|
1
|
--- |
|
2
|
name: podcast-marketing |
|
3
|
description: Repurpose podcast episodes into social posts, show notes, clips, and quote cards. |
|
4
|
--- |
|
5
|
|
|
6
|
# Podcast Marketing |
|
7
|
|
|
8
|
Turn podcast episodes into a full suite of marketing content. Supports four input modes, extracts content atoms with viral scoring, and produces platform-ready content pieces that can be published immediately. |
|
9
|
|
|
10
|
## Requirements & Setup |
|
11
|
|
|
12
|
### Tier 1: Text Content (always needed) |
|
13
|
|
|
14
|
These are required for transcript extraction and content generation (show notes, blog, social, newsletter, pull quotes): |
|
15
|
|
|
16
|
| Dependency | Install | Purpose | |
|
17
|
|
|
18
|
|-----------|---------|---------| |
|
19
|
|
|
20
|
| Python 3.10+ | `installProgrammingLanguage({ language: "python-3.11" })` | Runtime for all scripts | |
|
21
|
|
|
22
|
| youtube-transcript-api | `installLanguagePackages({ language: "python", packages: ["youtube-transcript-api"] })` | YouTube caption extraction | |
|
23
|
|
|
24
|
| yt-dlp | `installLanguagePackages({ language: "python", packages: ["yt-dlp"] })` | YouTube metadata + video/audio download | |
|
25
|
|
|
26
|
| LLM API | Set up via Anthropic or OpenAI AI integration (Replit proxy — no user API key needed) | Content atom extraction and content generation | |
|
27
|
|
|
28
|
### Tier 2: Quote Card Images |
|
29
|
|
|
30
|
These are required to generate visual quote card PNG files: |
|
31
|
|
|
32
|
| Dependency | Install | Purpose | |
|
33
|
|
|
34
|
|-----------|---------|---------| |
|
35
|
|
|
36
|
| Pillow | `installLanguagePackages({ language: "python", packages: ["Pillow"] })` | Image generation | |
|
37
|
|
|
38
|
| System fonts | Already available on Nix (DejaVu Sans at `/usr/share/fonts/truetype/dejavu/`) | Typography | |
|
39
|
|
|
40
|
No API keys needed. Run `scripts/generate_quote_cards.py` directly. |
|
41
|
|
|
42
|
### Tier 3: Short-Form Video Clips (basic) |
|
43
|
|
|
44
|
These produce trimmed, vertical-reformatted video clips without animated captions: |
|
45
|
|
|
46
|
| Dependency | Install | Purpose | |
|
47
|
|
|
48
|
|-----------|---------|---------| |
|
49
|
|
|
50
|
| ffmpeg | `installSystemDependencies({ packages: ["ffmpeg"] })` | Video trimming, reformatting, subtitle burning | |
|
51
|
|
|
52
|
| yt-dlp | (already in Tier 1) | Download source video from YouTube | |
|
53
|
|
|
54
|
### Tier 4: Animated Captions (Opus Clips style) |
|
55
|
|
|
56
|
These add word-by-word highlighted captions to video clips: |
|
57
|
|
|
58
|
| Dependency | Install | Purpose | |
|
59
|
|
|
60
|
|-----------|---------|---------| |
|
61
|
|
|
62
|
| OpenAI API | Set up OpenAI AI integration OR user provides `OPENAI_API_KEY` | Whisper word-level timestamps | |
|
63
|
|
|
64
|
| Pillow | (already in Tier 2) | Caption frame generation (alternative to ASS subtitles) | |
|
65
|
|
|
66
|
The animated caption pipeline re-transcribes each trimmed clip through Whisper with `timestamp_granularities=["word"]` to get per-word timing. Without this, clips still work but use segment-level static captions instead. |
|
67
|
|
|
68
|
### Setup Checklist |
|
69
|
|
|
70
|
Before running the pipeline, verify: |
|
71
|
|
|
72
|
```python |
|
73
|
|
|
74
|
# Tier 1 |
|
75
|
|
|
76
|
import youtube_transcript_api \# YouTube captions |
|
77
|
|
|
78
|
import yt_dlp \# YouTube metadata/download |
|
79
|
|
|
80
|
import anthropic \# or openai — for content generation |
|
81
|
|
|
82
|
# Tier 2 (if generating quote cards) |
|
83
|
|
|
84
|
from PIL import Image, ImageFont |
