lecture

Name: lecture
Author: tuan3w

32stars2forks

How to install lecture?

tuan3w/obsidian-vault-agent/lecture

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

Installs into the current project.

Summary

Extract transcript and key slides from a local video file, then create a vault-formatted lecture note. Use this skill whenever a user provides a local video file (.mp4, .mov, .mkv, .avi, .webm) and wants notes from it. Triggers on "lecture", "take notes", "I have a recording of",

Files · 1

View on GitHub

SKILL.md

1	<Purpose>
2	Transcribe a local video lecture using mlx-whisper (Apple Silicon), extract key
3	slide frames with timestamps, then synthesize into a vault-formatted lecture note
4	with embedded screenshots. Auto-detects transcript language — works with any language,
5	outputs English notes.
6	</Purpose>
7
8	<Use_When>
9	- User provides a local video file and wants lecture notes
10	- User says "take notes from this lecture/video"
11	- User uses /lecture with a file path
12	- User has an MP4/MOV/MKV file to process
13	</Use_When>
14
15	<Do_Not_Use_When>
16	- User has a YouTube URL (use /youtube instead)
17	- User wants to process an existing vault note (use /process)
18	- User wants audio-only transcription without note synthesis
19	</Do_Not_Use_When>
20
21	<Prerequisites>
22	- `ffmpeg` installed (for audio extraction and frame capture)
23	- `uv` installed (`brew install uv` or `curl -LsSf https://astral.sh/uv/install.sh \| sh`)
24	- Apple Silicon Mac (mlx-whisper is optimized for M-series chips)
25	</Prerequisites>
26
27	<Execution_Policy>
28	- Extract first, synthesize second, integrate third
29	- Always check vault for existing notes on the same topic before creating
30	- Create note as type: lecture with processing_status: inbox
31	- The note is a starting point — user can /process it later for deeper engagement
32	- Transcription can take several minutes for long videos — inform the user
33	</Execution_Policy>
34
35	<Steps>
36
37	## Stage 1: EXTRACT
38
39	Parse the video file path from $ARGUMENTS. If no path provided, ask the user.
40	Verify the file exists and is a video format (mp4, mov, mkv, avi, webm).
41
42	Run the extraction script:
43
44	```bash
45	SKILL_DIR="${CLAUDE_SKILL_DIR}"
46	LECTURE_OUTPUT="temp/lecture-extract-output.json"
47	uv run "$SKILL_DIR/scripts/extract_lecture.py" "VIDEO_PATH" > "$LECTURE_OUTPUT" 2>&1 &
48	```
49
50	IMPORTANT: This script takes time (several minutes for a 30-60 min video).
51	Inform the user: "Extracting audio and transcribing — this will take a few minutes for a [duration] video."
52
53	Run it and wait for completion. Then read the output JSON.
