sprint-review reads what your team shipped and writes the deck the business understands. Every line traced to a real commit, real screenshots of what changed, ready before the meeting.
How it works
A model alone gets the words right and the facts wrong. So the words come from the model, and everything that has to be true is done in code.
Commits, changed files and pull request descriptions from your git history, by date or by tag.
Themes in the language of your audience: what changed for them, not which files moved.
Each point must name real commits. Plumbing moves under the hood. A second reader flags what's unclear, and it gets rewritten.
One self-contained HTML deck with speaker notes and a presenter view. Also PowerPoint and Markdown.
Features
A line that can't name the commits behind it doesn't make the deck. Every correction is logged in checks.md.
Screenshots diffed pixel by pixel: from your visual tests, a folder, or taken by the tool from your app's URLs.
A separate call reads only the slides, like your audience will. Lines it stumbles on are rewritten once and checked again.
A GitHub Action and a GitLab template: the deck is ready on review day, posted to Slack if you like.
sprint-review edit opens the content beside the live deck. Change a word, see the slide, save.
Colors, fonts and logo from your own PowerPoint. Runs from Claude Code, Cursor and other MCP clients.
Get started
Run it in any git repository with an API key set. Add a config later to name your product, set the audience and point it at your screenshots.
# a preview of what it reads, no AI call npx @guuskeunen/sprint-review --dry-run # write the deck for the last sprint export ANTHROPIC_API_KEY=... npx @guuskeunen/sprint-review # change anything, with the deck beside it npx @guuskeunen/sprint-review edit sprint-reviews/2026-11-16/review.json
on: schedule: [{ cron: '0 7 * * 5' }] jobs: deck: runs-on: ubuntu-latest steps: - uses: actions/checkout@v4 with: { fetch-depth: 0 } - uses: guuskeunen/sprint-review@v0 with: anthropic-api-key: ${{ secrets.ANTHROPIC_API_KEY }}
# install the plugin: the skill and the MCP server /plugin marketplace add guuskeunen/sprint-review /plugin install sprint-review@sprint-review # then just ask > make our sprint review for the last two weeks
“I finally understand what the developers did.”Project manager, on the first generated review
Questions
You can, once. sprint-review does it every sprint with no one in the loop: it reads the exact range, refuses lines it can't trace to commits, diffs real screenshots, and keeps a log of every correction. The result is something you can present without checking it line by line.
Commit subjects, bodies, changed file names and, if you allow it, pull request descriptions. Never your code. It runs on your machine or in your CI with your own key: there's no sprint-review server. Run --dry-run to see exactly what would be sent, and read the security and privacy page.
Two or three model calls per sprint, typically a few cents to a few dimes depending on the size of the sprint and the model.
Claude by default, with a fallback model on errors. Also Amazon Bedrock, Azure OpenAI, OpenAI, OpenRouter, Gemini, Mistral, and local models with Ollama or LM Studio, where nothing leaves your network.
Set a color and logo, or point it at an existing PowerPoint: it takes the accent color, fonts and logo from the theme. There's a light and a dark deck.
Yes. Screenshots make the deck better, but themes, speaker notes, milestones and usage numbers all work without them.