Install all 5 with one prompt
Here is the one prompt I promised. Copy it, paste it into Claude Code, and hit enter. It writes all 5 skills, downloads the free video editor that /edit runs on, checks whether Blotato is connected (and walks you through it if not), then tells you exactly what to type to try each one.
You are installing 5 Claude skills on my computer that run my social media: /schedule, /analyze-competitors, /edit, /trial-reels and /comment-to-dm. Install all 5, connect what they need, check everything works, then report back.
I am not technical. Before you run anything, tell me in one plain sentence what it does. Do the steps in order and confirm each one worked before you move on.
STEP 1: Where am I?
- Check whether you can run terminal commands and write to ~/.claude/skills/. In Claude Code (the terminal app, or the Code tab in the Claude desktop app) you can.
- If you CAN'T (for example you're in Cowork, claude.ai or the phone app): don't try to fake it. Instead, save each skill below as its own folder (schedule/SKILL.md, analyze-competitors/SKILL.md and so on), zip each folder, and give me the files. Then tell me to upload them one at a time under Settings, then Skills (or Customize, then Skills), and toggle each one on. Tell me /edit only works in Claude Code on a Mac, so skip it there. Then go to STEP 4.
STEP 2: Write the 5 skill files
Below this prompt are 5 files. Each one starts with a line like
===== BEGIN FILE: ~/.claude/skills/schedule/SKILL.md =====
and ends with
===== END FILE =====
For each one:
- Make the folder if it doesn't exist (mkdir -p ~/.claude/skills/<name>).
- If a SKILL.md is already there, don't overwrite it silently. Rename the old one to SKILL.md.bak, tell me, then write the new one.
- Write everything between the BEGIN and END lines into that file, exactly as it is. Don't include the BEGIN and END lines themselves. Don't shorten, summarize or "improve" anything.
- Use your file-writing tool, not echo or a heredoc in the shell, so backticks, quotes and dollar signs don't get mangled.
- After writing all 5, read the first 5 lines of each file back and confirm each one starts with --- and has the right name: line.
STEP 3: Get the video editor for /edit (Mac only)
/edit runs on a free open-source editor: https://github.com/TakeOff-Team/claude-video-editor-blotato
- Run uname -s. If it isn't Darwin, I'm not on a Mac. Skip this step and tell me /edit needs a Mac.
- Check if I already have it in ~/claude-video-editor-blotato or ~/Documents/claude-video-editor-blotato. If I do, skip the clone.
- If not, clone it: git clone https://github.com/TakeOff-Team/claude-video-editor-blotato ~/claude-video-editor-blotato
(If git isn't installed, running git --version offers to install Apple's command line tools. Tell me to click Install, then try again.)
- Save the full path of the editor folder into ~/.claude/skills/edit/editor-path.txt
- Then ASK me: "Want me to run the editor's setup now? It installs ffmpeg, Whisper, Remotion and a few other free tools (about 5 GB, can take 10+ minutes). Or I can do it the first time you type /edit." If I say now, run ./setup.sh inside the editor folder and keep fixing and re-running until it says "All set." Homebrew is the one thing you can't install for me, because it needs my Mac password: if setup asks for it, give me the exact command to paste into Terminal and wait.
STEP 4: Check the Blotato connection
/schedule, /trial-reels and /comment-to-dm post to my accounts through Blotato. /analyze-competitors and /edit work without it.
- Check if you have Blotato tools (a tool called blotato_list_accounts). In Claude Code you can also run: claude mcp list
- IF BLOTATO IS CONNECTED: call blotato_list_accounts and show me which accounts are connected. Tell me if Instagram isn't one of them.
- IF IT ISN'T: tell me what I need, one step at a time, and wait for me between steps:
1. A paid Blotato plan. Claude can only talk to Blotato on a paid plan, the free trial doesn't include it. Sign up at https://blotato.com/?ref=zach and use the code ZACH30 for 30% off.
2. Connect my social accounts at https://my.blotato.com/accounts. My Instagram has to be a Business or Creator account (Instagram app: Settings, then Account type and tools).
3. Add Blotato to Claude. In Claude Code, offer to run this for me: claude mcp add --transport http --scope user blotato https://mcp.blotato.com/mcp
Then tell me to fully quit Claude Code, open it again, type /mcp, pick blotato, choose Authenticate and approve it in the browser.
In the Claude desktop app or claude.ai: Settings, then Connectors, add a custom connector with the URL https://mcp.blotato.com/mcp, then sign in.
If any of this looks out of date, Blotato's own guide wins: https://help.blotato.com/start-with-an-ai-agent/mcp/setup
4. Once I'm back, ask Claude "List my connected Blotato accounts" to check it.
- Remind me: trial reels only work once my Instagram has 1,000+ followers.
STEP 5: Optional power-up for /analyze-competitors
Tell me in one or two sentences: /analyze-competitors works for free if I paste competitors' view counts from the app, but it's faster with Apify (a scraper, about $0.002 per reel, free account at apify.com). Don't set it up unless I say yes.
STEP 6: Report back
In plain English:
- Which of the 5 skills installed, with the path of each file.
- Whether the editor is downloaded, and whether its setup has run yet.
- Whether Blotato is connected, and which accounts it sees.
- Anything that failed, with the exact error and the exact fix.
- Tell me to restart Claude Code (fully quit and reopen) so the new skills load.
- One heads up: some versions of Claude Code have their own built-in /schedule command for scheduling cloud agents. If I type /schedule and it starts talking about routines or cloud agents instead of posting my video, tell me to just say "schedule this video for tomorrow at 9am" (that picks the right skill), or offer to rename ours to /schedule-post.
- Then one thing to type to try each one:
/schedule (with a finished video: "schedule this for tomorrow at 9am on Instagram")
/analyze-competitors (tell it your niche and 3 to 10 accounts you watch)
/edit (drag in a raw video; the first time it interviews you about your style)
/trial-reels (with a finished video, the day before you'd post it)
/comment-to-dm (tell it the keyword and the link you want to send)
Here are the 5 files.
===== BEGIN FILE: ~/.claude/skills/schedule/SKILL.md =====
---
name: schedule
description: Schedules a finished video (or any post) to Instagram, TikTok, YouTube, Facebook, LinkedIn, X, Threads, Bluesky or Pinterest through Blotato, for whatever time you say. Uploads the video file, writes or takes the caption, runs a quick pre-publish check, schedules it, and hands back proof (post ids and times). Saves a copy-paste file instead if Blotato isn't connected. Triggers on "/schedule", "schedule this", "schedule this video", "post this tomorrow at 9", "put this on Instagram", "post this to TikTok and YouTube".
argument-hint: "[video file] [platforms] [time]"
allowed-tools: Read, Write, Bash, Glob, AskUserQuestion, mcp__blotato__blotato_list_accounts, mcp__blotato__blotato_create_presigned_upload_url, mcp__blotato__blotato_create_post, mcp__blotato__blotato_get_post_status, mcp__blotato__blotato_list_posts
---
# /schedule
You take a finished video and schedule it as a post, at the time the user wants, on the platforms they pick. You don't edit the video and you don't rewrite their caption unless they ask. Your job is to get it live, correctly, and prove it.
Everything runs through **Blotato** (blotato.com), which connects Claude to social accounts. The tools come from the Blotato MCP server: `blotato_list_accounts`, `blotato_create_presigned_upload_url`, `blotato_create_post`, `blotato_get_post_status`, `blotato_list_posts`. If the user's Blotato tools carry a different prefix (for example a desktop connector), use those. Same tools, same names after the prefix.
## Step 0: Is Blotato connected?
Call `blotato_list_accounts`.
- **The tool doesn't exist.** Blotato isn't connected to Claude yet. Walk them through "Connect Blotato" at the bottom of this file. Stop until it works, or offer the fallback file (Step 6).
- **403 or a subscription error.** Their plan doesn't include the MCP. Claude can only talk to Blotato on a **paid** plan (the free trial doesn't include it). Point them to https://blotato.com/?ref=zach and the code `ZACH30` (30% off). Stop.
- **401.** The sign-in went stale. In Claude Code: type `/mcp`, pick blotato, choose Authenticate. Then try again.
- **Empty list.** Connected, but no social accounts yet. Have them connect accounts at https://my.blotato.com/accounts, then call it again.
## Step 1: Get the 3 things
1. **The video.** A file path (they can drag the file into the chat), or a public URL. If it's a local file, check it with `ffprobe` if it's available: vertical 9:16 for Reels/TikTok/Shorts, has audio, under 3 minutes for Instagram Reels, under 60 seconds if they want a YouTube Short. Tell them if something's off. Don't block on it.
2. **Where.** Show their connected accounts as a short list and ask which ones. If a platform has 2+ accounts, ask which.
3. **When.** Whatever they say: "tomorrow at 9am", "Friday 6pm", "in 2 hours", "next free slot", "now". Convert it to ISO 8601 **in their timezone** (run `date +%Z` or ask if you can't tell) and say it back in plain words before scheduling: "Tomorrow, Wed Nov 12, 9:00 AM Eastern."
Then the caption. Ask: "Got a caption, or want me to write one?" If you write it: first line says what they get, then 2 to 5 short lines, then one call to action, 3 to 5 hashtags for Instagram and TikTok, none for LinkedIn, X, Threads or Bluesky. If they're running a comment-to-DM on this post, the call to action is the keyword: `Comment "KEYWORD" and I'll send it to you`. Show it, take edits.
## Step 2: Quick pre-publish check
- [ ] Instagram has a video or image attached (Instagram rejects text-only posts)
- [ ] YouTube has a title under 100 characters
- [ ] TikTok has a privacy level (default `PUBLIC_TO_EVERYONE`)
- [ ] No external link in a LinkedIn post body (it kills reach, put it in the first comment instead)
- [ ] Caption fits the platform (Bluesky 300 characters, X 280 unless they have Premium)
- [ ] Hashtags: 3 to 5 max on Instagram and TikTok
- [ ] The time is in the future and in their timezone
If something fails, say what and ask: fix it, skip that platform, or ship as-is. Don't auto-fix without asking.
## Step 3: Upload the video (local files only)
A local file has to be uploaded before Blotato can post it:
1. `blotato_create_presigned_upload_url` with the filename (for example `final.mp4`). You get back `presignedUrl` and `publicUrl`.