54
55	The JSON contains:
56	- `filename`, `duration`, `duration_seconds`, `width`, `height`
57	- `transcript.full_text`, `transcript.segments` (with start/end times), `transcript.language`
58	- `transcript.error` (null if success)
59	- `frames[]` — array of `{path, timestamp_seconds, timestamp}` for each extracted slide
60	- `output_dir` — temp directory with extracted frames
61
62	If transcript.error is not null: inform the user and stop. Check if mlx-whisper is installed.
63
64	If transcript is very long (>80,000 chars): warn the user. Send first 60,000 chars to the agent
65	with a note about total length.
66
67	## Stage 2: PREPARE FRAMES
68
69	Copy the extracted frames to the vault's assets directory with a descriptive naming scheme:
70
71	```bash
72	# Generate a slug from the video filename
73	SLUG=$(echo "VIDEO_FILENAME" \| sed 's/\.[^.]*$//' \| tr '[:upper:]' '[:lower:]' \| tr ' ' '-' \| sed 's/[^a-z0-9-]//g' \| cut -c1-30)
74	ASSETS_DIR="assets"
75
76	for frame in FRAME_PATHS; do
77	FRAME_NUM=$(basename "$frame" \| grep -o '[0-9]*')
78	cp "$frame" "$ASSETS_DIR/lecture-${SLUG}-${FRAME_NUM}.jpg"
79	done
80	```
81
82	Build a frame manifest for the noter agent — each frame gets:
83	- Its vault filename (for `![[embedding]]`)
84	- Its timestamp in the video (e.g., "12:30")
85
86	Example manifest:
87	```
88	FRAMES WITH TIMESTAMPS:
89	- lecture-risk-mgmt-01.jpg (timestamp: 0:10)
90	- lecture-risk-mgmt-02.jpg (timestamp: 2:00)
91	- lecture-risk-mgmt-03.jpg (timestamp: 4:00)
92	...
93	```
94
95	## Stage 3: SYNTHESIZE
96
97	Read the agent definition:
98	```
99	Read("${CLAUDE_SKILL_DIR}/agents/lecture-noter.md")
100	```
101
102	Search the vault for existing notes related to the lecture's topics using the MCP tool:
103	```
104	search_notes(query="KEYWORD", limit=20)
105	```
106	Or fall back to Grep if MCP is unavailable:
107	```
108	Grep(pattern="KEYWORD", path="notes/", glob="*.md", head_limit=20)
109	```
110
111	Review each frame using the Read tool to see what's on each slide.
112	Build a brief description of each frame's content (1 line each) to include in the agent prompt.
113
114	Launch the lecture-noter agent:
115
116	```
117	Agent(
118	subagent_type="general-purpose",
119	model="sonnet",
120	run_in_background=false,
121	prompt="You are Lecture Noter. Follow these instructions exactly:
122
123	[INSERT FULL CONTENT OF agents/lecture-noter.md HERE]
124
125	VIDEO METADATA:
126	- Filename: [filename]
127	- Duration: [duration]
128	- Transcript language: [language from extraction JSON]
129
130	FRAMES WITH TIMESTAMPS AND DESCRIPTIONS:
131	- lecture-slug-01.jpg (timestamp: 0:10) — Title slide showing cou

