Skip to main content

Documentation Index

Fetch the complete documentation index at: https://ito.ai/docs/llms.txt

Use this file to discover all available pages before exploring further.

Test Instructions let you give Ito direct guidance about how to test your codebase. Think of them as a plain-English briefing that gets passed straight to Ito before it runs. The more specific and explicit you are, the better Ito will incorporate your instructions into the test plans. A good mental model: it’s a little like briefing a highly capable, eager-to-please engineer who knows your code but doesn’t know your business. They will follow your instructions closely, so be clear about what matters to you, what doesn’t, and any context about your product that isn’t obvious just from reading the code.

Do’s

  • Use the word “requirement” for anything non-negotiable. The AI treats this as an absolute. If you want something to always be tested or always be respected, call it a requirement explicitly. Example: “It is an absolute requirement that checkout works end-to-end if anything in the payments folder is touched.”
  • Be specific about folders, dimensions, and scenarios. Vague instructions produce vague results. Name the folders, pages, or flows you care about by name. If it’s relevant, explicitly list the directory to help Ito find exactly what you’re referencing.
  • Tell it what you don’t care about. Saying “NEVER test mobile dimensions” or “minor cosmetic bugs on admin pages are low priority” helps Ito spend test budget on what actually matters to you.
  • Give Ito context about your data and scale. Ito works from your code, not your production database. If it’s not obvious from the code that your platform needs to handle thousands (or millions) of users, tell Ito that explicitly.
  • List out all cases when something is situational. If a rule only applies in certain scenarios, list those scenarios. The more you enumerate, the less Ito has to guess.
  • Iterate on your instructions. After your first few runs with new instructions, review the results and adjust. Watch your next few tests to see if they reflect what you intended.

Don’ts

  • Don’t be vague or ambiguous. Hedging language like “maybe check the UI” gives the AI too much room to interpret. Say what you mean.
  • Don’t assume Ito knows your scale or data state. Ito doesn’t know what your production environment or customers are like. If your platform handles large organizations or high-traffic scenarios, Ito won’t know that unless you say so.
  • Don’t use absolute language unless you mean it. If you write “requirement” or “never,” Ito will treat those as firm rules. Reserve that language for things you truly mean every single time.
  • Don’t forget these settings apply to everyone at your org by default. You can control both the repositories and which PR authors each instruction applies to, but only if you set them specifically. Otherwise, any changes you make here are global and affect tests across all PRs in your organization.
  • Don’t forget to test the impact of what you added. When you make a change, watch your next couple of PR runs to confirm it’s doing what you expected. If not, come back and try changing the language to be more specific or detailed.
  • Don’t try to do everything at once. Start with one or two of the most important things you want to protect or customize. Add more as you learn what works, being mindful not to create conflicting test instructions.

Inspiration and examples

Not sure where to start? Here are a handful of examples to help you get going. These aren’t templates, but real use cases to consider depending on your product. Use your own wording that best fits your situation, or write your own from scratch. Protect a critical flow
"It is an absolute requirement that if anything touches the /checkout or /payments folder, the end-to-end checkout flow is verified to still work."
Set mobile testing expectations
"Always test responsive layouts at 375px (mobile) and 768px (tablet) in addition to standard desktop."
Or, if you don’t care:
"NEVER test mobile dimensions. Only test standard desktop (1280px)."
Give scale context
"Our platform supports organizations with up to 5,000 members. Any test involving org-level features like team pages, member lists, or permission flows should assume a large-org scale."
Set priority guidance
"User-facing bugs are 50x more important than admin-only bugs. Prioritize anything a regular end user would encounter over testing internal tooling."
Exclude low-priority areas
"Do not flag minor cosmetic or layout issues on admin-only pages. Focus on functional and access-control bugs in those areas instead."
Flag non-obvious performance concerns
"Our repository can have more than 10,000 open PRs. Any feature that queries or displays PR lists should be tested with that scale in mind."
Protect a key regression area
"If any change touches the billing or subscription code, verify that existing paid users cannot be accidentally downgraded or lose access."
Apply more rigorous testing for junior contributors
"For PRs authored by @junior-dev or @intern-username, apply more thorough testing: flag any change touching core product flows, API endpoints, or shared utilities, and require end-to-end verification even for small or seemingly low-risk changes. Treat these PRs with higher scrutiny than PRs from senior engineers."

This feature is experimental

Test Instructions are a new and evolving capability. Don’t expect to get your instructions perfect on the first try. Start small, watch the next few test runs, and adjust. We’ll be refining our own guidance as we see what patterns work best in the real world.
If you write an instruction and it produces exactly what you hoped for, tell us. If something didn’t work the way you expected, or if you’re not sure how to phrase something, reach out. We want to know what’s landing and what’s confusing, and can help you craft the right instructions. Email us at support@ito.ai to share feedback, suggestions, or ask for help.