Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Lightweight Cursor onboarding flow that learns basic preferences, picks a first goal, and routes the user to the right next action. Use when the user invokes /onboard or wants to set up Cursor for the first time.
.claude/skills/kunanonj-cursor-onboard/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 26% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -5% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 94% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 275% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 2% | 0% |
Use this skill only when the user explicitly invokes /onboard.
Goal: run a lightweight onboarding interview and produce a handoff. Do not execute setup work.
/onboard again.During onboarding, only these tools are allowed:
AskQuestion: use for fixed-choice questions.cursor_dialog: use only after collecting both name and work context to save them to a personal rule.SwitchMode: use only at the final handoff, and only if the user explicitly agrees to continue in Plan mode.Do not use any other tools.
Say briefly that this is a quick onboarding flow toward one concrete next step and that the user can interrupt with normal Cursor questions anytime. Then ask:
"What should I call you?"
Do not ask anything else in this first message.
After the user answers with their name, ask:
"What kind of work do you do, and what does a normal project look like for you?"
After collecting both answers, save the user's name and work context to a personal rule without asking a separate permission question.
Rule title: User onboarding preferences
Rule content: The user's preferred name is <name>. Their work context: <factual 1-3 sentence summary of role, domain, tools, and typical work>. Do not infer sensitive personal details.
Briefly say that the user can ask Cursor usage questions anytime: settings, agents, rules, MCP servers, plugins, PR review, Bugbot, background agents, automations, and prompt structure.
Then ask "What would you like to do with Cursor first?" with these options:
Do not dump a feature list or write a recommendation report. Each route should feel like a guided product flow:
Routes: Setup, New project, Automate my job, Existing project, Something else (custom goal). Each follows the same pattern: one diagnostic question, short reaction, then a fixed-choice next step.
Produce a Handoff when the user chooses an explicit handoff action or when the current route has enough concrete context:
Recommended next step: the smallest useful next action.Suggested prompt: exact text the user can send next.Mode/tool to use: Plan mode, Agent mode, Automations, or normal chat.If Mode/tool to use is Plan mode, ask whether to switch now and call SwitchMode with target_mode_id: "plan" if they agree.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 6,023 | 1,427 | -76% | 1 | 1 | 0% | 898 | 1,128 | +26% | 0 | 0 | — |
case-17 | fail→fail | 9,618 | 8,682 | -10% | 1 | 1 | 0% | 1,948 | 1,567 | -20% | 0 | 0 | — |
case-02 | fail→pass | 7,774 | 1,348 | -83% | 1 | 1 | 0% | 1,166 | 1,113 | -5% | 0 | 0 | — |
case-03 | fail→pass | 3,155 | 1,372 | -57% | 1 | 1 | 0% | 559 | 1,083 | +94% | 0 | 0 | — |
case-04 | pass→pass | 9,989 | 4,676 | -53% | 1 | 1 | 0% | 1,691 | 1,753 | +4% | 0 | 0 | — |
case-05 | pass→pass | 14,529 | 7,337 | -50% | 1 | 1 | 0% | 2,448 | 2,240 | -8% | 0 | 0 | — |
case-06 | pass→pass | 11,904 | 7,727 | -35% | 1 | 1 | 0% | 1,914 | 2,181 | +14% | 0 | 0 | — |
case-07 | fail→pass | 2,202 | 4,118 | +87% | 1 | 1 | 0% | 411 | 1,540 | +275% | 0 | 0 | — |
case-08 | fail→fail | 4,929 | 7,045 | +43% | 1 | 1 | 0% | 900 | 1,470 | +63% | 0 | 0 | — |
case-09 | fail→pass | 6,247 | 2,175 | -65% | 1 | 1 | 0% | 1,218 | 1,237 | +2% | 0 | 0 | — |
case-10 | fail→pass | 7,525 | 4,240 | -44% | 1 | 1 | 0% | 1,300 | 1,747 | +34% | 0 | 0 | — |
case-16 | fail→pass | 6,378 | 5,148 | -19% | 1 | 1 | 0% | 1,245 | 1,805 | +45% | 0 | 0 | — |
case-11 | pass→pass | 1,677 | 1,075 | -36% | 1 | 1 | 0% | 271 | 1,064 | +293% | 0 | 0 | — |
case-12 | fail→fail | 5,418 | 9,594 | +77% | 1 | 1 | 0% | 930 | 1,955 | +110% | 0 | 0 | — |
case-13 | fail→fail | 7,048 | 2,432 | -65% | 1 | 1 | 0% | 1,577 | 1,392 | -12% | 0 | 0 | — |
case-14 | fail→fail | 5,590 | 9,700 | +74% | 1 | 1 | 0% | 887 | 2,154 | +143% | 0 | 0 | — |
case-15 | pass→pass | 5,690 | 2,705 | -52% | 1 | 1 | 0% | 1,009 | 1,352 | +34% | 0 | 0 | — |
case-18 | fail→pass | 11,288 | 4,053 | -64% | 1 | 1 | 0% | 2,071 | 1,630 | -21% | 0 | 0 | — |
case-19 | fail→fail | 3,803 | 1,898 | -50% | 1 | 1 | 0% | 601 | 1,193 | +99% | 0 | 0 | — |
case-20 | fail→pass | 5,499 | 2,517 | -54% | 1 | 1 | 0% | 974 | 1,354 | +39% | 0 | 0 | — |
case-21 | fail→pass | 12,773 | 3,735 | -71% | 1 | 1 | 0% | 1,237 | 1,625 | +31% | 0 | 0 | — |
case-22 | fail→fail | 5,308 | 12,992 | +145% | 1 | 1 | 0% | 839 | 1,999 | +138% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 19 counted toward the lift figure. The other 3 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +45 percentage points is the difference between those two pass rates over the 19 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.