People who liked this also viewed

6 picks

Type

lecture

32stars2forks

View on GitHub

How to install lecture?

tuan3w/obsidian-vault-agent/lecture

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

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	<Purpose>
2	Transcribe a local video lecture using mlx-whisper (Apple Silicon), extract key
3	slide frames with timestamps, then synthesize into a vault-formatted lecture note
4	with embedded screenshots. Auto-detects transcript language — works with any language,
5	outputs English notes.
6	</Purpose>
7
8	<Use_When>
9	- User provides a local video file and wants lecture notes
10	- User says "take notes from this lecture/video"
11	- User uses /lecture with a file path
12	- User has an MP4/MOV/MKV file to process
13	</Use_When>
14
15	<Do_Not_Use_When>
16	- User has a YouTube URL (use /youtube instead)
17	- User wants to process an existing vault note (use /process)
18	- User wants audio-only transcription without note synthesis
19	</Do_Not_Use_When>
20
21	<Prerequisites>
22	- `ffmpeg` installed (for audio extraction and frame capture)
23	- `uv` installed (`brew install uv` or `curl -LsSf https://astral.sh/uv/install.sh \| sh`)
24	- Apple Silicon Mac (mlx-whisper is optimized for M-series chips)
25	</Prerequisites>
26
27	<Execution_Policy>
28	- Extract first, synthesize second, integrate third
29	- Always check vault for existing notes on the same topic before creating
30	- Create note as type: lecture with processing_status: inbox
31	- The note is a starting point — user can /process it later for deeper engagement
32	- Transcription can take several minutes for long videos — inform the user
33	</Execution_Policy>
34
35	<Steps>
36
37	## Stage 1: EXTRACT
38
39	Parse the video file path from $ARGUMENTS. If no path provided, ask the user.
40	Verify the file exists and is a video format (mp4, mov, mkv, avi, webm).
41
42	Run the extraction script:
43
44	```bash
45	SKILL_DIR="${CLAUDE_SKILL_DIR}"
46	LECTURE_OUTPUT="temp/lecture-extract-output.json"
47	uv run "$SKILL_DIR/scripts/extract_lecture.py" "VIDEO_PATH" > "$LECTURE_OUTPUT" 2>&1 &
48	```
49
50	IMPORTANT: This script takes time (several minutes for a 30-60 min video).
51	Inform the user: "Extracting audio and transcribing — this will take a few minutes for a [duration] video."
52
53	Run it and wait for completion. Then read the output JSON.
54
55	The JSON contains:
56	- `filename`, `duration`, `duration_seconds`, `width`, `height`
57	- `transcript.full_text`, `transcript.segments` (with start/end times), `transcript.language`
58	- `transcript.error` (null if success)
59	- `frames[]` — array of `{path, timestamp_seconds, timestamp}` for each extracted slide
60	- `output_dir` — temp directory with extracted frames
61
62	If transcript.error is not null: inform the user and stop. Check if mlx-whisper is installed.
63
64	If transcript is very long (>80,000 chars): warn the user. Send first 60,000 chars to the agent
65	with a note about total length.
66
67	## Stage 2: PREPARE FRAMES
68
69	Copy the extracted frames to the vault's assets directory with a descriptive naming scheme:
70
71	```bash
72	# Generate a slug from the video filename
73	SLUG=$(echo "VIDEO_FILENAME" \| sed 's/\.[^.]*$//' \| tr '[:upper:]' '[:lower:]' \| tr ' ' '-' \| sed 's/[^a-z0-9-]//g' \| cut -c1-30)
74	ASSETS_DIR="assets"
75
76	for frame in FRAME_PATHS; do
77	FRAME_NUM=$(basename "$frame" \| grep -o '[0-9]*')
78	cp "$frame" "$ASSETS_DIR/lecture-${SLUG}-${FRAME_NUM}.jpg"
79	done
80	```
81
82	Build a frame manifest for the noter agent — each frame gets:
83	- Its vault filename (for `![[embedding]]`)
84	- Its timestamp in the video (e.g., "12:30")
85
86	Example manifest:
87	```
88	FRAMES WITH TIMESTAMPS:
89	- lecture-risk-mgmt-01.jpg (timestamp: 0:10)
90	- lecture-risk-mgmt-02.jpg (timestamp: 2:00)
91	- lecture-risk-mgmt-03.jpg (timestamp: 4:00)
92	...
93	```
94
95	## Stage 3: SYNTHESIZE
96
97	Read the agent definition:
98	```
99	Read("${CLAUDE_SKILL_DIR}/agents/lecture-noter.md")
100	```
101
102	Search the vault for existing notes related to the lecture's topics using the MCP tool:
103	```
104	search_notes(query="KEYWORD", limit=20)
105	```
106	Or fall back to Grep if MCP is unavailable:
107	```
108	Grep(pattern="KEYWORD", path="notes/", glob="*.md", head_limit=20)
109	```
110
111	Review each frame using the Read tool to see what's on each slide.
112	Build a brief description of each frame's content (1 line each) to include in the agent prompt.
113
114	Launch the lecture-noter agent:
115
116	```
117	Agent(
118	subagent_type="general-purpose",
119	model="sonnet",
120	run_in_background=false,
121	prompt="You are Lecture Noter. Follow these instructions exactly:
122
123	[INSERT FULL CONTENT OF agents/lecture-noter.md HERE]
124
125	VIDEO METADATA:
126	- Filename: [filename]
127	- Duration: [duration]
128	- Transcript language: [language from extraction JSON]
129
130	FRAMES WITH TIMESTAMPS AND DESCRIPTIONS:
131	- lecture-slug-01.jpg (timestamp: 0:10) — Title slide showing cou

People who liked this also viewed

6 picks

Type