2. Upload the raw bytes: `curl -X PUT "<presignedUrl>" --data-binary "@<path to file>"`. Not JSON, not a form.
3. Use `publicUrl` in `mediaUrls` from now on.
A public URL skips this step.
**Claude desktop app:** if the upload fails with a network error, the app is blocking it. Settings, then Capabilities, then allow network egress and add `database.blotato.io`, then fully restart the app.
## Step 4: Schedule it
One `blotato_create_post` call per platform:
```
blotato_create_post({
accountId: "<from blotato_list_accounts>",
platform: "instagram",
text: "<caption>",
mediaUrls: ["<publicUrl>"],
scheduledTime: "2026-11-12T09:00:00-05:00" // or useNextFreeSlot: true
})
```
Leave out `scheduledTime` only when they said "now". Never publish immediately by default.
Platform fields:
- **Instagram:** `mediaType: "reel"` for video. Optional `coverImageUrl`, `firstComment`, `collaborators`. Trial reels (shown to non-followers first, needs 1,000+ followers): add `trial: { graduationStrategy: "MANUAL" }`. If they want 5 trial versions of one video, that's the `/trial-reels` skill.
- **TikTok:** `privacyLevel`, `disabledComments`, `disabledDuet`, `disabledStitch`, `isBrandedContent`, `isYourBrand`, `isAiGenerated` (all required; default false for the booleans unless they say otherwise).
- **YouTube:** `title`, `privacyStatus: "public"`, `shouldNotifySubscribers: true`.
- **Facebook:** `pageId` from the account's subaccounts. `mediaType: "reel"` for vertical video.
- **LinkedIn:** optional `pageId` for a company page. Leave it out for their personal profile.
- **Pinterest:** `boardId`.
If one platform fails, keep going with the rest. Don't roll back.
## Step 5: Hand back proof
```
Scheduled
| Platform | Account | Goes live | Post id |
|-----------|-----------|-------------------------------|---------|
| Instagram | @handle | Wed Nov 12, 9:00 AM Eastern | abc-123 |
| TikTok | @handle | Wed Nov 12, 9:00 AM Eastern | def-456 |
See or edit everything at https://my.blotato.com/scheduler
```
"Done" without a post id isn't done. If they want to double check, `blotato_list_posts` with status `scheduled` shows it in the queue.
## Step 6: Fallback (no Blotato)
If Blotato isn't connected and they don't want to set it up now, save `post-ready-to-paste.txt` in the current folder, one block per platform:
```
=== INSTAGRAM ===
Post at: Wed Nov 12, 9:00 AM Eastern
Video: /path/to/final.mp4
<caption>
=== END ===
```
Tell them the file path. They lose the scheduling, but nothing's lost.
## What NOT to do
- Don't publish right now unless they said "now" or "post it now".
- Don't guess a platform or an account. Ask.
- Don't schedule in UTC by accident. Say the time back in their timezone first.
- Don't send a local file path to `blotato_create_post`. Upload it first.
- Don't say it worked without a post id.
## Connect Blotato (one time, about 5 minutes)
Go one step at a time and wait for them between steps.
1. **A paid Blotato plan.** Sign up at https://blotato.com/?ref=zach. Claude can only talk to Blotato on a paid plan. Code `ZACH30` takes 30% off.
2. **Connect your social accounts** at https://my.blotato.com/accounts. Instagram has to be a **Business or Creator** account (Instagram app: Settings, then Account type and tools). Facebook needs a Facebook Page.
3. **Add Blotato to Claude.**
- Claude Code: run this in Terminal (or let Claude run it): `claude mcp add --transport http --scope user blotato https://mcp.blotato.com/mcp`. Then fully quit Claude Code, open it again, type `/mcp`, pick blotato, choose Authenticate, and approve it in the browser.
- Claude desktop app or claude.ai: Settings, then Connectors, add a custom connector with the URL `https://mcp.blotato.com/mcp`, then sign in.
- Blotato's own guide, if anything here looks out of date: https://help.blotato.com/start-with-an-ai-agent/mcp/setup
4. **Check it.** Ask Claude: "List my connected Blotato accounts." You should see your accounts.
===== END FILE =====
===== BEGIN FILE: ~/.claude/skills/analyze-competitors/SKILL.md =====
---
name: analyze-competitors
description: Studies the creators in your niche to find which video ideas are actually working right now, then tells you exactly which 7 videos to make next, with a hook for each. Gets your niche and 3 to 10 competitor accounts (or finds them), pulls their recent posts with real view counts, scores every post against that account's own median, keeps the outliers (3x+ median), groups them into the ideas and angles that keep winning, and writes a ranked "next 7 videos" plan to a file. Never invents numbers. Triggers on "/analyze-competitors", "analyze my competitors", "what's working in my niche", "what should my next videos be", "find me video ideas from my competitors", "study these creators".
argument-hint: "[your niche] [@competitor1 @competitor2 ...]"
allowed-tools: Read, Write, Bash, Glob, WebSearch, WebFetch, AskUserQuestion, mcp__blotato__blotato_create_source, mcp__blotato__blotato_list_top_posts
---
# /analyze-competitors
You find the video ideas that are provably working in the user's niche right now, and turn them into their next 7 videos.
The whole thing rests on one idea: **a video isn't good because it got a lot of views. It's good because it got a lot more views than that account usually gets.** A creator with 2 million followers getting 300K views is a normal day. A creator with 20K followers getting 300K views found something. So every post is scored against its own account's median, and only the outliers count.
## The rule you never break
**Never invent a view count, a follower count, a date or a link.** Every number in the report has to come from a real source: a scraper, a page you actually read, or the user pasting it. If you can't get real numbers for an account, say so plainly and ask the user to paste them. A plan built on made-up numbers is worse than no plan.
## Step 1: Get the inputs
Ask for whatever's missing, one question at a time:
1. **Your niche, in one sentence.** Who you make videos for and about what. Example: "AI tools for small business owners who aren't technical."
2. **Your platform.** Instagram Reels, TikTok or YouTube Shorts. Default Instagram.
3. **3 to 10 competitor accounts.** Handles or profile links. Creators in the same niche, ideally a mix of big ones and ones around your size (the smaller ones teach you more, because their outliers are about the idea, not the audience).
- **Don't have a list?** Offer to find them. Web search things like `top [niche] creators instagram`, `[niche] tiktok creators`, `best [niche] accounts to follow`. Only suggest accounts you can confirm exist and posted in the last 30 days. Show the list with one line each on why, and let them cut or add before you start.
4. **Your own handle** (optional). So the plan can avoid ideas they've already made.
Save the final list to `competitors.md` in the current folder so next time you can skip this step.
## Step 2: Get real posts with real view counts
For each account, you want its last 20 to 30 posts (or last 60 days), each with: link, date, view count, and the caption or what the video's about. Try these in order and tell the user which one you're using:
**A. Apify (best, optional, costs a little).** If the user has the Apify MCP connected or an `APIFY_TOKEN`:
- Instagram: the `apify/instagram-reel-scraper` actor (about $0.002 a reel at the time of writing, so 10 accounts x 30 reels is roughly $0.60). TikTok: `clockworks/tiktok-scraper`. Check the actor's current input schema before running.
- **Say the estimated cost and wait for a yes before running.** Never run a paid scrape without asking.
- Don't have it? Mention it once as an optional upgrade (free account at apify.com), then move on to B or C.
**B. Public pages.** Use web search and fetch on the profile and post pages. TikTok and YouTube pages often show view counts. Instagram mostly hides them from logged-out visitors, so expect this to fail for Instagram. Only keep a number if you actually read it on the page.
**C. The user pastes them (always works, takes about 2 minutes per account).** Ask them to open each competitor's Reels tab in the app, where every reel shows its view count on the grid, and paste a list like:
```
@handle
https://www.instagram.com/reel/ABC123/ 412K "the hook or what it's about"
https://www.instagram.com/reel/DEF456/ 38K "..."
```
Abbreviations are fine (412K, 1.2M). Mixing is fine too: scrape what you can, ask them to paste the rest.
**Blotato note.** If Blotato is connected, `blotato_create_source` can pull the transcript of a TikTok or YouTube video, which helps with Step 4 (what the video actually says). It doesn't list other accounts' posts or their views. `blotato_list_top_posts` shows the user's OWN best posts, which is useful for comparing their winners to their competitors'. Neither one replaces Step 2.
If an account ends up with fewer than 8 posts with real views, keep it in the report but flag it as thin data.
## Step 3: Score every post
For each account:
1. Find the **median** views of its posts (the middle number, not the average, so one viral video doesn't skew it).
2. Score each post: `views / median`. That's its outlier score. 4.0x means it did 4 times what that account normally does.
3. **Keep the outliers: 3x median or more.**
4. If there are fewer than 6 outliers across every account combined, lower the bar to 2x and say you did.
5. Drop anything under 2 days old (it hasn't finished getting views yet) and anything that's clearly a paid ad or a collab with a huge account (say which ones you dropped and why).
## Step 4: Find the patterns
For every outlier, write down:
- **The idea.** What the video is about in one line.
- **The angle.** Why someone would care: money, fear of missing out, a shortcut, a secret, a mistake, a before/after, a tool reveal, a "nobody's talking about this".
- **The format.** Listicle, tutorial, story, news, reaction, myth-bust, side-by-side.
- **The hook.** The first line (from the caption, the on-screen text, or the transcript if you have it).
Then group them. A pattern is real when **2 or more outliers from 2 or more different accounts** share it. One viral video from one account is a fluke, not a pattern. Rank the patterns by how many outliers back them and how big the scores are.
Also note what's NOT working: topics every competitor keeps posting that never break their median. Those are the ideas to skip.
## Step 5: The next 7 videos
Turn the patterns into 7 specific videos for the user's niche. Not copies of the competitors' videos: the same winning idea, made for the user's audience, from the user's angle.
For each one:
- **The video:** one line on what it is
- **Why it should work:** the pattern, with the real outliers behind it (account, link, views, score)
- **Format and rough length**
- **Spoken hook:** the first sentence, word for word
- **On-screen text hook:** 3 to 7 words for the first frame
- **What to show:** 1 or 2 lines
Order them: the strongest-backed idea first. Mix formats so they don't film 7 of the same thing.
## Step 6: Write the file
Save to `competitor-analysis-YYYY-MM-DD.md` in the current folder:
```
# Competitor analysis: [niche], [date]
## Where the numbers came from
[Apify / public pages / pasted by you, per account. Any account with thin data.]
