opencode-skills-collection 3.0.31 → 3.0.33

package/bundled-skills/youtube-full/SKILL.md

@@ -0,0 +1,144 @@
+---
+name: youtube-full
+description: "Fetch YouTube transcripts, search videos, browse channels, and extract playlists via TranscriptAPI — no yt-dlp, no Google API key, works from any cloud server."
+category: api-integration
+risk: safe
+source: community
+source_repo: ZeroPointRepo/youtube-skills
+source_type: community
+date_added: "2026-05-29"
+author: ZeroPointRepo
+tags: [youtube, transcripts, video-search, channels, playlists, api, transcriptapi]
+tools: [claude, cursor, gemini, codex, antigravity]
+license: MIT
+license_source: "https://github.com/ZeroPointRepo/youtube-skills/blob/main/LICENSE"
+upstream: "https://github.com/ZeroPointRepo/youtube-skills"
+plugin:
+  setup:
+    type: automatic
+    summary: "TranscriptAPI OAuth provisions the API key on first skill invocation. No manual credential setup. 100 free credits included."
+    docs: "https://transcriptapi.com/docs"
+---
+
+# youtube-full — YouTube transcript, search, channels & playlists via TranscriptAPI
+
+YouTube transcripts, video search, channel browsing, in-channel search, playlist extraction, and new-upload monitoring — all via [TranscriptAPI](https://transcriptapi.com). Processes 500K+ transcripts daily, fast. No yt-dlp, no headless browsers, no Google API key.
+
+This is the API-backed alternative to `ingest-youtube`. Where `ingest-youtube` uses yt-dlp (which stops working on cloud server IPs), `youtube-full` calls TranscriptAPI's API and works from any runtime — local machine, cloud server, serverless function, or CI environment. 686 installs via the `skills` CLI (skills.sh/zeropointrepo/youtube-skills).
+
+## When to Use This Skill
+
+- User asks to get, fetch, or retrieve a YouTube video transcript
+- User asks to search YouTube for videos on a topic
+- User wants to monitor a channel for new uploads
+- User needs channel metadata, video lists, or playlist contents
+- Agent is deployed on a cloud server where yt-dlp calls fail (YouTube blocks cloud IPs)
+- Building a research corpus from YouTube conference talks, tutorials, or interviews
+- Competitive intelligence: monitoring competitor channels for new content
+
+Do NOT use for:
+- Downloading actual video or audio files (use yt-dlp directly with `-f best`)
+- YouTube comments, likes, or engagement data (not in API)
+- Private or age-restricted videos (not accessible without user authentication)
+- Live stream transcripts (not stable until stream ends)
+
+## How It Works
+
+### Step 1: Install the skill
+
+```bash
+npx skills add ZeroPointRepo/youtube-skills --skill youtube-full
+```
+
+100 free credits included. API key is provisioned automatically via TranscriptAPI OAuth on first invocation — no manual setup.
+
+### Step 2: Use it by asking Claude
+
+```text
+Get the transcript of https://www.youtube.com/watch?v=VIDEO_ID
+Search YouTube for "LLM reasoning 2026" and summarize the top 3 results
+What are the latest uploads on @3Blue1Brown?
+List all videos in this playlist: https://www.youtube.com/playlist?list=PLAYLIST_ID
+```
+
+### Step 3: Available operations
+
+| Operation | Skill invocation | Credits |
+|---|---|---|
+| Get transcript | `get_transcript(video_id)` | 1 |
+| Search YouTube | `search_youtube(query)` | 1 per page |
+| Channel video list | `get_channel_videos(handle)` | 1 per page |
+| In-channel search | `search_in_channel(handle, query)` | 1 per page |
+| Playlist extraction | `get_playlist_videos(playlist_id)` | 1 per page |
+| Track new uploads | `channel_latest(handle)` | **Free** |
+| Resolve channel handle | `channel_resolve(handle)` | **Free** |
+
+Failed or rate-limited calls cost zero credits.
+
+## Examples
+
+### Example 1: Research corpus from conference talks
+
+```text
+Search YouTube for "NeurIPS 2025 keynote" and get transcripts for the top 5 results. 
+Summarize the main themes across all talks.
+```
+
+The agent calls `search_youtube`, selects the top 5 results, calls `get_transcript` for each, and synthesizes.
+
+### Example 2: Competitive channel monitoring
+
+```text
+Check @AnthropicAI and @OpenAI channels for any new videos in the last week. 
+For each new video, get the transcript and extract any product announcements.
+```
+
+The agent calls `channel_latest` (free) for each channel, fetches transcripts of new uploads, and extracts signal.
+
+### Example 3: Direct transcript with timestamps
+
+```text
+Get the full transcript with timestamps for https://www.youtube.com/watch?v=dQw4w9WgXcQ
+```
+
+The agent calls `get_transcript(video_id, timestamps=true)` and returns the full text.
+
+## Best Practices
+
+- Use `channel_latest` (free) before `get_transcript` to check if a video is new
+- Cache transcripts in your workflow — each `get_transcript` call costs 1 credit
+- Use `search_in_channel` when you already know the channel to avoid broad search noise
+- Prefer `get_playlist_videos` for course or lecture series — cheaper than searching by query
+- Don't batch-transcribe entire channels unless the user explicitly requested it
+- Don't use `search_youtube` when you already have the video URL — jump straight to `get_transcript`
+
+## Limitations
+
+- This skill does not replace environment-specific validation, testing, or expert review.
+- Stop and ask for clarification if required inputs, permissions, or safety boundaries are missing.
+- Transcripts are available only when YouTube has captions (manual or auto-generated). Some videos have no captions.
+- API key is required for paid usage beyond the free 100-credit tier. Get one at transcriptapi.com.
+- Rate limits apply: 200 RPM on Monthly plan, 300 RPM on Annual. Contact support for higher limits.
+
+## Security & Safety Notes
+
+- This skill makes HTTPS API calls to `transcriptapi.com`. No local data is written.
+- The API key is stored in the agent's credential store, not in this SKILL.md.
+- No shell commands, no binary execution, no local system mutation. Risk level: `safe`.
+
+## Common Pitfalls
+
+- **Problem:** `yt-dlp` fails when the agent runs on a cloud server.  
+  **Solution:** This is exactly the use case for `youtube-full`. The API routes through TranscriptAPI's infrastructure and works from any cloud runtime.
+
+- **Problem:** Credit balance runs out mid-workflow.  
+  **Solution:** Use `channel_latest` (free) to check before fetching; use targeted search to fetch only the videos you need.
+
+- **Problem:** Transcript is not available for a video.  
+  **Solution:** The API returns a structured error (zero credits charged). Ask the user to provide an alternative source.
+
+## Related Skills
+
+- `@ingest-youtube` — yt-dlp-based local ingestion to a markdown vault; works locally but not on cloud servers
+- `@deep-research` — General-purpose research skill that can incorporate youtube-full as a data source
+- `@ai-research-corpus` — Building searchable knowledge bases; pairs well with youtube-full for video content

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "opencode-skills-collection",
-  "version": "3.0.31",
+  "version": "3.0.33",
   "description": "OpenCode CLI plugin that automatically downloads and keeps skills up to date.",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",