|
85
|
|
|
86
|
ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf", 48) |
|
87
|
|
|
88
|
# Tier 3 (if generating video clips) |
|
89
|
|
|
90
|
import subprocess |
|
91
|
|
|
92
|
subprocess.run(["ffmpeg", "-version"], capture_output=True, check=True) |
|
93
|
|
|
94
|
# Tier 4 (if generating animated captions) |
|
95
|
|
|
96
|
import openai \# OPENAI_API_KEY must be set |
|
97
|
|
|
98
|
``` |
|
99
|
|
|
100
|
## When to Use |
|
101
|
|
|
102
|
- User wants to repurpose a podcast episode into marketing content |
|
103
|
- User asks to create social media posts, blog articles, or newsletters from an episode |
|
104
|
|
|
105
|
- User wants show notes with timestamps and key takeaways |
|
106
|
- User asks to generate pull quotes, audiogram clips, or short-form video clips |
|
107
|
|
|
108
|
- User provides an RSS feed and wants to process episodes in batch |
|
109
|
- User shares a YouTube link to a podcast episode or video interview |
|
110
|
|
|
111
|
- User says "turn this episode into content", "repurpose this podcast", or "create marketing from my episode" |
|
112
|
- User wants to create a content calendar from podcast episodes |
|
113
|
|
|
114
|
## When NOT to Use |
|
115
|
|
|
116
|
- Creating a podcast from scratch (use the `podcast-generator` skill instead) |
|
117
|
- General social media management unrelated to podcast content |
|
118
|
|
|
119
|
## Input Modes |
|
120
|
|
|
121
|
### Mode 1: RSS Feed |
|
122
|
|
|
123
|
Auto-download and transcribe the latest episode(s) from a podcast RSS feed. |
|
124
|
|
|
125
|
1. Parse the RSS feed to extract episode metadata (title, description, publish date, audio URL). |
|
126
|
2. Download the audio file. |
|
127
|
|
|
128
|
3. Transcribe using OpenAI Whisper (produces timestamped transcript). |
|
129
|
4. Proceed to content generation. |
|
130
|
|
|
131
|
Read `rss-and-batch.md` for RSS parsing details and the Whisper transcription pipeline. |
|
132
|
|
|
133
|
### Mode 2: Raw Transcript |
|
134
|
|
|
135
|
Accept a transcript file directly — plain text, SRT, or VTT format. |
|
136
|
|
|
137
|
1. Detect format from file extension or content structure. |
|
138
|
2. Parse into a normalized transcript (list of segments with optional timestamps). |
|
139
|
|
|
140
|
3. Proceed to content generation. |
|
141
|
|
|
142
|
SRT/VTT files preserve timestamps, which enables show notes with time markers. Plain text works but won't have timestamp data. |
|
143
|
|
|
144
|
### Mode 3: YouTube Link |
|
145
|
|
|
146
|
Extract transcript and metadata from a YouTube video — works for podcast episodes, interviews, talks, or any video with speech. |
|
147
|
|
|
148
|
1. Accept a YouTube URL (full or short format). |
|
149
|
2. Attempt to fetch YouTube's auto-generated or manually uploaded captions first (fast, free, no API key needed). |
|
150
|
|
|
151
|
3. If no captions are available, download the audio track and transcribe via Whisper. |
|
152
|
4. Extract video metadata (title, channel, description, publish date, duration). |
|
153
|
|
|
154
|
5. Proceed to content generation. |
|
155
|
|
|
156
|
Read `youtube-processing.md` for the full YouTube extraction pipeline. |
|
157
|
|
|
158
|
### Mode 4: Batch Mode |
|
159
|
|
|
160
|
Process the last N episodes from an RSS feed automatically. |
|
161
|
|
|
162
|
1. Parse the RSS feed. |
|
163
|
2. Sort episodes by publish date (newest first). |
|
164
|
|
|
165
|
3. Download and transcribe each episode. |
|
166
|
4. Run the content generation pipeline for each episode. |
|
167
|
|
|
168
|
