" matters more than "what".
Format
Types:
feat:, fix:, refactor:, test:, chore:, docs:
Rules
- Concise — short message, present tense ("add" not "added")
- Lowercase after prefix
- No AI mentions — never reference Claude, AI, or assistants
- No Co-Authored-By — never add Co-Authored-By trailers
- No emojis in commit messages
- Specific files —
, nevergit add <files>git add .
Pre-commit Checklist
cargo test cargo clippy cargo fmt --check git diff --staged
Examples
Quick
git commit -m "feat: add batch matrix multiplication" git commit -m "fix: handle empty matrix in transpose" git commit -m "refactor: extract activation into trait"
Detailed
git commit -m "feat: add learning rate decay to training Training now supports configurable learning rate decay, reducing the rate by a factor each epoch. This prevents overshooting minima in later training stages when weights are already close to optimal. Decay is optional and defaults to 1.0 (no decay). Each gate training scenario uses hand-tuned decay values for faster convergence."