## The accounts
| Account | Posts checked | Median views | Outliers (3x+) |
## Every outlier
| Account | Link | Date | Views | Score | Idea | Angle | Format | Hook |
## The patterns that win
1. [Pattern]: backed by [N] outliers from [N] accounts. [Why it works in one line.]
...
## What isn't working
[Topics that keep getting posted and keep flopping]
## Your next 7 videos
### 1. [Title]
...
```
Then tell the user, in chat: the top 3 patterns in one line each, and the 7 video titles. Ask if they want the full script for any of them.
## Running it again
Next time, read `competitors.md` and the last `competitor-analysis-*.md` first. Only pull posts newer than the last run, and call out what changed: new patterns, patterns that stopped working, any competitor who suddenly broke out.
## What NOT to do
- Don't invent or estimate a view count. Real numbers or ask.
- Don't run a paid scrape without saying the cost and getting a yes.
- Don't rank by raw views. Always by the score against that account's median.
- Don't call one viral video a pattern.
- Don't hand back 7 copies of someone else's video. Same idea, their angle.
===== END FILE =====
===== BEGIN FILE: ~/.claude/skills/edit/SKILL.md =====
---
name: edit
description: Edits your raw talking-head video into a finished short-form video (Reels, TikTok, Shorts) in YOUR editing style. The first time, it interviews you about your style and studies 3 to 5 videos you'd love yours to look like, then it clips out every retake and dead second, captions it, adds graphics that show what you're saying the moment you say it, shows you a storyboard to approve, renders it and masters the audio. Runs on Zach Geleott's free open-source editor (github.com/TakeOff-Team/claude-video-editor-blotato), which it installs for you. Needs Claude Code on a Mac. Triggers on "/edit", "edit this video", "edit my video", "edit this raw footage", "make this a reel", "build my editing style".
argument-hint: "[path to your raw video]"
allowed-tools: Read, Write, Edit, Bash, Glob, AskUserQuestion
---
# /edit
You are the user's video editor. You don't do the editing from scratch in this file: you install and drive a free open-source editor that already does it well, the **Claude Video Editor (Blotato edition)** by Zach Geleott:
- Repo: https://github.com/TakeOff-Team/claude-video-editor-blotato
- What it does: clips raw footage (keeps the last good take of every line, cuts stutters, retakes and dead space), captions every word, plans graphics that show what's being said the moment it's said, shows a storyboard for approval, renders with Remotion, masters the audio to -14 LUFS, and checks its own work.
- It learns the user's style once (a short interview + 3 to 5 reference videos) and remembers every note they give it in `my-style/`.
## Step 0: Can this computer run it?
- **It needs Claude Code on a Mac** (Apple Silicon or Intel) with about 5 GB free. Run `uname -s`. If it isn't `Darwin`, stop and say plainly: "/edit runs on a Mac with Claude Code. It can't run on Windows yet, on your phone, or inside the regular Claude chat app, because it needs to run real video tools (ffmpeg, Whisper, Remotion) on your computer."
- If you're running somewhere without a terminal (claude.ai, the phone app, Cowork), say the same thing and stop.
## Step 1: Find or install the editor
Look for an existing copy, in this order:
1. A path saved in `~/.claude/skills/edit/editor-path.txt`
2. `~/claude-video-editor-blotato`
3. `~/Documents/claude-video-editor-blotato`
4. `find ~ -maxdepth 3 -type d -name claude-video-editor-blotato 2>/dev/null | head -1`
**Not found?** Tell them what you're about to do, then:
```bash
git clone https://github.com/TakeOff-Team/claude-video-editor-blotato ~/claude-video-editor-blotato
cd ~/claude-video-editor-blotato
./setup.sh
```
- `setup.sh` installs ffmpeg, Node, Python, Whisper (speech to text, runs on the Mac), yt-dlp and Remotion. It takes a while the first time. If anything fails, fix it and run it again until it says "All set."
- The one thing you can't do for them is install **Homebrew**, because it needs their Mac password. If setup asks for it, give them the exact command to paste into Terminal, wait for them to say it's done, then keep going.
- No `git`? Running `git --version` on a Mac offers to install the Apple command line tools. Have them click Install, then try again.
Save the editor's full path to `~/.claude/skills/edit/editor-path.txt`.
**Found it?** Run `./edit doctor` inside it. If anything fails, run `./setup.sh` and fix what it reports. If it's been a while, offer `git pull` to get the latest version (ask first if they've changed any files in it).
## Step 2: Work from inside the editor
The editor's own instructions live in its `.claude/skills/` folder, and every command (`./edit ...`) runs from the editor folder.
- Do all the work below with the editor folder as your working directory (`cd` into it for every command).
- **Read the editor's own skill files before doing anything and follow them exactly.** They're the source of truth, and they're more detailed than this file:
- `.claude/skills/edit-video/SKILL.md` (the full edit, end to end) and every file in `.claude/skills/edit-video/references/` it tells you to read
- `.claude/skills/build-style/SKILL.md` (the style interview)
- Tip for the user: for the smoothest run, they can also open Claude Code directly inside the editor folder (`cd ~/claude-video-editor-blotato && claude`). Then its own commands (`/edit-video`, `/build-style`, `/clip`, `/captions`, `/storyboard`, `/get-video`) all work directly.
## Step 3: The style interview (first time only)
If `my-style/style.md` doesn't exist in the editor folder, they don't have a style yet. Don't edit in a style you made up. Run the editor's `/build-style` flow first (`.claude/skills/edit-video/references/onboarding.md`):
- One question at a time, like a conversation with an editor, not a form: what the videos are about and who they're for, what they film, 3 words for how it should feel, **3 to 5 videos they'd love theirs to look like** (links or files, their own best videos count), brand colours and fonts, how their videos end, captions, music, and the names and brands they say a lot.
- Study every reference with `./edit study <link or file>` and actually look at the frames it produces.
- Write `my-style/style.md`, `style.json` and `feedback.md`, and show a sample before the first real edit.
Tell them plainly: the editor is only as good as the videos it learns from, so pick ones they genuinely want to look like.
## Step 4: Edit the video
Ask for the video file (they can drag it into the chat) and the script if they have one (optional). Then follow `.claude/skills/edit-video/SKILL.md` step by step:
1. `./edit new <slug> "<video>" [script]`
2. Clip it if it's raw, and **get their OK on the cut**
3. Captions, then fix any misheard names
4. Plan the beats, using their style and the one rule: every visual shows what's being said, the moment it's said
5. **Storyboard, and wait for their approval.** Never render before they approve
6. Render, master the audio, QA
7. Hand them the finished file path
Every note they give goes into `my-style/feedback.md`, so the next video needs less back and forth. Set the expectation: the first 2 or 3 videos take some back and forth on the storyboard. After that it knows their style.
## Step 5: After the edit
Tell them what's next: `/schedule` posts it at the time they want, `/trial-reels` tests 5 hooks on it first, and `/comment-to-dm` sends a free resource to everyone who comments. (The editor also has its own `/post` and `/automation` commands that do the same through Blotato, if they're working inside the editor folder.)
## What NOT to do
- Don't pretend this works on a phone, in claude.ai or on Windows. It needs Claude Code on a Mac.
- Don't render before they approve the storyboard.
- Don't edit before they have a style. Interview first.
- Don't invent numbers, names, results or screenshots in the graphics. Real assets only, or ask.
- Don't re-cut footage they already cut themselves unless they ask (`./edit precut` uses it as-is).
===== END FILE =====
===== BEGIN FILE: ~/.claude/skills/trial-reels/SKILL.md =====
---
name: trial-reels
description: Turns one finished video into 5 Instagram trial reels with 5 genuinely different hooks (spoken opening line, on-screen text hook and caption for each), schedules all 5 through Blotato for tomorrow, spaced out, so they go to non-followers first. About 3 days later it pulls each trial's numbers and tells you which one to share with your followers. Needs a paid Blotato plan and 1,000+ Instagram followers. Triggers on "/trial-reels", "trial reels", "test hooks on this", "make 5 versions of this", "test this video before I post it", "which hook won".
argument-hint: "[finished video, or 5 videos] | results"
allowed-tools: Read, Write, Bash, Glob, AskUserQuestion, mcp__blotato__blotato_list_accounts, mcp__blotato__blotato_create_presigned_upload_url, mcp__blotato__blotato_create_post, mcp__blotato__blotato_list_posts, mcp__blotato__blotato_get_post_analytics
---
# /trial-reels
Instagram trial reels are shown to people who **don't** follow you first. Your followers never see a trial unless you share it with them. That makes them the cleanest free test on Instagram: post 5 versions of one video as trials, see which one strangers actually watch, and only put the winner in front of your followers.
This skill has two jobs:
| Job | When |
|---|---|
| **Launch** | They have a finished video and want to test it |
| **Results** | It's been about 3 days and they want to know which one won |
If they say "/trial-reels results" or ask which one won, go straight to Results.
## What this skill does and doesn't do (say this up front, once)
- **It does:** write 5 different hooks for the same video (spoken line + on-screen text + caption), match recorded hook clips to them if they filmed more than one, schedule 5 trial reels for tomorrow, spaced out, and read the results.
- **It doesn't re-edit the video.** It can't add the on-screen text or swap the spoken hook into the video file by itself. If the `/edit` skill is installed (the free Claude Video Editor), it can make the 5 versions for them. Otherwise they add the text hook in their own editor (Instagram's Edits app or CapCut, about 30 seconds a version) and give it the 5 files.
- **The rule that makes it a real test:** the 5 trials need to start differently. 5 copies of the same file with 5 captions only tests the caption, because most people decide in the first 2 seconds, before they read anything. Instagram may also treat identical uploads as duplicates. Say this if they only give one file.
## Before anything: check
1. `blotato_list_accounts`. No Blotato tools means it isn't connected. 403 means their plan doesn't include the MCP: Claude can only use Blotato on a **paid** plan (https://blotato.com/?ref=zach, code `ZACH30` for 30% off). Setup steps are in the `/schedule` skill, or https://help.blotato.com/start-with-an-ai-agent/mcp/setup.
2. **Instagram has to be connected and be a Business or Creator account.**
3. **Trial reels need 1,000+ followers.** Ask if they're past 1,000. If not, stop and tell them: Instagram doesn't offer trial reels under 1,000 followers. Offer to schedule the best version as a normal reel with `/schedule` instead.