5. Organize output by episode (one directory per episode). |
|
169
|
|
|
170
|
Read `rss-and-batch.md` for batch processing details. |
|
171
|
|
|
172
|
## Content Generation Pipeline |
|
173
|
|
|
174
|
Once you have a transcript and episode metadata, follow this pipeline. The order matters — each step builds on the previous one. |
|
175
|
|
|
176
|
### Step 1: Extract Content Atoms |
|
177
|
|
|
178
|
This replaces a simple summary. Analyze the full transcript and extract **7 types of content atoms** — these are the raw material for everything downstream. |
|
179
|
|
|
180
|
| Atom Type | What to Look For | Example | |
|
181
|
|
|
182
|
|-----------|-----------------|---------| |
|
183
|
|
|
184
|
| **Narrative Arcs** | Stories with setup, tension, resolution | "When we almost lost the company in 2019..." | |
|
185
|
|
|
186
|
| **Quotable Moments** | Punchy, memorable one-liners | "Culture eats strategy for breakfast, but execution eats both" | |
|
187
|
|
|
188
|
| **Controversial Takes** | Opinions that challenge mainstream thinking | "I think remote work is actually worse for junior employees" | |
|
189
|
|
|
190
|
| **Data Points** | Statistics, numbers, research findings | "We saw a 340% increase after switching to..." | |
|
191
|
|
|
192
|
| **Frameworks** | Mental models, step-by-step processes | "The 3-bucket prioritization system..." | |
|
193
|
|
|
194
|
| **Stories** | Personal anecdotes, case studies, examples | "Our first customer was actually..." | |
|
195
|
|
|
196
|
| **Predictions** | Forward-looking claims about trends | "In 5 years, every company will need to..." | |
|
197
|
|
|
198
|
For each atom, capture: the verbatim text, timestamps (start/end), speaker, atom type, and a one-line hook that could open a social post. |
|
199
|
|
|
200
|
**Timestamp requirement:** Every atom MUST include `start_timestamp`and`end_timestamp` in seconds. Match the verbatim text against the transcript segments to find the correct timestamps. Without timestamps, video clip selection is impossible. |
|
201
|
|
|
202
|
**Long transcript handling:** If the transcript exceeds 50,000 characters, split into overlapping chunks (5,000 char overlap) and extract atoms from each chunk separately. Deduplicate by comparing verbatim_text similarity (>80% overlap = duplicate). Merge and re-rank. |
|
203
|
|
|
204
|
Read `content-atoms.md` for the full extraction prompt and output format. |
|
205
|
|
|
206
|
### Step 2: Score and Assign Content Atoms |
|
207
|
|
|
208
|
Score each atom on three dimensions to prioritize the strongest content: |
|
209
|
|
|
210
|
| Dimension | Weight | What It Measures | |
|
211
|
|
|
212
|
|-----------|--------|-----------------| |
|
213
|
|
|
214
|
| **Novelty** | 40% | Is this new, surprising, or counter-intuitive? | |
|
215
|
|
|
216
|
| **Controversy** | 30% | Will people disagree, debate, or have strong reactions? | |
|
217
|
|
|
218
|
| **Utility** | 30% | Can someone apply this immediately? | |
|
219
|
|
|
220
|
#### Score thresholds |
|
221
|
|
|
222
|
- **80-100**: Priority publish — lead with these |
|
223
|
- **60-79**: Strong fill — solid content for the calendar |
|
224
|
|
|
225
|
- **40-59**: Gap filler — use when scheduling needs variety |
|
226
|
- **Below 40**: Cut — not worth publishing |
|
227
|
|
|
228
|
**Atom-to-platform assignment (anti-repetition rule):** After scoring, assign each atom to AT MOST 2 content pieces. No single quote should appear verbatim in more than 2 outputs. Use this mapping to assign the highest-scoring atoms first: |
|
229
|
|
|
230
|
| Platform | Best Atom Types | Lead Atom | |
|
231
