commit-report

A Claude Code / Claude skill that turns your git history into a polished, self-contained HTML dashboard.

Point it at one or more repositories and a date range, and it produces:

• Per-contributor statistics — commit count and lines added/removed, per repo and total.
• Per-repo commit detail — every commit with author, message, file count and clickable links back to the commit on GitHub/GitLab.
• Code-quality review — each commit's diff is graded ✅ / 🟡 / 🟠 / 🔴. When a repo ships a CLAUDE.md or AGENTS.md, it's reviewed against those rules (🛡️ guidelines mode); otherwise it falls back to general best practices for the detected stack, using the model's knowledge and occasional web lookups (🌐 generic mode). Each report states which mode it used. Turn the review off per-repo ("qualityReview": false) or by asking for "stats only".

Everything is rendered as a single self-contained HTML file (inline CSS, vanilla JS, inline SVG charts — no CDN, no external assets) and published through the Artifact tool.

Install

Copy the commit-report/ folder into your skills directory:

# Claude Code (user-level skills)
cp -r commit-report ~/.claude/skills/

# or a project-level skill
cp -r commit-report /path/to/project/.claude/skills/

The skill is then available to Claude automatically.

Requirements

The skill is pure instructions — no install step, no runtime of its own. It only leans on tools the host already has:

Tool	When needed
git (CLI)	Optional but recommended — required to read commit history from local repos and to shallow-clone remote ones. Without it, only the (more limited) host fallbacks apply.
Network access	Only to analyze a remote repo (the shallow clone) or for the occasional mgrep --web lookup in generic quality mode.
gh / SSH key / token	Only to clone private remote repos.

Usage

Just ask, in natural language:

• commit report — yesterday, current repo
• commit report for this week
• daily commit report for alice
• how many commits last week across frontend and backend
• commit report from 2026-06-20 to 2026-06-23
• commit report for https://github.com/owner/repo this week — analyze a remote repo

Analyzing a remote repo (no local clone needed)

Pass a repo URL (https://github.com/owner/repo, git@github.com:owner/repo.git, GitLab, etc.) and the skill shallow-clones it to a temp folder, runs the full report (stats and quality review), then deletes the clone. You don't have to be inside a git folder. Private repos work only if credentials are already set up (gh auth login, an SSH key, or a token). Bounding the request to a date range keeps even huge repos fast.

Repo resolution

The skill is zero-setup. It figures out which repos to analyze in this order:

1. A repo URL in your message → shallow-cloned to a temp folder and analyzed (see above).
2. commit-report.config.json in the current directory, if present.
3. Otherwise, if you're inside a git repo, it uses the current repository and tells you so.
4. Otherwise, if the current directory contains several git repos as subfolders, it proposes the list and asks you to confirm.
5. Otherwise, it asks you for the repo paths.

Configuration (optional)

For multi-repo reports or custom behavior, drop a commit-report.config.json next to where you run the skill. See config.example.json. All fields are optional:

Field	Purpose	Default
repos[].path	Repo location	—
repos[].name	Display name	folder basename
repos[].commitUrlTemplate	Commit link, with {hash}	derived from origin remote
repos[].guidelines	Best-practices files for the quality review	auto-detect CLAUDE.md / AGENTS.md
publish	Extra destination (SCP/rsync) on top of the Artifact	Artifact only

Date ranges

yesterday (default) · a specific date · this week · last week · last N days · explicit from … to … ranges. The HTML output matches the language of your request.

Publishing elsewhere

By default the report lives only as an Artifact. Ask the skill to "also save it to /some/path" or to SCP it to a host, and it will — or configure a persistent publish block.

What the quality review checks

Two modes, picked automatically per repo:

• 🛡️ Guidelines mode — when a CLAUDE.md / AGENTS.md exists, the review is driven entirely by it and imposes no external conventions. If your doc forbids a library or pattern, or mandates a commit-message format, commits that break those rules are flagged.
• 🌐 Generic mode — with no such file, commits are reviewed against widely-accepted best practices for the detected stack (correctness, security, error handling, naming, tests, commit hygiene), using the model's knowledge plus occasional mgrep --web lookups for fast-moving ecosystems. Grades are badged as generic so they aren't mistaken for repo-endorsed rules.

Set "qualityReview": false on a repo, or ask for "stats only", to skip the review entirely.

Bots & merges

Merge commits (--no-merges) and bot authors (emails containing bot or noreply, e.g. dependabot, github-actions[bot]) are excluded from both the stats and the review.

Credits

Created by Francesco Strappini at Next.

License

MIT © Next S.r.l. unipersonale