youtube

Name: youtube
Author: tuan3w

32stars2forks

How to install youtube?

tuan3w/obsidian-vault-agent/youtube

$npx -y skills add tuan3w/obsidian-vault-agent --skill youtube

Installs into the current project.

Summary

Analyze a YouTube video and create a vault note from its transcript. Use this skill whenever a YouTube URL is pasted — even with casual commentary like "check this out" or "interesting talk." Also triggers on "youtube", "take notes from this video", "summarize this video", or any

Files · 1

View on GitHub

SKILL.md

1	<Purpose>
2	Extract transcript and metadata from a YouTube video, then synthesize it into
3	a vault-formatted post note. Uses youtube-transcript-api for transcript and
4	yt-dlp for rich metadata (with oEmbed fallback).
5	</Purpose>
6
7	<Use_When>
8	- User shares a YouTube URL and wants notes taken
9	- User says "take notes from this video"
10	- User pastes a YouTube link with /youtube
11	- User wants to add a video's insights to the vault
12	</Use_When>
13
14	<Do_Not_Use_When>
15	- User wants to watch or download the video itself
16	- User has a local video file (not YouTube)
17	- Video has no transcript/captions at all
18	- User wants to process an existing vault note (use /process)
19	</Do_Not_Use_When>
20
21	<Execution_Policy>
22	- Extract first, synthesize second, integrate third
23	- Always check vault for existing notes on the same video before creating
24	- Create note as type: post with processing_status: inbox
25	- The note is a starting point — user can /process it later for deeper engagement
26	</Execution_Policy>
27
28	<Steps>
29
30	## Stage 1: EXTRACT
31
32	Parse the YouTube URL/ID from $ARGUMENTS. If no URL provided, ask the user.
33
34	Run the extraction script. Redirect stdout to a temp file (stderr has progress messages that break piping):
35
36	```bash
37	SKILL_DIR="${CLAUDE_SKILL_DIR}"
38	YT_OUTPUT="$SKILL_DIR/_output.json"
39	uv run "$SKILL_DIR/scripts/fetch_youtube.py" "VIDEO_URL" --lang en > "$YT_OUTPUT"
40	```
41
42	Then read `$YT_OUTPUT` with the Read tool to get the JSON. Clean up the file after use.
43
44	The JSON contains:
45	- `title`, `channel`, `duration`, `upload_date`, `description`, `chapters`
46	- `transcript.full_text`, `transcript.segments`, `transcript.language`
47	- `transcript.error` (null if success)
48
49	If transcript.error is not null: inform the user and stop. No note without content.
50
51	If transcript is very long (>50,000 chars): warn the user this is a long video.
52	Chunk the transcript for the agent if needed — send first 40,000 chars with a note
53	about total length. For very long videos (>2hrs), consider suggesting the user
54	watch key sections instead.
55
56	## Stage 2: SYNTHESIZE
