What You Can Ask Your AI
A one-page reference of everything you can ask your Specnote-connected AI to do, with ready-to-use example phrases. Feed this whole page to your AI and it will understand you even better.
Once Specnote is connected to your AI, everything below can be done with a single sentence. There's no need to memorize tool names — describe what you want in your own words, and your AI picks the right tool.
One tip: click Copy as Markdown at the top right of this page and paste the whole thing to your AI. It will understand what Specnote can do even more precisely.
Managing workspaces
| Try saying | What happens |
|---|---|
| "Show me my Specnote workspaces" | Lists the workspaces you've made |
| "Create a workspace called 'My Shop'" | Creates a new workspace |
| "Change this workspace's one-line description" | Edits the title and description |
Creating and refining tests
New tests only get created through the approval flow in Creating tests. You approve on screen between the first two rows below. The remaining three rework tests you already have, so they need no approval again.
| Try saying | What happens |
|---|---|
| "I want to check that signing up sends a welcome email" | Drafts a plan review (PRD) of what to check and by what rules |
| "Submit tests from the approved review" | Submits only tests inside the scope you approved and runs the first verification |
| "Rebuild the steps for this test" | Browses the real screens and re-measures the steps |
| "Rename this test / reorder its steps" | Tidies up tests and steps |
| "Group similar tests into areas" | Organizes tests by area |
Starting a run group — the PRD approval flow
A new run group (round) starts only on a plan review (PRD) that a person has approved. Ask your AI like this.
| Try saying | What happens |
|---|---|
| "Start a verification round for this workspace" | Puts together a plan review (PRD) of what to verify and by what rules |
| "Approve this PRD" | Your AI can't — it refuses and hands back the approval screen's address (a person approves) |
| "Submit tests from the approved PRD" | Submits the tests per the approved review and runs the first verification automatically |
Running verifications and reading results
| Try saying | What happens |
|---|---|
| "Run this test" | Starts a real-browser verification |
| "Re-verify everything" | Re-verifies several tests in place at their existing group, refreshing pass rate |
| "What was the last verification result?" | Reads pass/fail and the round report |
| "Analyze why it failed" | Fetches the fix report (where, why, how) — you can hand it straight into a fix |
| "Suggest a fix for this failure" | Gets fix suggestions grounded in the failure cause |
Code connection and sync
| Try saying | What happens |
|---|---|
| "Send my current code to Specnote" | Your AI uploads the code directly with secrets masked (useful without GitHub, or for un-pushed local changes) |
| "Which tests does this change affect?" | Points out the tests and steps touched by the change |
Test environments, accounts & secrets
Everything you'd do on screen in Test Environments & Accounts can be said out loud instead.
| Try saying | What happens |
|---|---|
| "Register a test environment at ○○○" | Registers the site address to test |
| "Make Staging the default environment" | Switches the default environment |
| "Add the test account ○○○" | Registers the sign-in account (encrypted immediately) |
| "Verify this test as a member" | Sets whether the test signs in (member/visitor) |
Service knowledge and docs
| Try saying | What happens |
|---|---|
| "Write one line on what this service is" | Fills in the one-line description (the only service blurb Specnote stores) |
| "Summarize what this service is" | Drafts a service description grounded in the code |
What your AI can't do (human-only decisions)
Almost everything is one sentence away — but two things are set by a person, on screen, by design:
- Approving the verification plan (PRD) — confirming "what to verify and by what rules" before a new run group starts. Tell your AI to "approve it" and it refuses, handing back the approval screen's address instead.
- Approving pass criteria — confirming "what counts as success" is human-only.
The reasoning behind this is covered in How Verification Works. Not connected yet? Start with Connect Your AI (Install MCP).