Agent skill
commit-smart
Analyze staged/unstaged changes and create semantic conventional commits with context about WHY, not just WHAT. Auto-detects commit type and scope from the diff. Supports optional type/scope arguments. Usage - /commit-smart, /commit-smart fix, /commit-smart refactor api
Install this agent skill to your Project
npx add-skill https://github.com/davila7/claude-code-templates/tree/main/cli-tool/components/skills/git/commit-smart
SKILL.md
Smart Commit
Create meaningful conventional commits by analyzing your actual changes.
Workflow
Step 1: Assess the working tree
Run these commands to understand the current state:
git status
git diff --stat
git diff --cached --stat
Step 2: Handle unstaged changes
If nothing is staged (git diff --cached is empty):
- Show the user what files have changed
- Suggest what to stage based on logical grouping (e.g., "these 3 files are all related to the auth refactor")
- Ask if they want to stage all, or select specific files
- Stage the approved files with
git add <files>
If changes are already staged, proceed to analysis.
Step 3: Analyze the diff
Read the full staged diff:
git diff --cached
Determine the commit type from the changes:
| Signal | Type |
|---|---|
| New files with new functionality | feat |
| New test files or test additions | test |
| Changes to existing logic fixing incorrect behavior | fix |
| Structural changes without behavior change | refactor |
| package.json, tsconfig, CI config changes | chore |
| Build/bundler config changes | build |
| README, docs, comments only | docs |
| Formatting, whitespace, semicolons only | style |
| Performance improvements | perf |
Determine the scope from the primary directory or module affected:
src/api/->apisrc/components/auth/->authtests/->tests- Root config files -> omit scope
- Multiple unrelated areas -> omit scope
Step 4: Check for user overrides
If the user provided arguments via $ARGUMENTS:
- Single word (e.g.,
fix) -> use as commit type - Two words (e.g.,
refactor api) -> use as type and scope - Otherwise -> use auto-detected values
Step 5: Compose the commit message
Format: type(scope): imperative short description
Rules:
- Subject line max 72 characters
- Use imperative mood ("add", "fix", "refactor", not "added", "fixes")
- Don't end with a period
- Body explains WHY this change was made, not what changed (the diff shows what)
- If changes are trivial (typo fix, formatting), skip the body
Example:
feat(auth): add JWT refresh token rotation
Tokens were expiring mid-session for users with slow connections.
Rotating refresh tokens extends the session without compromising
security, since each refresh token can only be used once.
Step 6: Confirm and commit
Show the user the proposed commit message and ask for confirmation.
If confirmed, run:
git commit -m "<message>"
Then verify with:
git log --oneline -1
Show the committed hash and message.
Tips
- Run after completing a logical unit of work, not after every file change
- If the diff is too large for one commit, suggest splitting into multiple commits
- For breaking changes, add
!after the scope:feat(api)!: change response format - The body should answer "if someone reads this commit in 6 months, will they understand WHY?"
Recommended Agent Skills
Expand your agent's capabilities with these related and highly-rated skills.
verl-rl-training
Provides guidance for training LLMs with reinforcement learning using verl (Volcano Engine RL). Use when implementing RLHF, GRPO, PPO, or other RL algorithms for LLM post-training at scale with flexible infrastructure backends.
openrlhf-training
High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3. 2× faster than DeepSpeedChat with distributed architecture and GPU resource sharing.
gguf-quantization
GGUF format and llama.cpp quantization for efficient CPU/GPU inference. Use when deploying models on consumer hardware, Apple Silicon, or when needing flexible quantization from 2-8 bit without GPU requirements.
Claude Code Guide
Master guide for using Claude Code effectively. Includes configuration templates, prompting strategies "Thinking" keywords, debugging techniques, and best practices for interacting with the agent.
qdrant-vector-search
High-performance vector similarity search engine for RAG and semantic search. Use when building production RAG systems requiring fast nearest neighbor search, hybrid search with filtering, or scalable vector storage with Rust-powered performance.
behavioral-modes
AI operational modes (brainstorm, implement, debug, review, teach, ship, orchestrate). Use to adapt behavior based on task type.
Didn't find tool you were looking for?