## Job 1: Launch
### Step 1: The video and what it says
Ask for the finished video (or 5 versions of it, or the video plus extra spoken hook clips). Then get what the video says: the script, or a quick summary of every line. If they don't have the script, ask them to paste the first 3 lines and the main point. Don't guess what the video says.
### Step 2: Write 5 genuinely different hooks
Different means a different reason to keep watching, not the same sentence reworded. Use 5 different angles, for example:
1. **The result:** lead with the outcome ("This one prompt schedules a week of posts")
2. **The secret:** "Nobody's talking about this Claude feature"
3. **The mistake:** "You're posting your reels wrong"
4. **The number:** a real, specific number from the video (never made up)
5. **The callout:** "If you post on Instagram, steal this"
For each one, write:
- **Spoken hook:** the first sentence, word for word, under 2.5 seconds out loud
- **On-screen text:** 3 to 7 words for the first frame
- **Caption:** first line restates the hook, 2 to 4 short lines, one call to action, 3 to 5 hashtags. If a comment-to-DM is running on these, every caption ends with the same keyword: `Comment "KEYWORD" and I'll send it to you`
Label them H1 to H5 and show all 5. Take edits.
### Step 3: Match the files to the hooks
- **They recorded several spoken hooks** (separate clips, or one video with them back to back): list the clips with a short description of each (what's said in the first seconds; transcribe it if Whisper or the `/edit` editor is installed), and let them pick which clip goes with which hook. If `/edit` is installed, offer to cut each hook clip onto the body of the video for them.
- **They have 5 finished versions:** ask which file is H1, H2 and so on. Don't assume from the file names.
- **They have 1 file:** give them the 5 on-screen text hooks to add in their editor and wait for the 5 files. If they'd rather go ahead with one file, say once that it's a caption test, then do it.
### Step 4: Schedule 5 trials for tomorrow
Default timing: tomorrow, in their timezone, spaced out so each gets its own window: **9:00 AM, 11:30 AM, 2:00 PM, 4:30 PM, 7:00 PM**. Change it if they want. Say the times back in plain words before scheduling.
For each version:
1. Upload it (local files): `blotato_create_presigned_upload_url` with the filename, then `curl -X PUT "<presignedUrl>" --data-binary "@<file>"`, then use the `publicUrl`.
2. Schedule it:
```
blotato_create_post({
accountId: "<instagram account id>",
platform: "instagram",
text: "<that version's caption>",
mediaUrls: ["<publicUrl>"],
mediaType: "reel",
trial: { graduationStrategy: "MANUAL" },
scheduledTime: "<ISO 8601 with their timezone offset>"
})
```
`MANUAL` means none of them go to followers until they choose the winner. If they'd rather let Instagram decide, `SS_PERFORMANCE` shares a trial with followers automatically if it does well. Ask once, default to `MANUAL`.
If they have a cover image for each version, pass it as `coverImageUrl` (upload it the same way).
### Step 5: Save the log and hand back proof
Write `trial-reels-log.md` in the current folder:
```
# Trial reels: [video name]
Scheduled: [date] Check results: [3 days after the last one goes live]
| Version | Goes live | Post id | Spoken hook | On-screen text | File |
|---------|-----------|---------|-------------|----------------|------|
| H1 | Wed 9:00 AM | abc-123 | ... | ... | ... |
```
Show them the table and tell them: "Come back in about 3 days and type /trial-reels results." Mention they can see the queue at https://my.blotato.com/scheduler.
## Job 2: Results (about 3 days after)
1. Read `trial-reels-log.md`. If it's not there, ask which trials to check.
2. `blotato_list_posts` (platform `instagram`, status `published`, since the scheduled date) to match each post id to its published reel, then `blotato_get_post_analytics` for each one.
3. **If Blotato has no metrics yet** (metrics come back null, or a trial isn't synced), don't guess. Ask them to open each trial in the Instagram app, check its insights, and paste the views, likes, shares and saves for each version.
4. If it's been less than 48 hours since the last trial went live, say the numbers are early and offer to check again tomorrow.
5. Rank them. **Views decide it, then shares and saves as the tiebreaker** (shares and saves are the strongest signal that strangers liked it, not just stopped on it). Show a table: version, hook, views, likes, shares, saves.
6. Call the winner, and say if it's close: if the top 2 are within about 15% of each other, it's basically a tie. Then they pick the one they like more.
7. **What to do with the winner:** open the winning trial in the Instagram app and share it with your followers from there. Or, if they'd rather post it fresh as a normal reel, schedule the winning file with `/schedule`.
8. One line on what the winner tells them for next time: which angle won (result, secret, mistake, number or callout). Add the results to `trial-reels-log.md`.
## What NOT to do
- Don't schedule trials for someone under 1,000 followers. Instagram won't run them.
- Don't write 5 hooks that are the same sentence reworded.
- Don't say the video was re-edited when it wasn't. Be clear about who adds the on-screen text.
- Don't pick a winner from made-up or estimated numbers. Real numbers, or ask them to paste.
- Don't share a trial with followers yourself. The user decides.
===== END FILE =====
===== BEGIN FILE: ~/.claude/skills/comment-to-dm/SKILL.md =====
---
name: comment-to-dm
description: Builds, runs, checks and upgrades Instagram and Facebook comment-to-DM automations through the Blotato MCP, replacing a separate ManyChat subscription. Use it to promote a digital product, a free resource or your newsletter. When someone comments a keyword they get a DM with a link, optionally after following the account and giving their email. Three jobs. Build a new keyword automation, Check whether a live one is working, and Upgrade one to capture emails through a webhook. Use when someone says "set up a comment to DM", "comment X and I'll send it", "build the DM automation for this reel", "is my DM automation working", "why isn't my automation sending", "capture emails from my DMs", or /comment-to-dm.
argument-hint: "[build | check | upgrade] [keyword] [link]"
allowed-tools: Read, AskUserQuestion, mcp__blotato__blotato_list_accounts, mcp__blotato__blotato_list_posts, mcp__blotato__blotato_list_automations, mcp__blotato__blotato_create_automation, mcp__blotato__blotato_update_automation, mcp__blotato__blotato_update_automation_trigger, mcp__blotato__blotato_get_automation_analytics, mcp__blotato__blotato_list_automation_runs, mcp__blotato__blotato_list_automation_logs
---
# Comment to DM
Someone comments your keyword. They get a DM with your link. Optionally they have to follow you first, and optionally they give you their email on the way through. That loop is the highest-converting free thing a creator can run, and Blotato runs it without a separate DM tool.
Use it for whatever you want people to get: a digital product, a free guide, a template, or your newsletter signup page. The link is the only thing that changes.
**Before any job:** call `blotato_list_accounts`. No Blotato tools means Blotato isn't connected to Claude yet (setup steps are in the `/schedule` skill, or https://help.blotato.com/start-with-an-ai-agent/mcp/setup). A 403 means their plan doesn't include the MCP: Claude can only use Blotato on a **paid** plan (https://blotato.com/?ref=zach, code `ZACH30` for 30% off). The Instagram account has to be a Business or Creator account.
| Job | When |
|---|---|
| **Build** | New post, new keyword, nothing exists yet |
| **Check** | It's live and they want to know if it's working, or it seems broken |
| **Upgrade** | It works, and now they want email addresses out of it |
If it's unclear which job they want, ask.
## How Blotato automations work (read this before any job)
- **Triggers.** A comment trigger fires when a comment contains one of the keywords. It watches either one post (`postId`) or every post (`postId` left out). An automation can also carry one DM trigger, so people who DM the keyword instead of commenting it get the same thing.
- **Scoping to one post needs a Blotato post id.** `postId` is the post's id from `blotato_list_posts`, which only lists posts published or scheduled through Blotato. A reel posted straight from the Instagram app won't be there, so it can't be scoped, and the automation has to run on every post. If scoping matters, publish the reel through Blotato.
- **Gates run before the DM.** A follow gate (Instagram only) holds the DM until they follow: they get the gate message, and once they've followed and tapped the button (or replied), the DM goes out. An email gate asks for an email first. People with an email already on file skip it, and a reply with no email in it gets the gate message again.
- **Webhook.** Fires after the DM is sent. Every method except GET sends a body of `{"email": ..., "isFollower": ...}`, each null when unknown. The URL has to be publicly reachable. Private, loopback and link-local addresses are rejected.
- **Limits.** DM and gate messages are 1 to 640 characters. Up to 3 link buttons, each label 20 characters or fewer, each an http(s) URL.
- **New automations start as inactive drafts** unless created with `isActive: true`. **Edits to an active automation go live the moment they're saved.**
- **Waiting windows.** A run waiting on a follow gate expires after 30 days. A run waiting on an email expires after 72 hours. Expired runs count as triggered, but never as completed or failed.
---
## Job 1: Build
### Step 1: Collect 4 things
Don't start building with a gap. Ask for whatever's missing.
1. **KEYWORD.** One word, the word they tell people to comment.
2. **LINK.** The URL the DM sends people to.
3. **THE FREE THING.** What they're sending, in one short phrase, the way they'd say it out loud.
4. **WHICH POST.** One specific post, or every post.
Defaults unless they say otherwise: follow gate on (Instagram), no email gate, one link button, DM trigger off.
### Step 2: Check the keyword before building anything
This is the step that quietly breaks launches.
1. **Existing automations.** Run `blotato_list_automations`. If an active automation on the same account already uses this keyword on every post, the two will fight over the same comments. The fix is scoping, not a new keyword: scope the new one to its post, and flag the old any-post one so they can scope it or pause it.
2. **Everyday words.** Matching is on "contains", so assume ordinary comments that include the word will fire it. On an any-post automation, warn plainly about short or common words (AI, ME, GO, LINK). Scoped to one post, a common word is usually fine, because anyone saying it on that post probably wants the thing.
3. **Case.** Pass the keyword in caps, lowercase and title case (for example `CONTENT`, `content`, `Content`). It costs nothing and removes a guess.
4. **Already filmed?** If the video already says the keyword, the keyword stays. Scope instead.
### Step 3: Find the post (only if scoping)
Run `blotato_list_posts` with platform `instagram` and status `published`. Match on the caption or the postUrl, and confirm the match with the user.
If the post isn't there, it was posted natively. Explain the Blotato post id rule and give them the two options: publish the reel through Blotato (the `/schedule` skill can), or run the automation on every post and accept the collision risk. Their call.