|
|
|
232
|
|----------|----------------|-----------| |
|
233
|
|
|
234
|
| Twitter thread | Controversial takes, data points | Assign atom \#1 by score | |
|
235
|
|
|
236
|
| LinkedIn post | Frameworks, data points | Assign a DIFFERENT atom than Twitter | |
|
237
|
|
|
238
|
| Blog post | Narrative arcs, stories | Assign a DIFFERENT atom than Twitter/LinkedIn | |
|
239
|
|
|
240
|
| Newsletter | Top 2-3 by score, any type | Can reuse 1 atom from blog | |
|
241
|
|
|
242
|
| Quote cards | Quotable moments | Use atoms NOT already assigned to other platforms | |
|
243
|
|
|
244
|
| Video clips | Controversial takes, stories, data points | Based on timestamp availability | |
|
245
|
|
|
246
|
This ensures each platform leads with a unique angle rather than repeating the same quote everywhere. |
|
247
|
|
|
248
|
### Step 3: Show Notes |
|
249
|
|
|
250
|
Generate structured show notes with: |
|
251
|
|
|
252
|
- Episode title and one-paragraph description |
|
253
|
- Timestamped topic outline (if timestamps available) |
|
254
|
|
|
255
|
- Key takeaways as bullet points (5-7) |
|
256
|
- **ALL links mentioned** in the episode — parse from YouTube description, transcript mentions of URLs, and episode metadata. Include every reference link, not just a subset |
|
257
|
|
|
258
|
- Guest bio (if interview format — pull from transcript intro or episode metadata) |
|
259
|
|
|
260
|
### Step 4: Blog Post |
|
261
|
|
|
262
|
Write an 800-1200 word article that: |
|
263
|
|
|
264
|
- Opens with the atom assigned to the blog (from Step 2) as the hook (not "In this episode...") |
|
265
|
- Expands on 2-3 top atoms with context and examples from the conversation |
|
266
|
|
|
267
|
- Includes 1-2 pull quotes from the transcript |
|
268
|
- Ends with a CTA to listen to the full episode |
|
269
|
|
|
270
|
- Uses subheadings for scannability |
|
271
|
|
|
272
|
**Hard limit: 1,200 words maximum.** Count words before saving. If over, cut the weakest section. |
|
273
|
|
|
274
|
### Step 5: Social Media Posts |
|
275
|
|
|
276
|
Generate platform-specific posts. Each platform MUST lead with a DIFFERENT atom (from Step 2 assignment): |
|
277
|
|
|
278
|
#### Twitter/X Thread (5-8 tweets) |
|
279
|
|
|
280
|
- Tweet 1: Hook — the atom assigned to Twitter (controversial take or novel insight) |
|
281
|
- Tweets 2-6: One atom per tweet, using the speakers' own words when possible |
|
282
|
|
|
283
|
- Final tweet: CTA with episode link |
|
284
|
- **HARD LIMIT: Every tweet MUST be under 280 characters including spaces and URLs.** Count characters before saving. Split long quotes across two tweets if needed. |
|
285
|
|
|
286
|
##### LinkedIn Post (1 long-form) |
|
287
|
|
|
288
|
- Open with the atom assigned to LinkedIn (framework or data point — NOT the same as Twitter hook) |
|
289
|
- Share 2-3 insights with brief context |
|
290
|
|
|
291
|
- Professional tone, first-person perspective |
|
292
|
- End with a question to drive engagement + episode link |
|
293
|
|
|
294
|
- 1,200-1,500 characters |
|
295
|
|
|
296
|
###### Instagram Caption |
|
297
|
|
|
298
|
- Conversational, accessible tone |
|
299
|
- Lead with the "why should I care" angle |
|
300
|
|
|
301
|
- Include 3-5 relevant hashtags |
|
302
|
- End with CTA to listen (link in bio) |
|
303
|
|
|
304
|
- 300-500 characters |
|
305
|
|
|
306
|
**Multi-topic episodes:** If the episode covers 3+ distinct topics, also generate 1 standalone tweet per topic (not threads — single tweets) for drip-posting throughout the week. Include these in the calendar. |
|
307
|
|
|
308
|
