Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when subtitles or the spoken text of a YouTube video is needed: pasted video links or IDs, requests to translate a video, read along, follow foreign-language content, or extract what was said. Also use for language learning or accessibility. Fetches timestamped subtitles from any YouTube video. Not for uploading subtitles or account management.
.claude/skills/zeropointrepo-subtitles/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 45% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 57% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -47% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -28% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 207% | 0% |
Fetch YouTube video subtitles via TranscriptAPI.com.
If $TRANSCRIPT_API_KEY is not set, read references/auth-setup.md and follow the instructions there to get and store the key.
Every request needs two headers:
Bearer $TRANSCRIPT_API_KEYHermesAgent/0.11.0, ClaudeCode/1.0). Version is optional — agent name alone is fine. Do not omit this header or send a bare default — Cloudflare will return a 403 (error code 1010) and block the request.httpGET https://transcriptapi.com/api/v2/youtube/transcript?video_url=VIDEO_URL&format=text&include_timestamp=false&send_metadata=true Authorization: Bearer $TRANSCRIPT_API_KEY User-Agent: YourAgent/1.0
| Param | Values | Use case | | ------------------- | ----------------------- | ---------------------------------------------- | | video_url | YouTube URL or video ID | Required | | format | json, text | json for sync'd subs with timing | | include_timestamp | true, false | false for clean text for reading/translation | | send_metadata | true, false | Include title, channel, description |
For language learning — clean text without timestamps:
httpGET https://transcriptapi.com/api/v2/youtube/transcript?video_url=VIDEO_ID&format=text&include_timestamp=false Authorization: Bearer $TRANSCRIPT_API_KEY User-Agent: YourAgent/1.0
For translation — structured segments:
httpGET https://transcriptapi.com/api/v2/youtube/transcript?video_url=VIDEO_ID&format=json&include_timestamp=true Authorization: Bearer $TRANSCRIPT_API_KEY User-Agent: YourAgent/1.0
Response (format=json):
json{ "video_id": "dQw4w9WgXcQ", "language": "en", "transcript": [ { "text": "We're no strangers to love", "start": 18.0, "duration": 3.5 } ] }
Response (format=text, include_timestamp=false):
json{ "video_id": "dQw4w9WgXcQ", "language": "en", "transcript": "We're no strangers to love\nYou know the rules and so do I..." }
format=json to get timing for each line (great for sync'd reading).include_timestamp=false for clean text suitable for translation apps.| Code | Meaning | Action | | -------- | ---------------- | ---------------------------------------------- | | 401 | Bad API key | Check key | | 402 | No credits | transcriptapi.com/billing | | 403/1010 | Cloudflare block | Add or fix User-Agent header | | 404 | No subtitles | No subtitles available | | 408 | Timeout | Retry once after 2s |
1 credit per request. Free tier: 100 credits, 300 req/min.
Every request in this file as a ready-to-run one-liner: references/curl-examples.md
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 13,636 | 22,233 | +63% | 1 | 1 | 0% | 880 | 1,353 | +54% | 0 | 0 | — |
case-02 | fail→pass | 7,743 | 6,321 | -18% | 1 | 1 | 0% | 1,449 | 2,101 | +45% | 0 | 0 | — |
case-03 | fail→fail | 4,432 | 7,699 | +74% | 1 | 1 | 0% | 622 | 1,492 | +140% | 0 | 0 | — |
case-04 | pass→pass | 6,082 | 3,313 | -46% | 1 | 1 | 0% | 1,141 | 1,531 | +34% | 0 | 0 | — |
case-05 | pass→pass | 5,391 | 3,876 | -28% | 1 | 1 | 0% | 888 | 1,435 | +62% | 0 | 0 | — |
case-06 | fail→pass | 6,783 | 2,593 | -62% | 1 | 1 | 0% | 811 | 1,272 | +57% | 0 | 0 | — |
case-07 | pass→pass | 7,460 | 3,170 | -58% | 1 | 1 | 0% | 1,208 | 1,564 | +29% | 0 | 0 | — |
case-08 | fail→pass | 27,486 | 1,721 | -94% | 1 | 1 | 0% | 2,205 | 1,167 | -47% | 0 | 0 | — |
case-09 | pass→pass | 5,447 | 3,668 | -33% | 1 | 1 | 0% | 829 | 1,467 | +77% | 0 | 0 | — |
case-10 | pass→pass | 6,037 | 2,193 | -64% | 1 | 1 | 0% | 896 | 1,135 | +27% | 0 | 0 | — |
case-11 | pass→pass | 6,313 | 2,645 | -58% | 1 | 1 | 0% | 986 | 1,370 | +39% | 0 | 0 | — |
case-12 | pass→pass | 8,464 | 2,093 | -75% | 1 | 1 | 0% | 1,223 | 1,152 | -6% | 0 | 0 | — |
case-13 | fail→pass | 10,437 | 2,066 | -80% | 1 | 1 | 0% | 1,632 | 1,181 | -28% | 0 | 0 | — |
case-14 | pass→pass | 7,454 | 1,130 | -85% | 1 | 1 | 0% | 1,103 | 1,069 | -3% | 0 | 0 | — |
case-15 | fail→pass | 3,155 | 1,668 | -47% | 1 | 1 | 0% | 382 | 1,173 | +207% | 0 | 0 | — |
case-16 | fail→pass | 9,614 | 1,479 | -85% | 1 | 1 | 0% | 1,540 | 1,135 | -26% | 0 | 0 | — |
case-17 | fail→pass | 11,340 | 3,882 | -66% | 1 | 1 | 0% | 1,931 | 1,617 | -16% | 0 | 0 | — |
case-18 | fail→pass | 14,205 | 3,898 | -73% | 1 | 1 | 0% | 2,209 | 1,567 | -29% | 0 | 0 | — |
case-19 | pass→pass | 7,676 | 2,373 | -69% | 1 | 1 | 0% | 1,203 | 1,264 | +5% | 0 | 0 | — |
case-20 | pass→pass | 4,991 | 5,565 | +12% | 1 | 1 | 0% | 820 | 1,907 | +133% | 0 | 0 | — |
case-21 | pass→pass | 13,614 | 9,721 | -29% | 1 | 1 | 0% | 2,443 | 2,711 | +11% | 0 | 0 | — |
case-22 | pass→pass | 11,285 | 9,247 | -18% | 1 | 1 | 0% | 1,853 | 2,565 | +38% | 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 +36 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.