### Step 4: Show the build, then wait
Run `blotato_list_accounts` and name the account this is about to attach to. For Facebook, it also needs the `pageId` from that account's subaccounts, and it can't have a follow gate.
Then show the whole thing:
```
Account: [platform, @handle]
Name: KEYWORD - comment to DM
Trigger: comment-received
Keywords: KEYWORD, keyword, Keyword
Post: [the post, or every post]
Follow gate: ON
Message: "Yes! Follow me first and I'll send it straight over.
Tap the button below once you're following
(or just reply "done" if you don't see a button)."
Button: "I'm following"
DM message: "Here it is, THE FREE THING. Tap below and it's yours."
Link button: "Send it over" -> LINK
Active on create: yes
```
Write the DM like a person talking: what it is, then "tap below". Keep it short. Nobody reads a long DM.
**Wait for an explicit go. Don't create anything before they confirm.**
### Step 5: Create it, then hand back proof
Create it with `isActive: true`. Report the automation id, the account, the keywords, the scope, and that it's active.
Then tell them the thing everyone forgets: **your own comments never trigger it.** Test from a second Instagram account. People assume it's broken when it's working fine.
---
## Job 2: Check
1. **Find it.** `blotato_list_automations` with metrics `triggered`, `completed`, `failed`, `issues`. Match on name or keyword. If nothing matches, list them all and ask which one.
2. **Is it active?** An inactive draft never fires. That's the most common "it's broken."
3. **Triggered is 0?** Work through this in order: the automation is active; its trigger is active; it's scoped to the right post id; they didn't test from their own account; the keyword they typed matches.
4. **Triggered is high, completed is low?** Run `blotato_list_automation_runs` with status `waiting` and `expired`. **Waiting** means sitting at a gate (up to 30 days for a follow, 72 hours for an email). **Expired** means they never answered. **Superseded** means the same person triggered it again and the newer run took over. None of these are failures. Report them before calling anything broken.
5. **Failed above 0?** Run `blotato_list_automation_runs` with status `failed`, then `blotato_list_automation_logs` with that run's `flowRunId`. Quote the error and give the exact fix. `issues` is the failures from the last 7 days, which tells you whether the problem is current or old.
6. **Answer in plain English:**
- Is it live right now?
- How many times has it fired?
- How many people actually got the DM?
- How many are waiting at a gate, and how many gave up?
- If anything really failed: the reason, and the exact fix.
---
## Job 3: Upgrade to capture email
### Step 1: Get 2 things
1. **WHICH ONE.** The automation name or its keyword.
2. **WEBHOOK URL.** Their endpoint: Zapier, Make, n8n, or their own.
### Step 2: Say the part that matters, before building
Blotato saves the email to the contact, but **the webhook is the only thing that sends it on to their email list.** No working webhook means the email never reaches their list.
So before going live:
- Confirm the URL is publicly reachable. Private, loopback and link-local addresses get rejected.
- Use POST. A GET webhook sends no body, which means no email.
- Once it's live, check the first delivery in their tool's history (the Zapier, Make or n8n run log) before trusting it.
### Step 3: Add only 2 things
- **Email gate.** Message: "One thing first, what's the best email to send this to? I'll send it right after."
- **Webhook.** Method POST, URL = their webhook.
Keep everything else exactly as it is: same keyword, same scope, same follow gate, same DM, same button.
This automation is live, so the update publishes the moment it saves. **Show the full updated automation and wait for a go.**
### Step 4: Tell them what the person experiences
Comment, then the gate or gates, then the DM with the link. The webhook fires after the DM goes out. On the first test run, check the logs to see the exact gate order.
Say plainly that every gate costs some conversion. Two gates in front of a link is worth it when they want an email list. It isn't worth it for a pure follower push.
---
## What NOT to do
- **Don't create or edit anything before an explicit go.** Edits to an active automation go live instantly.
- **Don't skip the keyword check.** Two any-post automations on one keyword is the most common silent failure.
- **Don't put a link button and a plain-text URL in the same DM.** On desktop the URL stops being clickable, and those readers end up with no working link at all.
- **Don't call low completion broken** until you've counted waiting and expired runs.
- **Don't promise email capture without a working POST webhook.**
- **Don't change a keyword that's already been filmed.** Scope the automation instead.
- **Don't add a follow gate on Facebook.** The automation can't go live with one.
===== END FILE =====
- Open Claude Code. The terminal app, or the Code tab in the Claude desktop app. Paste the prompt and send it.
- Answer its questions. It explains every command before it runs it, and asks before it installs the video editor's tools.
- Restart Claude Code when it says so, so the 5 skills load. Then type
/and you will see them.
1. /schedule
Upload a finished video, type /schedule, and it goes out whenever you want.
You drag in the video and say something like "tomorrow at 9am on Instagram and TikTok." Claude uploads the file to Blotato, writes a caption if you want one, runs a quick check, and schedules it. Then it hands back a table with every platform, the exact time in your timezone, and the post id, so you know it actually worked.
It posts to Instagram, TikTok, YouTube, Facebook, LinkedIn, X, Threads, Bluesky and Pinterest. If Blotato is not connected yet, it does not fail. It saves the caption and the time to a file you can paste by hand.
What it needs
- Claude Claude Code, Claude desktop or claude.ai
- Blotato Paid plan, accounts connected
---
name: schedule
description: Schedules a finished video (or any post) to Instagram, TikTok, YouTube, Facebook, LinkedIn, X, Threads, Bluesky or Pinterest through Blotato, for whatever time you say. Uploads the video file, writes or takes the caption, runs a quick pre-publish check, schedules it, and hands back proof (post ids and times). Saves a copy-paste file instead if Blotato isn't connected. Triggers on "/schedule", "schedule this", "schedule this video", "post this tomorrow at 9", "put this on Instagram", "post this to TikTok and YouTube".
argument-hint: "[video file] [platforms] [time]"
allowed-tools: Read, Write, Bash, Glob, AskUserQuestion, mcp__blotato__blotato_list_accounts, mcp__blotato__blotato_create_presigned_upload_url, mcp__blotato__blotato_create_post, mcp__blotato__blotato_get_post_status, mcp__blotato__blotato_list_posts
---
# /schedule
You take a finished video and schedule it as a post, at the time the user wants, on the platforms they pick. You don't edit the video and you don't rewrite their caption unless they ask. Your job is to get it live, correctly, and prove it.
Everything runs through **Blotato** (blotato.com), which connects Claude to social accounts. The tools come from the Blotato MCP server: `blotato_list_accounts`, `blotato_create_presigned_upload_url`, `blotato_create_post`, `blotato_get_post_status`, `blotato_list_posts`. If the user's Blotato tools carry a different prefix (for example a desktop connector), use those. Same tools, same names after the prefix.
## Step 0: Is Blotato connected?
Call `blotato_list_accounts`.
- **The tool doesn't exist.** Blotato isn't connected to Claude yet. Walk them through "Connect Blotato" at the bottom of this file. Stop until it works, or offer the fallback file (Step 6).
- **403 or a subscription error.** Their plan doesn't include the MCP. Claude can only talk to Blotato on a **paid** plan (the free trial doesn't include it). Point them to https://blotato.com/?ref=zach and the code `ZACH30` (30% off). Stop.
- **401.** The sign-in went stale. In Claude Code: type `/mcp`, pick blotato, choose Authenticate. Then try again.
- **Empty list.** Connected, but no social accounts yet. Have them connect accounts at https://my.blotato.com/accounts, then call it again.
## Step 1: Get the 3 things
1. **The video.** A file path (they can drag the file into the chat), or a public URL. If it's a local file, check it with `ffprobe` if it's available: vertical 9:16 for Reels/TikTok/Shorts, has audio, under 3 minutes for Instagram Reels, under 60 seconds if they want a YouTube Short. Tell them if something's off. Don't block on it.
2. **Where.** Show their connected accounts as a short list and ask which ones. If a platform has 2+ accounts, ask which.
3. **When.** Whatever they say: "tomorrow at 9am", "Friday 6pm", "in 2 hours", "next free slot", "now". Convert it to ISO 8601 **in their timezone** (run `date +%Z` or ask if you can't tell) and say it back in plain words before scheduling: "Tomorrow, Wed Nov 12, 9:00 AM Eastern."
Then the caption. Ask: "Got a caption, or want me to write one?" If you write it: first line says what they get, then 2 to 5 short lines, then one call to action, 3 to 5 hashtags for Instagram and TikTok, none for LinkedIn, X, Threads or Bluesky. If they're running a comment-to-DM on this post, the call to action is the keyword: `Comment "KEYWORD" and I'll send it to you`. Show it, take edits.
## Step 2: Quick pre-publish check
- [ ] Instagram has a video or image attached (Instagram rejects text-only posts)
- [ ] YouTube has a title under 100 characters
- [ ] TikTok has a privacy level (default `PUBLIC_TO_EVERYONE`)
- [ ] No external link in a LinkedIn post body (it kills reach, put it in the first comment instead)
- [ ] Caption fits the platform (Bluesky 300 characters, X 280 unless they have Premium)
- [ ] Hashtags: 3 to 5 max on Instagram and TikTok
- [ ] The time is in the future and in their timezone
If something fails, say what and ask: fix it, skip that platform, or ship as-is. Don't auto-fix without asking.
## Step 3: Upload the video (local files only)
A local file has to be uploaded before Blotato can post it:
1. `blotato_create_presigned_upload_url` with the filename (for example `final.mp4`). You get back `presignedUrl` and `publicUrl`.
2. Upload the raw bytes: `curl -X PUT "<presignedUrl>" --data-binary "@<path to file>"`. Not JSON, not a form.
3. Use `publicUrl` in `mediaUrls` from now on.
A public URL skips this step.
**Claude desktop app:** if the upload fails with a network error, the app is blocking it. Settings, then Capabilities, then allow network egress and add `database.blotato.io`, then fully restart the app.
## Step 4: Schedule it
One `blotato_create_post` call per platform:
```
blotato_create_post({
accountId: "<from blotato_list_accounts>",
platform: "instagram",
text: "<caption>",
mediaUrls: ["<publicUrl>"],
scheduledTime: "2026-11-12T09:00:00-05:00" // or useNextFreeSlot: true
})
```
Leave out `scheduledTime` only when they said "now". Never publish immediately by default.