Read `content-templates.md` for detailed templates and examples for each platform. |
|
309
|
|
|
310
|
### Step 6: Email Newsletter |
|
311
|
|
|
312
|
Write newsletter copy that: |
|
313
|
|
|
314
|
- Subject line: curiosity-driven, under 50 characters |
|
315
|
- Preview text: complements the subject line, under 90 characters |
|
316
|
|
|
317
|
- Body: 200-300 words built around the top 2-3 atoms |
|
318
|
- Include 1 pull quote (can reuse 1 from blog) |
|
319
|
|
|
320
|
- Clear CTA button text (e.g., "Listen Now", "Hear the Full Conversation") |
|
321
|
|
|
322
|
### Step 7: Pull Quotes |
|
323
|
|
|
324
|
Select 5-8 standalone quotes that: |
|
325
|
|
|
326
|
- Make sense without context |
|
327
|
- Are surprising, insightful, or emotionally resonant |
|
328
|
|
|
329
|
- Are under 150 characters each (for shareable graphics) |
|
330
|
- Include speaker attribution |
|
331
|
|
|
332
|
- Have timestamps (for linking to the exact moment) |
|
333
|
|
|
334
|
### Step 8: Quote Cards |
|
335
|
|
|
336
|
Generate visual quote card images for the top 3-5 pull quotes. |
|
337
|
|
|
338
|
**Requires Tier 2 dependencies** (Pillow + system fonts). Run `scripts/generate_quote_cards.py`: |
|
339
|
|
|
340
|
```python |
|
341
|
|
|
342
|
from generate_quote_cards import batch_generate |
|
343
|
|
|
344
|
quotes = [ |
|
345
|
|
|
346
|
{"text": "Quote text here", "speaker_name": "Speaker", "speaker_title": "Title"}, |
|
347
|
|
|
348
|
... |
|
349
|
|
|
350
|
] |
|
351
|
|
|
352
|
paths = batch_generate(quotes, output_dir="quote-cards/", podcast_name="Podcast Name") |
|
353
|
|
|
354
|
``` |
|
355
|
|
|
356
|
Each quote produces 3 files: |
|
357
|
|
|
358
|
- `quote-N-instagram-1080x1080.png` (feed posts, carousels) |
|
359
|
- `quote-N-twitter-1200x675.png` (tweets, LinkedIn) |
|
360
|
|
|
361
|
- `quote-N-stories-1080x1920.png` (Instagram Stories, TikTok) |
|
362
|
|
|
363
|
Read `quote-cards.md` for design specs, color customization, and AI background generation. |
|
364
|
|
|
365
|
### Step 9: Short-Form Video Clips (Opus Clips Style) |
|
366
|
|
|
367
|
Automatically identify and extract the best 30-90 second moments from the episode for YouTube Shorts, TikTok, and Instagram Reels. |
|
368
|
|
|
369
|
**Requires Tier 3 dependencies** (ffmpeg + yt-dlp). For animated captions, also requires Tier 4 (OpenAI Whisper). |
|
370
|
|
|
371
|
The pipeline: |
|
372
|
|
|
373
|
1. Use content atoms (especially controversial takes, stories, and data points) to identify clip-worthy moments. **Atoms MUST have start/end timestamps** — atoms without timestamps are skipped. |
|
374
|
2. Download the source video if from YouTube (`scripts/generate_video_clips.download_youtube_video`). |
|
375
|
|
|
376
|
3. Trim clips at the identified timestamps with fade in/out. |
|
377
|
4. Reformat to 9:16 vertical (center-crop or letterbox). |
|
378
|
|
|
379
|
5. **If Tier 4 available:** Add animated word-by-word captions (Opus Clips style) via Whisper re-transcription of each clip. |
|
380
|
6. **If Tier 4 unavailable:** Add static segment-level captions using transcript data (still readable, just not word-highlighted). |
|
381
|
|
|
382
|
7. Burn in a progress bar and speaker name lower third. |
|
383
|
|
|
384
|
Target: 3-5 clips per episode, each scored by viral potential. Generate a hook title for each clip. |
|
385
|
|
|
386
|
```python |
|
387
|
|
|
388
|
from generate_video_clips import extract_clips |
|
389
|
|
|
390
|
results = extract_clips( |
|
391
|
|
|
392
|
video_path="episode.mp4", |
|
393
|
|
|
394
|
atoms=scored_atoms, \# must have start_timestamp/end_timestamp |
