Get watched: 10 AI skills for short-form video
series-builder
turn one video into a series people come back for
How the two work together
Claude thinks it through. Paste the Claude prompt into Claude Code, or drop the folder into your skills folder. Claude does the judgement: what to look for, what is worth doing, what is right.
Codex gets it done. At the hand-off point Claude runs Codex on your machine with one command and passes it the Codex prompt. Codex does the mechanical part and hands the result back. Claude checks it before you see it.
No API key to set up: Claude calls the Codex you already have installed. If Codex is not installed, Claude does that half itself and tells you.
Prompt for Claude
---
name: series-builder
description: Turns one video that worked into a named repeatable series, with the next three episodes written and 30 more titled, batched and dated. Use when posting feels random, or right after a video outperforms.
---
# Turn one video into a series people come back for
A named format, its rules, episodes 2 to 4 written in full, and 30 more scheduled.
## Claude does
1. Ask for five things: the video that worked (script or transcript), who it is for, what the viewer should do next, filming days per week, and the platforms. Refuse to guess the audience or the offer. Ask once, then wait.
2. Isolate the one or two load-bearing drivers (topic, format, person, surprise, sound). Counterfactual test: remove it, would the video still have worked? Only a survivor becomes the series.
3. Name 3 to 5 pillars, tagged Hub, Hero or Help. Four-part test: the audience searches or argues about it, you can speak on it credibly, it connects to what you sell, you can film it weekly. A pillar that cannot yield 10 concrete titles in one sitting fails and is replaced.
4. Write the format spine, six fixed parts: the recurring hook stem (same shape every episode, inside the first second, never past three), the promise, the constant (a number, a place, a rule), the variable that changes each episode, the payoff shape, and the sign-off naming the series and teasing the next number.
5. Write episodes 2, 3 and 4 in full at 50 to 60 seconds (130 to 150 spoken words), foreshadowing the conflict straight after the hook, strongest emotion last, because the final beat decides whether they return.
6. Fix the cadence: Hub in one slot on one weekday, Hero monthly, Help on the remaining shoot days.
## Then Codex does
The bulk expansion: one approved spine into 30 distinct episodes, grouped into shoot batches by location, framing and outfit, laid on a dated calendar. Repetition and date arithmetic, no judgement.
```
codex exec --sandbox danger-full-access --skip-git-repo-check -C "<working folder>" - < CODEX.md
```
Claude fills in first: {SERIES_SPINE} (the six fixed parts), {PILLARS} (3 to 5, tagged Hub, Hero or Help), {SHOOT_DAYS} (days per week, posting slot and weekday), {PLATFORMS}, {START_DATE}.
## Claude checks
1. Every episode names a supplied pillar. Orphans rejected.
2. The 30 variables are different subjects, not one idea reworded. Under 25 distinct, rerun.
3. The stem repeats, the hook content never does. No duplicate opening lines.
4. No view count, sound or competitor claim the user did not supply. Video repos quote benchmarks with no citation; house numbers are never this account's data.
5. Every date is today or later, checked by subtracting from today, and lands on a declared posting day.
6. Batches share location, framing and outfit, so one session films a whole batch.
Reject and rerun: send failing rows back with the reason and the corrected constraint. Never hand-patch the file, or the next run repeats the fault.
## Rules
- A pillar that cannot produce 10 titles in one sitting is not a pillar.
- A series is a promise; changing the format removes the reason to return.
- Repeat the shape, never the sentence.
- Public information only.
- Never invent a fact, a number or a quote.
- Anything sent in someone's name says whose name it is.
## If Codex is not installed
Claude builds the 30 episodes, batches and calendar itself, and says so.
## Built from
- AgriciDaniel/claude-youtube, https://github.com/AgriciDaniel/claude-youtube, 350 stars on api.github.com, 5 Sep 2026; Hub, Hero, Help pillars and cadence.
- social-media-skills/skills, https://github.com/social-media-skills/skills, 66 stars, 5 Sep 2026; four-part pillar test, 10-ideas rule, counterfactual.
- sergebulaev/tiktok-skills, https://github.com/sergebulaev/tiktok-skills, 22 stars, MIT; per-day hook, length and goal, plus hook batching.
- Jenny Hoyos playbook, https://www.marketingexamined.com/blog/jenny-hoyos-short-form-video-playbook, roughly 10m views per Short, interviewed by YouTube's Shorts product lead; sub-three-second hook, final-emotion rule.
Prompt for Codex
# Turn one video into a series people come back for - Codex task
You are expanding an approved short-video series into a full backlog and calendar. You have no memory of the conversation that produced the spine. Use only what is below.
## You are given
- {SERIES_SPINE}: six fixed parts of the format - the recurring hook stem, the promise, the constant, the variable that changes each episode, the payoff shape, the sign-off.
- {PILLARS}: 3 to 5 content pillars, each tagged Hub, Hero or Help.
