Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Create scheduled tasks for recurring or one-time automated execution. Use when users want to set up tasks that run automatically at specified times (daily, weekly, monthly, cron, or one-time).
.claude/skills/kxiandaoyan-scheduled-task/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | 775% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 178% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 25% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 182% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 6% | 0% |
Use this skill when users want to:
Confirm the following with the user (if not provided):
One-time execution (at):
json{ "type": "at", "datetime": "2026-03-15T09:00:00" }
Cron expression (cron) — 5-field format: minute hour day month weekday
json{ "type": "cron", "expression": "0 9 * * *" }
Common cron examples: | Expression | Meaning | |--------|------| | 0 9 * * * | Every day at 9:00 AM | | 0 8 * * 1 | Every Monday at 8:00 AM | | 0 9 * * 1-5 | Weekdays at 9:00 AM | | 0 0 1 * * | First day of month at midnight | | */30 * * * * | Every 30 minutes | | 0 * * * * | Every hour on the hour | | 0 9,18 * * * | Every day at 9:00 AM and 6:00 PM |
Use the backend API to create scheduled tasks. The API endpoint should support the following payload structure:
json{ "name": "Task name", "schedule": { "type": "cron", "expression": "0 9 * * *" }, "prompt": "Detailed instructions Claude will execute when task runs...", "workingDirectory": "/path/to/project", "description": "Optional detailed description", "systemPrompt": "Optional custom system prompt", "executionMode": "auto", "expiresAt": "2026-12-31", "notifyPlatforms": ["dingtalk", "feishu", "telegram", "discord"], "enabled": true }
| Field | Required | Description | |------|------|------| | name | ✅ | Short task name | | prompt | ✅ | Instructions Claude receives when task runs (should be clear and complete) | | schedule | ✅ | Schedule configuration (see types above) | | workingDirectory | ❌ | Execution directory (defaults to empty) | | description | ❌ | Detailed description (defaults to empty) | | systemPrompt | ❌ | Custom system prompt (defaults to empty) | | executionMode | ❌ | "auto" / "local" / "sandbox" (defaults to "local") | | expiresAt | ❌ | Expiration date "YYYY-MM-DD" (defaults to null, no expiration) | | notifyPlatforms | ❌ | Notification platform array: ["dingtalk","feishu","telegram","discord"] (defaults to []) | | enabled | ❌ | Whether to enable immediately (defaults to true) |
API returns JSON response:
{ "success": true, "task": { "id": "...", "name": "...", ... } }{ "success": false, "error": "error message" }Confirm the following with the user:
prompt should describe "what to do when the task triggers", not pre-execute the task and embed static results. Example: write "Fetch yesterday's AI news and send summary" instead of fetching news first and embedding the list in the prompt.bash node -e 'const d=new Date();const p=n=>String(n).padStart(2,"0");console.log(${d.getFullYear()}-${p(d.getMonth()+1)}-${p(d.getDate())}T${p(d.getHours())}:${p(d.getMinutes())}:${p(d.getSeconds())})'
prompt is the only instruction Claude receives when the task runs independently, so write it clearly and completelytype: "at" are automatically disabled after execution| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-22 | pass→pass | 11,429 | 5,414 | -53% | 1 | 1 | 0% | 1,968 | 2,191 | +11% | 0 | 0 | — |
case-01 | fail→fail | 14,548 | 10,934 | -25% | 1 | 1 | 0% | 2,916 | 2,239 | -23% | 0 | 0 | — |
case-02 | fail→fail | 7,349 | 16,504 | +125% | 1 | 1 | 0% | 1,172 | 4,322 | +269% | 0 | 0 | — |
case-03 | fail→pass | 7,994 | 24,820 | +210% | 1 | 1 | 0% | 558 | 4,882 | +775% | 0 | 0 | — |
case-04 | fail→pass | 4,667 | 4,768 | +2% | 1 | 1 | 0% | 822 | 2,286 | +178% | 0 | 0 | — |
case-05 | fail→pass | 9,792 | 5,161 | -47% | 1 | 1 | 0% | 1,777 | 2,219 | +25% | 0 | 0 | — |
case-06 | pass→pass | 8,439 | 3,990 | -53% | 1 | 1 | 0% | 1,564 | 2,104 | +35% | 0 | 0 | — |
case-21 | pass→pass | 17,787 | 12,527 | -30% | 1 | 1 | 0% | 3,271 | 3,709 | +13% | 0 | 0 | — |
case-07 | fail→pass | 4,029 | 4,177 | +4% | 1 | 1 | 0% | 745 | 2,099 | +182% | 0 | 0 | — |
case-08 | fail→fail | 7,488 | 6,529 | -13% | 1 | 1 | 0% | 1,366 | 2,477 | +81% | 0 | 0 | — |
case-09 | pass→pass | 7,744 | 3,969 | -49% | 1 | 1 | 0% | 1,417 | 2,117 | +49% | 0 | 0 | — |
case-10 | fail→pass | 10,492 | 5,483 | -48% | 1 | 1 | 0% | 2,158 | 2,291 | +6% | 0 | 0 | — |
case-11 | fail→pass | 6,627 | 4,019 | -39% | 1 | 1 | 0% | 1,259 | 2,109 | +68% | 0 | 0 | — |
case-12 | fail→pass | 4,300 | 4,349 | +1% | 1 | 1 | 0% | 729 | 2,123 | +191% | 0 | 0 | — |
case-13 | fail→pass | 4,149 | 4,849 | +17% | 1 | 1 | 0% | 756 | 2,256 | +198% | 0 | 0 | — |
case-14 | fail→fail | 8,498 | 8,264 | -3% | 1 | 1 | 0% | 1,599 | 2,807 | +76% | 0 | 0 | — |
case-15 | pass→pass | 7,727 | 4,926 | -36% | 1 | 1 | 0% | 1,446 | 2,208 | +53% | 0 | 0 | — |
case-16 | pass→pass | 6,817 | 4,130 | -39% | 1 | 1 | 0% | 1,266 | 2,099 | +66% | 0 | 0 | — |
case-17 | pass→pass | 4,210 | 2,876 | -32% | 1 | 1 | 0% | 721 | 1,826 | +153% | 0 | 0 | — |
case-18 | pass→pass | 4,600 | 3,028 | -34% | 1 | 1 | 0% | 821 | 1,941 | +136% | 0 | 0 | — |
case-19 | fail→pass | 9,318 | 4,299 | -54% | 1 | 1 | 0% | 2,020 | 2,130 | +5% | 0 | 0 | — |
case-20 | fail→pass | 5,307 | 13,945 | +163% | 1 | 1 | 0% | 944 | 2,662 | +182% | 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 20 counted toward the lift figure. The other 2 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 20 comparable cases.
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.