|
395
|
|
|
396
|
output_dir="clips/", |
|
397
|
|
|
398
|
speaker_name="Speaker Name", |
|
399
|
|
|
400
|
max_clips=5, |
|
401
|
|
|
402
|
) |
|
403
|
|
|
404
|
``` |
|
405
|
|
|
406
|
Read `video-clips.md` for the full video clip extraction and captioning pipeline. |
|
407
|
|
|
408
|
### Step 10: Content Calendar |
|
409
|
|
|
410
|
Auto-schedule all generated content across platforms using best-practice timing: |
|
411
|
|
|
412
|
| Platform | Best Times | Frequency | Content Types | |
|
413
|
|
|
414
|
|----------|-----------|-----------|--------------| |
|
415
|
|
|
416
|
| Twitter/X | 9-11am, 1-3pm weekdays | 1-2 posts/day | Threads, quote cards, clip links, topic tweets | |
|
417
|
|
|
418
|
| LinkedIn | Tue-Thu 8-10am | 2-3 posts/week | Articles, quote cards, insights | |
|
419
|
|
|
420
|
| Instagram | Mon/Wed/Fri 11am-1pm | 3-5 posts/week | Quote cards, carousels, Reels | |
|
421
|
|
|
422
|
| TikTok | Tue/Thu/Sat 7-9pm | 3-5 posts/week | Short-form clips | |
|
423
|
|
|
424
|
| YouTube | Sat/Sun 9am-12pm | 1-2 Shorts/week | Short-form clips | |
|
425
|
|
|
426
|
| Newsletter | Tue/Thu morning | 1/week | Newsletter copy | |
|
427
|
|
|
428
|
| Blog | Any weekday | 1/episode | Blog post | |
|
429
|
|
|
430
|
Read `content-calendar.md` for scheduling logic, content mix optimization, and calendar output format. |
|
431
|
|
|
432
|
## Quality Checklist |
|
433
|
|
|
434
|
Run this checklist before delivering content to the user: |
|
435
|
|
|
436
|
- [ ] **No quote recycling**: No verbatim quote appears in more than 2 content pieces |
|
437
|
- [ ] **Platform leads are unique**: Twitter, LinkedIn, and blog each open with a different atom/angle |
|
438
|
|
|
439
|
- [ ] **Twitter character limits**: Every tweet is under 280 characters (count them) |
|
440
|
- [ ] **Blog word count**: Blog post is 800-1,200 words (count them) |
|
441
|
|
|
442
|
- [ ] **LinkedIn character count**: Post is 1,200-1,500 characters |
|
443
|
- [ ] **Instagram length**: Caption is 300-500 characters before hashtags |
|
444
|
|
|
445
|
- [ ] **Newsletter subject line**: Under 50 characters |
|
446
|
- [ ] **All links captured**: Show notes include every link from the YouTube description/episode metadata |
|
447
|
|
|
448
|
- [ ] **Timestamps on atoms**: Every atom has start*timestamp and end*timestamp |
|
449
|
- [ ] **Pull quotes are short**: Each under 150 characters |
|
450
|
|
|
451
|
- [ ] **Content pieces generated individually**: Each piece was generated in its own LLM call (not batched with ---SEPARATOR--- patterns that cause content to bleed between files) |
|
452
|
|
|
453
|
## Generation Best Practices |
|
454
|
|
|
455
|
**Generate each content piece in a separate LLM call.** Do NOT batch multiple outputs (e.g., "generate blog + social + newsletter separated by ---SEPARATOR---"). This causes: |
|
456
|
|
|
457
|
- Content bleeding between files (e.g., Twitter thread ends up in LinkedIn file) |
|
458
|
- Quality degradation on later pieces as the model runs out of attention |
|
459
|
|
|
460
|
- Inconsistent formatting |
|
461
|
|
|
462
|
Instead: generate atoms first, then generate each content piece independently with the atoms + relevant assignment as context. This produces better quality and avoids file-splitting bugs. |
|
463
|
|
|
464
|
**For long transcripts (>50K chars):** Split into 2-3 chunks with 5K char overlap. Extract atoms from each chunk. Merge and deduplicate before scoring. |
|
465
|
|
|
466
|
## Output Structure |
|
467
|
|
|
468
|
```text |
|
469
|
|
|
470
|
podcast-content/ |
|
471
|
|
|
472