- {SHOOT_DAYS}: how many days a week filming happens, plus the fixed posting slot and weekday.
- {PLATFORMS}: the platforms this series posts to.
- {START_DATE}: the first posting date. Never schedule anything before it.
## Produce
Write three files into the working folder.
1. `series-backlog.csv`, exactly 30 rows plus a header, columns: `episode_no,pillar,pillar_type,variable,working_title,hook_line,payoff,sign_off,batch_id,shoot_location,framing,outfit,target_seconds,platform`.
- `hook_line` uses the spine's hook stem shape every time with different content every time. No two hook lines may share more than four consecutive words.
- `variable` must be 30 distinct subjects. Rewording one subject is a failure.
- `target_seconds` is between 50 and 60.
- Every `pillar` value comes from {PILLARS} verbatim.
2. `shoot-batches.md`: the same 30 episodes grouped into batches. A batch shares one location, one framing and one outfit, so it can be filmed in a single session. Give each batch an id matching `batch_id`, list its episode numbers, and state what needs setting up once.
3. `calendar.csv`, columns: `date,weekday,episode_no,pillar_type,platform,slot`. Dates start at {START_DATE}, fall only on declared posting weekdays, keep the Hub episodes in the fixed slot, place one Hero per calendar month, and fill the rest with Help. Cover all 30 episodes.
## Rules
- Do not invent view counts, watch-time figures, trending sounds, competitor names or any statistic. If a field would need one, leave it blank.
- Do not change the spine. The hook stem, constant, payoff shape and sign-off are fixed inputs.
- No date before {START_DATE}, and no date on a weekday not declared in {SHOOT_DAYS}.
- Plain text and CSV only. No em dash characters. British English.
- Do not research the web, and do not add pillars beyond {PILLARS}.
## Return
Print, in this order:
1. The three file paths written.
2. `DISTINCT VARIABLES: <n>` - the count of genuinely different subjects across the 30 rows.
3. `DUPLICATE HOOK PAIRS: <n>` - pairs sharing five or more consecutive words.
4. `PILLARS USED: <name>=<count>` for each pillar.
5. `DATE RANGE: <first> to <last>` and `OFF-SLOT DATES: <n>`.
6. `BATCHES: <n>` and the episode count in each.
7. Any field left blank because it would have required an invented number.
Built from the best public work on this
Sources
- **AgriciDaniel/claude-youtube** - https://github.com/AgriciDaniel/claude-youtube - 350 stargazers read from api.github.com on 5 Sep 2026 (created 5 Mar 2026, pushed 10 Apr 2026). A three-layer skill: orchestrator, 14 sub-skills and 9 reference guides, installable as a plugin. Its `/youtube strategy` and `/youtube calendar` commands supply the Hub, Hero and Help pillar split and the cadence model used here. Caveat worth repeating: its benchmark percentages are stated in-repo with no citation, so they are house numbers, not verified data, and are deliberately not carried into this skill.
- **social-media-skills/skills** - https://github.com/social-media-skills/skills - 66 stargazers from api.github.com on 5 Sep 2026 (created 16 Jul 2026, pushed 3 Sep 2026, actively maintained), 106 skill directories confirmed by listing the contents endpoint. Its `content-pillars` skill supplies the four-part pillar test and the rule that a pillar failing to yield 10 ideas is not a pillar; its `viral-reverse-engineering` skill supplies the counterfactual test used to isolate what actually made the first video work.
- **sergebulaev/tiktok-skills** - https://github.com/sergebulaev/tiktok-skills - 22 stargazers from api.github.com on 5 Sep 2026 (created 2 Jul 2026, pushed 3 Sep 2026), MIT licensed. Low stars but unusually prescriptive. Its `content-planner` supplies the weekly plan with a hook, a length and a goal attached to each day, plus hook batching, which is where the shoot-batch grouping comes from.
- **Jenny Hoyos short-form playbook (Marketing Examined)** - https://www.marketingexamined.com/blog/jenny-hoyos-short-form-video-playbook - roughly 10 million views per Short, 3 million subscribers, and interviewed on YouTube's own blog by the Shorts product lead. Supplies the hook window of no more than three seconds, the instruction to foreshadow the conflict immediately after the hook, and the rule that the viewer judges the whole video by the intensity of the final emotion, which is why the sign-off is a fixed part of the spine.
Best public prompt we found for this job
https://github.com/social-media-skills/skills - the `content-pillars` skill. It is the only public source that states a pass or fail condition for a pillar rather than describing pillars in the abstract: three to five pillars, each tested on four criteria, and 10 or more concrete ideas produced on the spot or the pillar does not qualify. That single failure condition is what stops a series plan drifting into a list of themes nobody can film.
Want this running in your business?
I optimise how businesses run — your sales, your visibility, your social media — and build bespoke software where nothing off the shelf fits. The first conversation is free. Work starts from £150 a day.
Foxera