Practical browser-native agent workflows that are easy to demo, easy to adapt, and credible enough to adopt.
This repository is a content-first catalog of repeatable browser-agent tasks. Each recipe is explicit about inputs, outputs, risks, operator handoff, and business value so teams can evaluate real use, not just watch a flashy demo.
- Includes concrete recipes for release notes, docs audits, pricing watch, signup checks, support reproduction, invoice collection, marketplace sync, and customer proof harvest.
- Every recipe follows the same operating model: goal, inputs, outputs, workflow, demo, risks, and value.
- Comes with a categorized Recipe Index so readers can jump to the right use case quickly.
- Need the fastest overview? Open the Recipe Index.
- Want a safe first example? Start with Documentation Audit.
- Want an obviously operational use case? Try Vendor Invoice Collection or Signup Flow Smoke Test.
If this kind of browser-agent playbook is useful to your team, star the repo and adapt the closest recipe before inventing a workflow from scratch.
- It depends on the browser, not hidden APIs.
- It has a clear operator handoff and a clear deliverable.
- It can be shown in a short demo without too much setup.
- It is useful enough that a team could adopt it after the demo.
Need a quicker path through the catalog? Start with the categorized Recipe Index.
| Recipe | Best for | Main deliverable |
|---|---|---|
| GitHub Release Assistant | shipping release notes from merged PRs | draft release title and notes |
| Documentation Audit | catching docs drift against product reality | mismatch report with fixes |
| SaaS Pricing Watch | tracking competitor pricing and packaging changes | structured change brief |
| Signup Flow Smoke Test | checking signup or checkout regressions | pass/fail run log with evidence |
| Support Ticket Reproduction | turning vague customer reports into actionable bugs | reproduction notes and evidence |
| Vendor Invoice Collection | downloading invoices from vendor portals | organized invoice bundle and exceptions list |
| Marketplace Listing Sync Check | keeping website and marketplace listings aligned | sync gap report |
| Customer Proof Harvest | collecting proof points from public web sources | citation-ready proof set |
Goal: the task the browser agent is trying to completeWhen to use: the operating context and triggerInputs: what the operator must supplyOutputs: what the run should produceWorkflow: a practical run outlineDemonstration: how to show the recipe in a short demoRisks and guardrails: where the workflow can go wrongWhy it is valuable: why a team would keep using it
- Pick a recipe close to your workflow.
- Replace the sample inputs with your own URLs, accounts, and constraints.
- Keep the listed guardrails, especially around login state, side effects, and evidence capture.
- Treat the output section as the minimum bar for a successful run.
Contributions should expand the catalog, improve clarity, or make the recipes easier to execute in real teams. See CONTRIBUTING.md, the authoring template in recipes/TEMPLATE.md, and the issue and PR templates in GitHub's standard contribution flow.
- founders packaging AI workflow demos
- devtools teams exploring browser-native automation
- operations teams dealing with portal-heavy work
- product and growth teams that need evidence, not just clicks