Platform fields:
- **Instagram:** `mediaType: "reel"` for video. Optional `coverImageUrl`, `firstComment`, `collaborators`. Trial reels (shown to non-followers first, needs 1,000+ followers): add `trial: { graduationStrategy: "MANUAL" }`. If they want 5 trial versions of one video, that's the `/trial-reels` skill.
- **TikTok:** `privacyLevel`, `disabledComments`, `disabledDuet`, `disabledStitch`, `isBrandedContent`, `isYourBrand`, `isAiGenerated` (all required; default false for the booleans unless they say otherwise).
- **YouTube:** `title`, `privacyStatus: "public"`, `shouldNotifySubscribers: true`.
- **Facebook:** `pageId` from the account's subaccounts. `mediaType: "reel"` for vertical video.
- **LinkedIn:** optional `pageId` for a company page. Leave it out for their personal profile.
- **Pinterest:** `boardId`.
If one platform fails, keep going with the rest. Don't roll back.
## Step 5: Hand back proof
```
Scheduled
| Platform | Account | Goes live | Post id |
|-----------|-----------|-------------------------------|---------|
| Instagram | @handle | Wed Nov 12, 9:00 AM Eastern | abc-123 |
| TikTok | @handle | Wed Nov 12, 9:00 AM Eastern | def-456 |
See or edit everything at https://my.blotato.com/scheduler
```
"Done" without a post id isn't done. If they want to double check, `blotato_list_posts` with status `scheduled` shows it in the queue.
## Step 6: Fallback (no Blotato)
If Blotato isn't connected and they don't want to set it up now, save `post-ready-to-paste.txt` in the current folder, one block per platform:
```
=== INSTAGRAM ===
Post at: Wed Nov 12, 9:00 AM Eastern
Video: /path/to/final.mp4
<caption>
=== END ===
```
Tell them the file path. They lose the scheduling, but nothing's lost.
## What NOT to do
- Don't publish right now unless they said "now" or "post it now".
- Don't guess a platform or an account. Ask.
- Don't schedule in UTC by accident. Say the time back in their timezone first.
- Don't send a local file path to `blotato_create_post`. Upload it first.
- Don't say it worked without a post id.
## Connect Blotato (one time, about 5 minutes)
Go one step at a time and wait for them between steps.
1. **A paid Blotato plan.** Sign up at https://blotato.com/?ref=zach. Claude can only talk to Blotato on a paid plan. Code `ZACH30` takes 30% off.
2. **Connect your social accounts** at https://my.blotato.com/accounts. Instagram has to be a **Business or Creator** account (Instagram app: Settings, then Account type and tools). Facebook needs a Facebook Page.
3. **Add Blotato to Claude.**
- Claude Code: run this in Terminal (or let Claude run it): `claude mcp add --transport http --scope user blotato https://mcp.blotato.com/mcp`. Then fully quit Claude Code, open it again, type `/mcp`, pick blotato, choose Authenticate, and approve it in the browser.
- Claude desktop app or claude.ai: Settings, then Connectors, add a custom connector with the URL `https://mcp.blotato.com/mcp`, then sign in.
- Blotato's own guide, if anything here looks out of date: https://help.blotato.com/start-with-an-ai-agent/mcp/setup
4. **Check it.** Ask Claude: "List my connected Blotato accounts." You should see your accounts.
2. /analyze-competitors
It finds which video ideas are working in your niche, then tells you exactly what to make for your next 7 videos.
You tell it your niche and 3 to 10 creators you watch (or it finds them for you). It pulls each one's recent posts with their real view counts, then scores every post against that creator's own median. That part matters. 300K views is a normal day for a huge account and a breakout for a small one, so it only keeps the real outliers: posts that did 3x or more what that account usually does.
Then it groups the outliers into the ideas and angles that keep winning across different creators, flags the topics everyone posts that keep flopping, and writes you a ranked plan for your next 7 videos. Every one comes with the spoken hook, the on-screen text, and the real posts it is based on.
Where the numbers come from: you can paste them from the app (open a creator's Reels tab, every reel shows its views), or connect Apify's Instagram reel scraper, which costs about $0.002 a reel. It asks before it spends anything.
What it needs
- Claude Claude Code, Claude desktop or claude.ai
- Blotato Not needed
- Optional Apify, about $0.002 a reel
--- name: analyze-competitors description: Studies the creators in your niche to find which video ideas are actually working right now, then tells you exactly which 7 videos to make next, with a hook for each. Gets your niche and 3 to 10 competitor accounts (or finds them), pulls their recent posts with real view counts, scores every post against that account's own median, keeps the outliers (3x+ median), groups them into the ideas and angles that keep winning, and writes a ranked "next 7 videos" plan to a file. Never invents numbers. Triggers on "/analyze-competitors", "analyze my competitors", "what's working in my niche", "what should my next videos be", "find me video ideas from my competitors", "study these creators". argument-hint: "[your niche] [@competitor1 @competitor2 ...]" allowed-tools: Read, Write, Bash, Glob, WebSearch, WebFetch, AskUserQuestion, mcp__blotato__blotato_create_source, mcp__blotato__blotato_list_top_posts --- # /analyze-competitors You find the video ideas that are provably working in the user's niche right now, and turn them into their next 7 videos. The whole thing rests on one idea: **a video isn't good because it got a lot of views. It's good because it got a lot more views than that account usually gets.** A creator with 2 million followers getting 300K views is a normal day. A creator with 20K followers getting 300K views found something. So every post is scored against its own account's median, and only the outliers count. ## The rule you never break **Never invent a view count, a follower count, a date or a link.** Every number in the report has to come from a real source: a scraper, a page you actually read, or the user pasting it. If you can't get real numbers for an account, say so plainly and ask the user to paste them. A plan built on made-up numbers is worse than no plan. ## Step 1: Get the inputs Ask for whatever's missing, one question at a time: 1. **Your niche, in one sentence.** Who you make videos for and about what. Example: "AI tools for small business owners who aren't technical." 2. **Your platform.** Instagram Reels, TikTok or YouTube Shorts. Default Instagram. 3. **3 to 10 competitor accounts.** Handles or profile links. Creators in the same niche, ideally a mix of big ones and ones around your size (the smaller ones teach you more, because their outliers are about the idea, not the audience). - **Don't have a list?** Offer to find them. Web search things like `top [niche] creators instagram`, `[niche] tiktok creators`, `best [niche] accounts to follow`. Only suggest accounts you can confirm exist and posted in the last 30 days. Show the list with one line each on why, and let them cut or add before you start. 4. **Your own handle** (optional). So the plan can avoid ideas they've already made. Save the final list to `competitors.md` in the current folder so next time you can skip this step. ## Step 2: Get real posts with real view counts For each account, you want its last 20 to 30 posts (or last 60 days), each with: link, date, view count, and the caption or what the video's about. Try these in order and tell the user which one you're using: **A. Apify (best, optional, costs a little).** If the user has the Apify MCP connected or an `APIFY_TOKEN`: - Instagram: the `apify/instagram-reel-scraper` actor (about $0.002 a reel at the time of writing, so 10 accounts x 30 reels is roughly $0.60). TikTok: `clockworks/tiktok-scraper`. Check the actor's current input schema before running. - **Say the estimated cost and wait for a yes before running.** Never run a paid scrape without asking. - Don't have it? Mention it once as an optional upgrade (free account at apify.com), then move on to B or C. **B. Public pages.** Use web search and fetch on the profile and post pages. TikTok and YouTube pages often show view counts. Instagram mostly hides them from logged-out visitors, so expect this to fail for Instagram. Only keep a number if you actually read it on the page. **C. The user pastes them (always works, takes about 2 minutes per account).** Ask them to open each competitor's Reels tab in the app, where every reel shows its view count on the grid, and paste a list like: ``` @handle https://www.instagram.com/reel/ABC123/ 412K "the hook or what it's about" https://www.instagram.com/reel/DEF456/ 38K "..." ``` Abbreviations are fine (412K, 1.2M). Mixing is fine too: scrape what you can, ask them to paste the rest. **Blotato note.** If Blotato is connected, `blotato_create_source` can pull the transcript of a TikTok or YouTube video, which helps with Step 4 (what the video actually says). It doesn't list other accounts' posts or their views. `blotato_list_top_posts` shows the user's OWN best posts, which is useful for comparing their winners to their competitors'. Neither one replaces Step 2. If an account ends up with fewer than 8 posts with real views, keep it in the report but flag it as thin data. ## Step 3: Score every post For each account: 1. Find the **median** views of its posts (the middle number, not the average, so one viral video doesn't skew it). 2. Score each post: `views / median`. That's its outlier score. 4.0x means it did 4 times what that account normally does. 3. **Keep the outliers: 3x median or more.** 4. If there are fewer than 6 outliers across every account combined, lower the bar to 2x and say you did. 5. Drop anything under 2 days old (it hasn't finished getting views yet) and anything that's clearly a paid ad or a collab with a huge account (say which ones you dropped and why). ## Step 4: Find the patterns For every outlier, write down: - **The idea.** What the video is about in one line. - **The angle.** Why someone would care: money, fear of missing out, a shortcut, a secret, a mistake, a before/after, a tool reveal, a "nobody's talking about this". - **The format.** Listicle, tutorial, story, news, reaction, myth-bust, side-by-side. - **The hook.** The first line (from the caption, the on-screen text, or the transcript if you have it). Then group them. A pattern is real when **2 or more outliers from 2 or more different accounts** share it. One viral video from one account is a fluke, not a pattern. Rank the patterns by how many outliers back them and how big the scores are. Also note what's NOT working: topics every competitor keeps posting that never break their median. Those are the ideas to skip. ## Step 5: The next 7 videos Turn the patterns into 7 specific videos for the user's niche. Not copies of the competitors' videos: the same winning idea, made for the user's audience, from the user's angle. For each one: - **The video:** one line on what it is - **Why it should work:** the pattern, with the real outliers behind it (account, link, views, score) - **Format and rough length** - **Spoken hook:** the first sentence, word for word - **On-screen text hook:** 3 to 7 words for the first frame - **What to show:** 1 or 2 lines Order them: the strongest-backed idea first. Mix formats so they don't film 7 of the same thing. ## Step 6: Write the file Save to `competitor-analysis-YYYY-MM-DD.md` in the current folder: ``` # Competitor analysis: [niche], [date] ## Where the numbers came from [Apify / public pages / pasted by you, per account. Any account with thin data.] ## The accounts | Account | Posts checked | Median views | Outliers (3x+) | ## Every outlier | Account | Link | Date | Views | Score | Idea | Angle | Format | Hook | ## The patterns that win 1. [Pattern]: backed by [N] outliers from [N] accounts. [Why it works in one line.] ... ## What isn't working [Topics that keep getting posted and keep flopping] ## Your next 7 videos ### 1. [Title] ... ``` Then tell the user, in chat: the top 3 patterns in one line each, and the 7 video titles. Ask if they want the full script for any of them. ## Running it again Next time, read `competitors.md` and the last `competitor-analysis-*.md` first. Only pull posts newer than the last run, and call out what changed: new patterns, patterns that stopped working, any competitor who suddenly broke out. ## What NOT to do - Don't invent or estimate a view count. Real numbers or ask. - Don't run a paid scrape without saying the cost and getting a yes. - Don't rank by raw views. Always by the score against that account's median. - Don't call one viral video a pattern. - Don't hand back 7 copies of someone else's video. Same idea, their angle.
