| name | marketing-ideate |
| description | Use when deciding WHAT short video to make next for a channel — generate falsifiable viral bets (idea + hook + assumption + goal + theme). The AGENT does the ideation (web-search current trends + recall the channel's past winners), then persists each bet to the backlog. One lego-block of the ideate→deploy→measure→learn growth loop. Pairs with marketing-guru, which picks the next bet from the backlog and reads/writes channel state.
|
marketing-ideate — generate the next bet(s)
Produce N falsifiable bets. Each bet is idea · hook (literal 0–3s scroll-stopper) ·
assumption (WHY it should go viral — must be falsifiable) · goal (measurable, e.g.
">P75 velocity") · theme · tags. You — the agent — do the thinking; the CLI only persists.
Do this
- Read state — phase + learned direction:
studio marketing journal --channel <name>
- Gather live signal (don't bet from stale memory): use
WebSearch for trending
formats/hooks/narratives/news pegs in the niche + what comparable channels are blowing up
with this week.
- Recall what worked here:
studio marketing recall "<niche / current direction / candidate theme>" --channel <name>
- Reason the bets yourself:
- Cold-start (<10 deployed): maximize DIVERSITY — vary theme, hook style, emotion to map
what this audience rewards.
- Optimizing (≥10): lean into the recalled winning patterns, but make ≥1 exploration
bet into adjacent territory (the 40% — see the backlog section in marketing-guru).
- Persist each (no LLM, just I/O):
studio marketing add "<idea>" --hook "<hook>" --assumption "<why>" --goal "<target>" \
--theme "<theme>" --tags "a,b,c" --channel <name>
Fallback (scripted, non-agent)
studio marketing ideate --provider <llm> --signals <file> --niche "<niche>" --n 3 runs the
built-in LLM generator and writes the bets for you. Use only when you want a quick draft
without agent reasoning.
Writes planned entries → the backlog; reads strategy (long-term) + episodic recall.
Memory model: docs/50-marketing/memory.md.