|
├── episode-title/ |
|
473
|
|
|
474
|
│ ├── atoms.json \# Extracted + scored content atoms with timestamps |
|
475
|
|
|
476
|
│ ├── show-notes.md |
|
477
|
|
|
478
|
│ ├── blog-post.md |
|
479
|
|
|
480
|
│ ├── social/ |
|
481
|
|
|
482
|
│ │ ├── twitter-thread.md |
|
483
|
|
|
484
|
│ │ ├── linkedin-post.md |
|
485
|
|
|
486
|
│ │ ├── instagram-caption.md |
|
487
|
|
|
488
|
│ │ └── topic-tweets.md \# (multi-topic episodes only) |
|
489
|
|
|
490
|
│ ├── newsletter.md |
|
491
|
|
|
492
|
│ ├── pull-quotes.md |
|
493
|
|
|
494
|
│ ├── quote-cards/ \# (Tier 2) |
|
495
|
|
|
496
|
│ │ ├── quote-1-instagram-1080x1080.png |
|
497
|
|
|
498
|
│ │ ├── quote-1-twitter-1200x675.png |
|
499
|
|
|
500
|
│ │ ├── quote-1-stories-1080x1920.png |
|
501
|
|
|
502
|
│ │ └── ... |
|
503
|
|
|
504
|
│ ├── clips/ \# (Tier 3+) |
|
505
|
|
|
506
|
│ │ ├── clip-1-vertical.mp4 |
|
507
|
|
|
508
|
│ │ ├── clip-2-vertical.mp4 |
|
509
|
|
|
510
|
│ │ └── clips-metadata.json |
|
511
|
|
|
512
|
│ ├── audiograms/ \# (audio-only or requested) |
|
513
|
|
|
514
|
│ │ ├── clip-1.mp4 |
|
515
|
|
|
516
|
│ │ └── ... |
|
517
|
|
|
518
|
│ └── calendar.json \# Scheduled content calendar |
|
519
|
|
|
520
|
``` |
|
521
|
|
|
522
|
## Tone & Voice Calibration |
|
523
|
|
|
524
|
Before generating content, ask the user (or infer from their existing content): |
|
525
|
|
|
526
|
- **Brand voice** — formal, conversational, edgy, academic, playful? |
|
527
|
- **Target audience** — who reads/follows their content? |
|
528
|
|
|
529
|
- **Platform priorities** — which platforms matter most? |
|
530
|
- **Minimum viral score** — threshold for content to make the cut (default: 40) |
|
531
|
|
|
532
|
If not specified, default to conversational and professional. Mirror the energy of the podcast itself — if the hosts are casual and funny, the marketing content should reflect that. |
|
533
|
|
|
534
|
## Composability |
|
535
|
|
|
536
|
This skill works alongside the `podcast-generator` skill. A typical workflow might be: |
|
537
|
|
|
538
|
1. Use `podcast-generator` to create an episode from research |
|
539
|
2. Use `podcast-marketing` to repurpose that episode into content |
|
540
|
|
|
541
|
## References |
|
542
|
|
|
543
|
- `content-atoms.md` — Content atom extraction prompt, the 7 atom types, output format, and scoring rubric. Read first — atoms are the foundation for all content. |
|
544
|
- `content-templates.md` — Detailed templates and examples for each content type (social posts, blog, newsletter, show notes, pull quotes). Read when generating specific content types. |
|
545
|
- `quote-cards.md` — Quote card design specifications, image generation, and multi-format output. Read when generating quote cards. |
|
546
|
- `video-clips.md` — Opus Clips-style video clip extraction, vertical reformatting, animated captions, and progress bar overlay. Read when generating short-form video clips. |
|
547
|
- `content-calendar.md` — Platform-specific scheduling rules, content mix optimization, and calendar JSON output format. Read when generating a content calendar. |
|
548
|
- `audio-processing.md` — Whisper transcription setup, audiogram generation pipeline, and audio clip extraction. Read when handling audio input or generating audiograms. |
|
549
|
- `rss-and-batch.md` — RSS feed parsing, episode downloading, and batch processing logic. Read when the input is an RSS feed or batch mode is requested. |
|
550
|
- `youtube-processing.md` — YouTube URL handling, caption extraction, audio download, and metadata retrieval. Read when the input is a YouTube link. |