57
58	Read the agent definition:
59	```
60	Read("${CLAUDE_SKILL_DIR}/agents/video-noter.md")
61	```
62
63	Search the vault for existing notes related to the video's topics using the MCP tool:
64	```
65	search_notes(query="KEYWORD", limit=20)
66	```
67	Or fall back to Grep if MCP is unavailable:
68	```
69	Grep(pattern="KEYWORD", path="notes/", glob="*.md", head_limit=20)
70	```
71
72	Launch the video-noter agent:
73
74	```
75	Agent(
76	subagent_type="general-purpose",
77	model="sonnet",
78	run_in_background=false,
79	prompt="You are Video Noter. Follow these instructions exactly:
80
81	[INSERT FULL CONTENT OF agents/video-noter.md HERE]
82
83	VIDEO METADATA:
84	- Title: [title]
85	- Channel: [channel]
86	- Duration: [duration]
87	- Upload date: [upload_date]
88	- Chapters: [chapters if any]
89	- Description: [first 500 chars of description]
90
91	EXISTING VAULT NOTES ON RELATED TOPICS:
92	[List any matching notes found in grep]
93
94	TRANSCRIPT:
95	[full_text or chunked text]
96
97	Produce the note body following the Output Format. Do NOT include frontmatter —
98	only the body starting from the # title line."
99	)
100	```
101
102	## Stage 3: INTEGRATE
103
104	1. Generate timestamp ID:
105	```bash
106	date +%Y%m%d%H%M%S
107	```
108
109	2. Format the upload_date for frontmatter (YYYYMMDD → YYYY-MM-DD)
110
111	3. Check for duplicate notes:
112	```bash
113	grep -rl "VIDEO_TITLE" notes/ --include="*.md" \| head -5
114	```
115
116	4. Create the note file with frontmatter + agent output:
117
118	```markdown
119	---
120	id: YYYYMMDDHHMMSS
121	type: post
122	processing_status: inbox
123	author: Channel Name
124	link: "https://www.youtube.com/watch?v=VIDEO_ID"
125	created_date: YYYY-MM-DD
126	updated_date: YYYY-MM-DD
127	---
128
129	[AGENT OUTPUT HERE — starts with # title and 🏷️Tags line]
130	```
131
132	Place the note in `notes/` under the most appropriate topic subfolder:
133	- ML/AI content → `notes/ml/`
134	- Startup/business → `notes/startup/`
135	- Finance → `notes/finance/`
136	- Design → `notes/design/`
137	- Psychology → `notes/psychology/`
138	- General/unclear → `notes/`
139
140	5. Report to user:
141	- Note path and title
142	- Number of concepts suggested for extraction
143	- Any related vault notes found
144	- Remind: "Run /process on this note when you're ready to deepen it"
145
146	</Steps>
147
148	<Tool_Usage>
149	- Bash: Run fetch_youtube.py script (via uv run), generate timestamps, search vault
150	- Read: Read agent definition, read existing vault notes for context
151	- Write: Create the post note in vault
152	- Agent: Delegate synthesis to video-noter agent (sonnet)
153	- Grep/Glob: Search vault for duplicates and related notes
154	- TodoWrite: Track progre

