Pre-launch check

Finds what will break on real respondents while the link is still unsent

Free agent skill for Claude Code, Codex, Cursor and Gemini CLI. It works through the SurveyNinja MCP server, with your permissions.

Category
Check before launch
Tags
launch, logic and branching, notifications
Works with
Claude Code, Codex, Cursor, Gemini CLI
MCP tools
8, all read only
Version
1.0.1, October 20, 2026
Made by
SurveyNinja
Price
Free
Skill file
SKILL.md

What it does

  • Mistakes found before sending. After the link goes out, part of the responses are already collected through a broken survey and cannot be compared with the rest.
  • Every route checked to the end. Dead ends, unreachable questions and conditions on deleted questions are found before live respondents run into them.
  • You decide what to fix. The skill hands over a list sorted by severity and changes nothing on its own.

Example

You type

Check my survey before launch. Is everything ready?

What the assistant brings back

2
Breaks collectionfix before sending
3
Hurts data
4
Minor
3
Cleanchecked, nothing to report
  1. The "never used the service" branch leads nowhere logic, question 6
  2. No email recipients and no integrations where responses land
  3. Recipient address is not confirmed email, marketing@…
  4. Phone is required at question three of nine questions
  5. The finish screen has no text texts, finish
  6. Button labels are still the defaults texts
Not ready to publish yet

Two findings will lose responses outright: the dead end in the logic and leads that nobody receives. The rest can wait until after launch.

The findings above are an example. In your account the check runs against your own survey.

How it works

  1. Reads the whole survey firstQuestions, logic, texts and publication state. Before judging anything, the assistant looks at what is actually in the survey.
  2. Walks every route to the endDead ends, unreachable questions, conditions on deleted questions, a "greater than" on a text field. These are the things that surface on live respondents.
  3. Checks where the responses will landEmail recipients and whether their addresses are confirmed, Telegram, Zapier, Google Sheets and webhooks. An unconfirmed address looks configured but never receives mail.
  4. Hands over a list and waitsNothing is fixed quietly. Findings come sorted by severity, and you decide what to fix.

Try also

  • I am about to send the link. Did I forget anything?
  • Find logic and text problems before I publish

What it will not do

  • Do not fix silently
  • Do not publish on your own initiative
  • Do not rewrite question wording
  • Do not suggest a plan upgrade or lead to payment

The skill works with your permissions. Tools that change data are marked, so you check what the assistant is about to do before you confirm. To cut access, remove the connector in the assistant or delete the API key.

In short

What does the skill check?

The branching logic, the questions, the texts and the finish screen, where the responses will go, the limits left on your plan, and whether the survey is published and reachable. For a survey on your website it checks that the survey is attached to a site.

Which logic errors does it look for?

Dead ends with no way forward, questions that no route leads to, conditions on deleted questions, and operators that do not fit the question, such as "greater than" on a text field.

Will it check that I actually receive the responses?

Yes. It looks at email notifications and whether the recipient addresses are confirmed, at Telegram, Zapier and Google Sheets, and at webhooks, including one that was switched off after failed deliveries.

Will the skill fix what it finds?

No. It lists the findings and you decide what to repair. It will not publish the survey on its own or rewrite your questions.

What do I get at the end?

A list sorted by weight: what will break collection, what will hurt data quality, and what is just worth improving. If everything is clean, it says so without inventing remarks.

What do I need to use it?

A SurveyNinja account and an assistant that connects to MCP and loads skills from a folder, such as Claude Code, Codex, Cursor or Gemini CLI. Access goes through your own API key or "Allow access" in the assistant. The skill is plain text that runs nothing itself: your assistant does the work through the MCP server, with your permissions.

The full skill

The SKILL.md file the assistant reads, exactly as it is in the archive.

Show the full text

Checking a survey before launch

Survey mistakes surface after the link goes out, when part of the responses are already collected through a broken questionnaire and cannot be compared with anything.

Reply to the person in the language they write in.

Order

Read get_quiz_structure (questions, their logic, and the headings and texts of the welcome and thank-you screens), get_quiz_texts (interface and button labels) and get_quiz_list (the survey type, whether it is published and at what address), then walk the list below. Fix nothing silently: show the findings first, then ask what to repair.

What to check

Branching

  • Dead ends: a branch with no way forward and no way to finish.
  • Unreachable questions: nothing routes to them.
  • Conditions pointing at deleted or renamed questions.
  • An operator that does not fit the question type, such as "greater than" on a text field.

Questions

  • A choice question with no options.
  • Required questions that do not need to be required: each one costs responses.
  • Open text where a count is needed: a rating typed as text never reaches the charts in the report.
  • Excessive length: if there are more than fifteen questions, say so plainly.

Texts

  • Empty headings, default button labels, an empty text on a thank-you screen that exists.
  • No finish screen is not a defect: the respondent then sees the standard completion screen, and the Thank You screen depends on the plan. Mention it only if the survey needs its own final message, such as a promised next step.

Where the answers will go

  • get_quiz_email_settings: are notifications on, are recipients set, and are their addresses confirmed (the checked field). An unconfirmed address looks configured but receives nothing.
  • get_quiz_integrations: is Telegram, Zapier or Google Sheets connected, is sending on, and does the plan allow it.
  • get_quiz_webhooks: if there are webhooks, are they on, and was one switched off automatically after failed deliveries.
  • A survey with no notifications and no integrations collects responses in silence. That can be fine if the person reads the report, but say it.

Plan and access

  • get_workspace_tariff: how many responses, surveys and storage remain, and whether the features the survey relies on are included.
  • A regular survey: is it published and reachable by its link.
  • An on-site survey (type micro) has no link of its own: get_quiz_summary shows is_connected. If it is false, the survey is not on any site yet and nobody will see it.

What not to do

  • Do not fix silently. List the findings and let the person decide.
  • Do not publish on your own initiative.
  • Do not rewrite question wording. That is separate work.
  • Do not suggest a plan upgrade or lead to payment. If a limit is hit or a feature is not on the plan, state the fact and stop.

What to say at the end

Findings ordered by weight: what will break collection, what will degrade data quality, what is merely worth improving. If everything is clean, say so and do not invent remarks.

The SurveyNinja ninja on a rocket

Connect SurveyNinja to your assistant

The MCP server works on every plan, Free included. Sign up, allow access and ask your assistant to build the first survey.