3. /edit
Upload raw footage. It interviews you about your editing style, then edits the whole video in that style.
The first time, it runs a short interview like a real editor would: what your videos are about, how they should feel, and the most important question, 3 to 5 videos you would love yours to look like. It studies those frame by frame and writes your style guide.
After that, you drag in a raw video. It cuts every retake, stutter and dead second (it keeps your last good take of every line), captions every word, adds graphics that show what you are saying the moment you say it, and shows you a storyboard of every beat before it renders anything. You leave notes, it remembers every one, and the next video needs fewer.
What it needs
- Claude Claude Code on a Mac
- Space About 5 GB free
- Blotato Not needed
The editor is free and open source: claude-video-editor-blotato on GitHub. The install prompt downloads it for you.
--- name: edit description: Edits your raw talking-head video into a finished short-form video (Reels, TikTok, Shorts) in YOUR editing style. The first time, it interviews you about your style and studies 3 to 5 videos you'd love yours to look like, then it clips out every retake and dead second, captions it, adds graphics that show what you're saying the moment you say it, shows you a storyboard to approve, renders it and masters the audio. Runs on Zach Geleott's free open-source editor (github.com/TakeOff-Team/claude-video-editor-blotato), which it installs for you. Needs Claude Code on a Mac. Triggers on "/edit", "edit this video", "edit my video", "edit this raw footage", "make this a reel", "build my editing style". argument-hint: "[path to your raw video]" allowed-tools: Read, Write, Edit, Bash, Glob, AskUserQuestion --- # /edit You are the user's video editor. You don't do the editing from scratch in this file: you install and drive a free open-source editor that already does it well, the **Claude Video Editor (Blotato edition)** by Zach Geleott: - Repo: https://github.com/TakeOff-Team/claude-video-editor-blotato - What it does: clips raw footage (keeps the last good take of every line, cuts stutters, retakes and dead space), captions every word, plans graphics that show what's being said the moment it's said, shows a storyboard for approval, renders with Remotion, masters the audio to -14 LUFS, and checks its own work. - It learns the user's style once (a short interview + 3 to 5 reference videos) and remembers every note they give it in `my-style/`. ## Step 0: Can this computer run it? - **It needs Claude Code on a Mac** (Apple Silicon or Intel) with about 5 GB free. Run `uname -s`. If it isn't `Darwin`, stop and say plainly: "/edit runs on a Mac with Claude Code. It can't run on Windows yet, on your phone, or inside the regular Claude chat app, because it needs to run real video tools (ffmpeg, Whisper, Remotion) on your computer." - If you're running somewhere without a terminal (claude.ai, the phone app, Cowork), say the same thing and stop. ## Step 1: Find or install the editor Look for an existing copy, in this order: 1. A path saved in `~/.claude/skills/edit/editor-path.txt` 2. `~/claude-video-editor-blotato` 3. `~/Documents/claude-video-editor-blotato` 4. `find ~ -maxdepth 3 -type d -name claude-video-editor-blotato 2>/dev/null | head -1` **Not found?** Tell them what you're about to do, then: ```bash git clone https://github.com/TakeOff-Team/claude-video-editor-blotato ~/claude-video-editor-blotato cd ~/claude-video-editor-blotato ./setup.sh ``` - `setup.sh` installs ffmpeg, Node, Python, Whisper (speech to text, runs on the Mac), yt-dlp and Remotion. It takes a while the first time. If anything fails, fix it and run it again until it says "All set." - The one thing you can't do for them is install **Homebrew**, because it needs their Mac password. If setup asks for it, give them the exact command to paste into Terminal, wait for them to say it's done, then keep going. - No `git`? Running `git --version` on a Mac offers to install the Apple command line tools. Have them click Install, then try again. Save the editor's full path to `~/.claude/skills/edit/editor-path.txt`. **Found it?** Run `./edit doctor` inside it. If anything fails, run `./setup.sh` and fix what it reports. If it's been a while, offer `git pull` to get the latest version (ask first if they've changed any files in it). ## Step 2: Work from inside the editor The editor's own instructions live in its `.claude/skills/` folder, and every command (`./edit ...`) runs from the editor folder. - Do all the work below with the editor folder as your working directory (`cd` into it for every command). - **Read the editor's own skill files before doing anything and follow them exactly.** They're the source of truth, and they're more detailed than this file: - `.claude/skills/edit-video/SKILL.md` (the full edit, end to end) and every file in `.claude/skills/edit-video/references/` it tells you to read - `.claude/skills/build-style/SKILL.md` (the style interview) - Tip for the user: for the smoothest run, they can also open Claude Code directly inside the editor folder (`cd ~/claude-video-editor-blotato && claude`). Then its own commands (`/edit-video`, `/build-style`, `/clip`, `/captions`, `/storyboard`, `/get-video`) all work directly. ## Step 3: The style interview (first time only) If `my-style/style.md` doesn't exist in the editor folder, they don't have a style yet. Don't edit in a style you made up. Run the editor's `/build-style` flow first (`.claude/skills/edit-video/references/onboarding.md`): - One question at a time, like a conversation with an editor, not a form: what the videos are about and who they're for, what they film, 3 words for how it should feel, **3 to 5 videos they'd love theirs to look like** (links or files, their own best videos count), brand colours and fonts, how their videos end, captions, music, and the names and brands they say a lot. - Study every reference with `./edit study <link or file>` and actually look at the frames it produces. - Write `my-style/style.md`, `style.json` and `feedback.md`, and show a sample before the first real edit. Tell them plainly: the editor is only as good as the videos it learns from, so pick ones they genuinely want to look like. ## Step 4: Edit the video Ask for the video file (they can drag it into the chat) and the script if they have one (optional). Then follow `.claude/skills/edit-video/SKILL.md` step by step: 1. `./edit new <slug> "<video>" [script]` 2. Clip it if it's raw, and **get their OK on the cut** 3. Captions, then fix any misheard names 4. Plan the beats, using their style and the one rule: every visual shows what's being said, the moment it's said 5. **Storyboard, and wait for their approval.** Never render before they approve 6. Render, master the audio, QA 7. Hand them the finished file path Every note they give goes into `my-style/feedback.md`, so the next video needs less back and forth. Set the expectation: the first 2 or 3 videos take some back and forth on the storyboard. After that it knows their style. ## Step 5: After the edit Tell them what's next: `/schedule` posts it at the time they want, `/trial-reels` tests 5 hooks on it first, and `/comment-to-dm` sends a free resource to everyone who comments. (The editor also has its own `/post` and `/automation` commands that do the same through Blotato, if they're working inside the editor folder.) ## What NOT to do - Don't pretend this works on a phone, in claude.ai or on Windows. It needs Claude Code on a Mac. - Don't render before they approve the storyboard. - Don't edit before they have a style. Interview first. - Don't invent numbers, names, results or screenshots in the graphics. Real assets only, or ask. - Don't re-cut footage they already cut themselves unless they ask (`./edit precut` uses it as-is).
4. /trial-reels
It turns one finished video into 5 trial reels with 5 different hooks, schedules them for tomorrow, and tells you which one to post.
Trial reels go to people who do not follow you first. Your followers never see them unless you share one. So you can test 5 versions of the same video on strangers and only show your followers the winner.
It writes 5 genuinely different hooks, each one a different reason to keep watching: the result, the secret, the mistake, the number, the callout. Every hook gets a spoken opening line, an on-screen text hook and its own caption. If you filmed more than one spoken hook, it lets you match each clip to a hook. Then it schedules all 5 as trial reels for tomorrow, spaced out through the day.
About 3 days later you type /trial-reels results. It pulls the numbers for each one, ranks them, and tells you which to share with your followers.
What it needs
- Instagram 1,000+ followers, Business or Creator
- Blotato Paid plan, Instagram connected
---
name: trial-reels
description: Turns one finished video into 5 Instagram trial reels with 5 genuinely different hooks (spoken opening line, on-screen text hook and caption for each), schedules all 5 through Blotato for tomorrow, spaced out, so they go to non-followers first. About 3 days later it pulls each trial's numbers and tells you which one to share with your followers. Needs a paid Blotato plan and 1,000+ Instagram followers. Triggers on "/trial-reels", "trial reels", "test hooks on this", "make 5 versions of this", "test this video before I post it", "which hook won".
argument-hint: "[finished video, or 5 videos] | results"
allowed-tools: Read, Write, Bash, Glob, AskUserQuestion, mcp__blotato__blotato_list_accounts, mcp__blotato__blotato_create_presigned_upload_url, mcp__blotato__blotato_create_post, mcp__blotato__blotato_list_posts, mcp__blotato__blotato_get_post_analytics
---
# /trial-reels
Instagram trial reels are shown to people who **don't** follow you first. Your followers never see a trial unless you share it with them. That makes them the cleanest free test on Instagram: post 5 versions of one video as trials, see which one strangers actually watch, and only put the winner in front of your followers.
This skill has two jobs:
| Job | When |
|---|---|
| **Launch** | They have a finished video and want to test it |
| **Results** | It's been about 3 days and they want to know which one won |
If they say "/trial-reels results" or ask which one won, go straight to Results.
## What this skill does and doesn't do (say this up front, once)
- **It does:** write 5 different hooks for the same video (spoken line + on-screen text + caption), match recorded hook clips to them if they filmed more than one, schedule 5 trial reels for tomorrow, spaced out, and read the results.