People who liked this also viewed

6 picks

Type

youtube

32stars2forks

View on GitHub

How to install youtube?

tuan3w/obsidian-vault-agent/youtube

$npx -y skills add tuan3w/obsidian-vault-agent --skill youtube

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	<Purpose>
2	Extract transcript and metadata from a YouTube video, then synthesize it into
3	a vault-formatted post note. Uses youtube-transcript-api for transcript and
4	yt-dlp for rich metadata (with oEmbed fallback).
5	</Purpose>
6
7	<Use_When>
8	- User shares a YouTube URL and wants notes taken
9	- User says "take notes from this video"
10	- User pastes a YouTube link with /youtube
11	- User wants to add a video's insights to the vault
12	</Use_When>
13
14	<Do_Not_Use_When>
15	- User wants to watch or download the video itself
16	- User has a local video file (not YouTube)
17	- Video has no transcript/captions at all
18	- User wants to process an existing vault note (use /process)
19	</Do_Not_Use_When>
20
21	<Execution_Policy>
22	- Extract first, synthesize second, integrate third
23	- Always check vault for existing notes on the same video before creating
24	- Create note as type: post with processing_status: inbox
25	- The note is a starting point — user can /process it later for deeper engagement
26	</Execution_Policy>
27
28	<Steps>
29
30	## Stage 1: EXTRACT
31
32	Parse the YouTube URL/ID from $ARGUMENTS. If no URL provided, ask the user.
33
34	Run the extraction script. Redirect stdout to a temp file (stderr has progress messages that break piping):
35
36	```bash
37	SKILL_DIR="${CLAUDE_SKILL_DIR}"
38	YT_OUTPUT="$SKILL_DIR/_output.json"
39	uv run "$SKILL_DIR/scripts/fetch_youtube.py" "VIDEO_URL" --lang en > "$YT_OUTPUT"
40	```
41
42	Then read `$YT_OUTPUT` with the Read tool to get the JSON. Clean up the file after use.
43
44	The JSON contains:
45	- `title`, `channel`, `duration`, `upload_date`, `description`, `chapters`
46	- `transcript.full_text`, `transcript.segments`, `transcript.language`
47	- `transcript.error` (null if success)
48
49	If transcript.error is not null: inform the user and stop. No note without content.
50
51	If transcript is very long (>50,000 chars): warn the user this is a long video.
52	Chunk the transcript for the agent if needed — send first 40,000 chars with a note
53	about total length. For very long videos (>2hrs), consider suggesting the user
54	watch key sections instead.
55
56	## Stage 2: SYNTHESIZE
57
58	Read the agent definition:
59	```
60	Read("${CLAUDE_SKILL_DIR}/agents/video-noter.md")
61	```
62
63	Search the vault for existing notes related to the video's topics using the MCP tool:
64	```
65	search_notes(query="KEYWORD", limit=20)
66	```
67	Or fall back to Grep if MCP is unavailable:
68	```
69	Grep(pattern="KEYWORD", path="notes/", glob="*.md", head_limit=20)
70	```
71
72	Launch the video-noter agent:
73
74	```
75	Agent(
76	subagent_type="general-purpose",
77	model="sonnet",
78	run_in_background=false,
79	prompt="You are Video Noter. Follow these instructions exactly:
80
81	[INSERT FULL CONTENT OF agents/video-noter.md HERE]
82
83	VIDEO METADATA:
84	- Title: [title]
85	- Channel: [channel]
86	- Duration: [duration]
87	- Upload date: [upload_date]
88	- Chapters: [chapters if any]
89	- Description: [first 500 chars of description]
90
91	EXISTING VAULT NOTES ON RELATED TOPICS:
92	[List any matching notes found in grep]
93
94	TRANSCRIPT:
95	[full_text or chunked text]
96
97	Produce the note body following the Output Format. Do NOT include frontmatter —
98	only the body starting from the # title line."
99	)
100	```
101
102	## Stage 3: INTEGRATE
103
104	1. Generate timestamp ID:
105	```bash
106	date +%Y%m%d%H%M%S
107	```
108
109	2. Format the upload_date for frontmatter (YYYYMMDD → YYYY-MM-DD)
110
111	3. Check for duplicate notes:
112	```bash
113	grep -rl "VIDEO_TITLE" notes/ --include="*.md" \| head -5
114	```
115
116	4. Create the note file with frontmatter + agent output:
117
118	```markdown
119	---
120	id: YYYYMMDDHHMMSS
121	type: post
122	processing_status: inbox
123	author: Channel Name
124	link: "https://www.youtube.com/watch?v=VIDEO_ID"
125	created_date: YYYY-MM-DD
126	updated_date: YYYY-MM-DD
127	---
128
129	[AGENT OUTPUT HERE — starts with # title and 🏷️Tags line]
130	```
131
132	Place the note in `notes/` under the most appropriate topic subfolder:
133	- ML/AI content → `notes/ml/`
134	- Startup/business → `notes/startup/`
135	- Finance → `notes/finance/`
136	- Design → `notes/design/`
137	- Psychology → `notes/psychology/`
138	- General/unclear → `notes/`
139
140	5. Report to user:
141	- Note path and title
142	- Number of concepts suggested for extraction
143	- Any related vault notes found
144	- Remind: "Run /process on this note when you're ready to deepen it"
145
146	</Steps>
147
148	<Tool_Usage>
149	- Bash: Run fetch_youtube.py script (via uv run), generate timestamps, search vault
150	- Read: Read agent definition, read existing vault notes for context
151	- Write: Create the post note in vault
152	- Agent: Delegate synthesis to video-noter agent (sonnet)
153	- Grep/Glob: Search vault for duplicates and related notes
154	- TodoWrite: Track progre

People who liked this also viewed

6 picks

Type