| name | repurpose-content |
| description | Turn one long video (or podcast/audio) into many pieces — vertical Shorts/Reels clips, quote images, captions, and a thread. Use when the user says "repurpose this video", "turn my podcast into clips", "make shorts from this", "slice this up for social". |
| allowed-tools | Bash Read Write Edit Glob Grep |
Repurpose Content
One long asset → a week of content. Pipeline: transcribe → find the moments →
cut clips → derive posts → package per platform.
Commands used (argument shapes in workflows/WORKFLOWS.md § Publish & Repurpose):
transcribeVideo (needs OPENROUTER_API_KEY) · extractClip · packageContent
(both local/free) · generateSingleImage · generateCaption.
STEP 1: Intake
- Confirm the source file exists; note duration:
ffprobe -v error -show_entries format=duration -of csv=p=0 "<file>"
- Ask (one batch): which platforms? how many clips (default 3-5)? quote
images / caption thread wanted too?
- Run
/content-preflight if a project exists (brand tone for captions);
minimal mode is fine for a quick job.
STEP 2: Transcribe
node workflows/cli.cjs transcribeVideo '{"mediaPath":"<file>","outputPath":"projects/{name}/output-contents/{piece}/transcript.txt"}'
- No OPENROUTER_API_KEY → tell the user transcription needs one
(https://openrouter.ai/keys), or ask them to paste a transcript/subtitles
file instead. Don't guess the content.
- Cost is roughly $0.006/minute — state it for long files (>30 min) before running.
STEP 3: Find the moments
Read the transcript and select clip candidates — you do this, no API call
needed. Look for: strong hooks/claims, complete self-contained thoughts
(15-60s), emotional peaks, concrete tips, quotable one-liners.
The transcript has no timestamps. Estimate each moment's position from its
word offset (spoken English ≈ 150 words/min), then cut generously (start
~5s early) — you'll verify in the next step.
Present candidates and let the user pick:
Found 6 clip candidates:
1. ~02:10 "The roast date matters more than the price tag" (hook, ~25s)
2. ~08:45 The 3-signs-of-stale-beans tip (self-contained, ~40s)
...
Which ones? (numbers / all / adjust)
STEP 4: Cut and verify each clip
node workflows/cli.cjs extractClip '{"videoPath":"<file>","outputPath":".../clips/clip-01.mp4","start":"02:05","duration":35,"cropTo":"9:16"}'
cropTo: "9:16" center-crops for Shorts/Reels — warn if the speaker may be
off-center; omit cropTo to pad instead at packaging time.- Timestamps are estimates: after each cut, transcribe the CLIP (cheap — it's
short) or ask the user to check it, and adjust
start/duration if the cut
clips a sentence. Iterate until clean.
STEP 5: Derive the other formats (as requested)
- Quote images: pull the strongest lines →
/generate-image with brand
style (or a clean typographic card), 4:5.
- Captions per clip:
generateCaption (or /write-copy) from the clip's
transcript segment — hook first line, platform hashtags.
- Thread/text post: distill the transcript's core argument into a
numbered thread (twitter/linkedin), respecting char limits.
STEP 6: Package everything
For each clip/image: /package-content into
output-contents/{piece}/publish/{clip-id}/. Then report:
✅ REPURPOSED — podcast-ep12 (42 min source)
3 clips (9:16) → publish/clip-01..03/ (tiktok, youtube, instagram)
2 quote images → publish/quote-01..02/ (instagram, linkedin)
1 thread → publish/thread.txt (twitter)
Spend: $0.31 (transcription $0.25, images $0.06 — clips/packaging free)
recordCost any paid steps against the project budget.
Pipeline-first + audit trail (mandatory)
- Author
<content-id>.pipeline.json IN the content folder BEFORE running the
chain (nodes = CLI commands — transcribe → extract → package —
{{node.data.field}} refs wire outputs to inputs), then
node workflows/cli.cjs runPipeline @<file>. See workflows/pipelines/README.md.
- The content folder gets a
prompts.txt: quote-image prompts and caption/thread
copy; update it on every retry with a one-line RESULT note.
- Log each paid generation via
createGenerationManifest / addManifestEntry
(AGENT-GUIDE Step 5).