- **It doesn't re-edit the video.** It can't add the on-screen text or swap the spoken hook into the video file by itself. If the `/edit` skill is installed (the free Claude Video Editor), it can make the 5 versions for them. Otherwise they add the text hook in their own editor (Instagram's Edits app or CapCut, about 30 seconds a version) and give it the 5 files.
- **The rule that makes it a real test:** the 5 trials need to start differently. 5 copies of the same file with 5 captions only tests the caption, because most people decide in the first 2 seconds, before they read anything. Instagram may also treat identical uploads as duplicates. Say this if they only give one file.
## Before anything: check
1. `blotato_list_accounts`. No Blotato tools means it isn't connected. 403 means their plan doesn't include the MCP: Claude can only use Blotato on a **paid** plan (https://blotato.com/?ref=zach, code `ZACH30` for 30% off). Setup steps are in the `/schedule` skill, or https://help.blotato.com/start-with-an-ai-agent/mcp/setup.
2. **Instagram has to be connected and be a Business or Creator account.**
3. **Trial reels need 1,000+ followers.** Ask if they're past 1,000. If not, stop and tell them: Instagram doesn't offer trial reels under 1,000 followers. Offer to schedule the best version as a normal reel with `/schedule` instead.
## Job 1: Launch
### Step 1: The video and what it says
Ask for the finished video (or 5 versions of it, or the video plus extra spoken hook clips). Then get what the video says: the script, or a quick summary of every line. If they don't have the script, ask them to paste the first 3 lines and the main point. Don't guess what the video says.
### Step 2: Write 5 genuinely different hooks
Different means a different reason to keep watching, not the same sentence reworded. Use 5 different angles, for example:
1. **The result:** lead with the outcome ("This one prompt schedules a week of posts")
2. **The secret:** "Nobody's talking about this Claude feature"
3. **The mistake:** "You're posting your reels wrong"
4. **The number:** a real, specific number from the video (never made up)
5. **The callout:** "If you post on Instagram, steal this"
For each one, write:
- **Spoken hook:** the first sentence, word for word, under 2.5 seconds out loud
- **On-screen text:** 3 to 7 words for the first frame
- **Caption:** first line restates the hook, 2 to 4 short lines, one call to action, 3 to 5 hashtags. If a comment-to-DM is running on these, every caption ends with the same keyword: `Comment "KEYWORD" and I'll send it to you`
Label them H1 to H5 and show all 5. Take edits.
### Step 3: Match the files to the hooks
- **They recorded several spoken hooks** (separate clips, or one video with them back to back): list the clips with a short description of each (what's said in the first seconds; transcribe it if Whisper or the `/edit` editor is installed), and let them pick which clip goes with which hook. If `/edit` is installed, offer to cut each hook clip onto the body of the video for them.
- **They have 5 finished versions:** ask which file is H1, H2 and so on. Don't assume from the file names.
- **They have 1 file:** give them the 5 on-screen text hooks to add in their editor and wait for the 5 files. If they'd rather go ahead with one file, say once that it's a caption test, then do it.
### Step 4: Schedule 5 trials for tomorrow
Default timing: tomorrow, in their timezone, spaced out so each gets its own window: **9:00 AM, 11:30 AM, 2:00 PM, 4:30 PM, 7:00 PM**. Change it if they want. Say the times back in plain words before scheduling.
For each version:
1. Upload it (local files): `blotato_create_presigned_upload_url` with the filename, then `curl -X PUT "<presignedUrl>" --data-binary "@<file>"`, then use the `publicUrl`.
2. Schedule it:
```
blotato_create_post({
accountId: "<instagram account id>",
platform: "instagram",
text: "<that version's caption>",
mediaUrls: ["<publicUrl>"],
mediaType: "reel",
trial: { graduationStrategy: "MANUAL" },
scheduledTime: "<ISO 8601 with their timezone offset>"
})
```
`MANUAL` means none of them go to followers until they choose the winner. If they'd rather let Instagram decide, `SS_PERFORMANCE` shares a trial with followers automatically if it does well. Ask once, default to `MANUAL`.
If they have a cover image for each version, pass it as `coverImageUrl` (upload it the same way).
### Step 5: Save the log and hand back proof
Write `trial-reels-log.md` in the current folder:
```
# Trial reels: [video name]
Scheduled: [date] Check results: [3 days after the last one goes live]
| Version | Goes live | Post id | Spoken hook | On-screen text | File |
|---------|-----------|---------|-------------|----------------|------|
| H1 | Wed 9:00 AM | abc-123 | ... | ... | ... |
```
Show them the table and tell them: "Come back in about 3 days and type /trial-reels results." Mention they can see the queue at https://my.blotato.com/scheduler.
## Job 2: Results (about 3 days after)
1. Read `trial-reels-log.md`. If it's not there, ask which trials to check.
2. `blotato_list_posts` (platform `instagram`, status `published`, since the scheduled date) to match each post id to its published reel, then `blotato_get_post_analytics` for each one.
3. **If Blotato has no metrics yet** (metrics come back null, or a trial isn't synced), don't guess. Ask them to open each trial in the Instagram app, check its insights, and paste the views, likes, shares and saves for each version.
4. If it's been less than 48 hours since the last trial went live, say the numbers are early and offer to check again tomorrow.
5. Rank them. **Views decide it, then shares and saves as the tiebreaker** (shares and saves are the strongest signal that strangers liked it, not just stopped on it). Show a table: version, hook, views, likes, shares, saves.
6. Call the winner, and say if it's close: if the top 2 are within about 15% of each other, it's basically a tie. Then they pick the one they like more.
7. **What to do with the winner:** open the winning trial in the Instagram app and share it with your followers from there. Or, if they'd rather post it fresh as a normal reel, schedule the winning file with `/schedule`.
8. One line on what the winner tells them for next time: which angle won (result, secret, mistake, number or callout). Add the results to `trial-reels-log.md`.
## What NOT to do
- Don't schedule trials for someone under 1,000 followers. Instagram won't run them.
- Don't write 5 hooks that are the same sentence reworded.
- Don't say the video was re-edited when it wasn't. Be clear about who adds the on-screen text.
- Don't pick a winner from made-up or estimated numbers. Real numbers, or ask them to paste.
- Don't share a trial with followers yourself. The user decides.
5. /comment-to-dm
Someone comments your keyword, they get a DM with your link. No ManyChat.
This is how you promote a digital product, a free guide or your newsletter. You tell Claude the keyword, the link and which post, and it builds the whole automation in Blotato. It shows you everything first and waits for you to say go.
You can make people follow you before the DM goes out, and you can collect their email on the way through. It also does two more jobs: check whether a live automation is actually working, and upgrade one to send emails to your list through a webhook.
What it needs
- Instagram Business or Creator account
- Blotato Paid plan, Instagram connected
---
name: comment-to-dm
description: Builds, runs, checks and upgrades Instagram and Facebook comment-to-DM automations through the Blotato MCP, replacing a separate ManyChat subscription. Use it to promote a digital product, a free resource or your newsletter. When someone comments a keyword they get a DM with a link, optionally after following the account and giving their email. Three jobs. Build a new keyword automation, Check whether a live one is working, and Upgrade one to capture emails through a webhook. Use when someone says "set up a comment to DM", "comment X and I'll send it", "build the DM automation for this reel", "is my DM automation working", "why isn't my automation sending", "capture emails from my DMs", or /comment-to-dm.
argument-hint: "[build | check | upgrade] [keyword] [link]"
allowed-tools: Read, AskUserQuestion, mcp__blotato__blotato_list_accounts, mcp__blotato__blotato_list_posts, mcp__blotato__blotato_list_automations, mcp__blotato__blotato_create_automation, mcp__blotato__blotato_update_automation, mcp__blotato__blotato_update_automation_trigger, mcp__blotato__blotato_get_automation_analytics, mcp__blotato__blotato_list_automation_runs, mcp__blotato__blotato_list_automation_logs
---
# Comment to DM
Someone comments your keyword. They get a DM with your link. Optionally they have to follow you first, and optionally they give you their email on the way through. That loop is the highest-converting free thing a creator can run, and Blotato runs it without a separate DM tool.
Use it for whatever you want people to get: a digital product, a free guide, a template, or your newsletter signup page. The link is the only thing that changes.
**Before any job:** call `blotato_list_accounts`. No Blotato tools means Blotato isn't connected to Claude yet (setup steps are in the `/schedule` skill, or https://help.blotato.com/start-with-an-ai-agent/mcp/setup). A 403 means their plan doesn't include the MCP: Claude can only use Blotato on a **paid** plan (https://blotato.com/?ref=zach, code `ZACH30` for 30% off). The Instagram account has to be a Business or Creator account.
| Job | When |
|---|---|
| **Build** | New post, new keyword, nothing exists yet |
| **Check** | It's live and they want to know if it's working, or it seems broken |
| **Upgrade** | It works, and now they want email addresses out of it |
If it's unclear which job they want, ask.
## How Blotato automations work (read this before any job)
- **Triggers.** A comment trigger fires when a comment contains one of the keywords. It watches either one post (`postId`) or every post (`postId` left out). An automation can also carry one DM trigger, so people who DM the keyword instead of commenting it get the same thing.
- **Scoping to one post needs a Blotato post id.** `postId` is the post's id from `blotato_list_posts`, which only lists posts published or scheduled through Blotato. A reel posted straight from the Instagram app won't be there, so it can't be scoped, and the automation has to run on every post. If scoping matters, publish the reel through Blotato.
- **Gates run before the DM.** A follow gate (Instagram only) holds the DM until they follow: they get the gate message, and once they've followed and tapped the button (or replied), the DM goes out. An email gate asks for an email first. People with an email already on file skip it, and a reply with no email in it gets the gate message again.
- **Webhook.** Fires after the DM is sent. Every method except GET sends a body of `{"email": ..., "isFollower": ...}`, each null when unknown. The URL has to be publicly reachable. Private, loopback and link-local addresses are rejected.
- **Limits.** DM and gate messages are 1 to 640 characters. Up to 3 link buttons, each label 20 characters or fewer, each an http(s) URL.
- **New automations start as inactive drafts** unless created with `isActive: true`. **Edits to an active automation go live the moment they're saved.**
- **Waiting windows.** A run waiting on a follow gate expires after 30 days. A run waiting on an email expires after 72 hours. Expired runs count as triggered, but never as completed or failed.
---
## Job 1: Build
### Step 1: Collect 4 things
Don't start building with a gap. Ask for whatever's missing.
1. **KEYWORD.** One word, the word they tell people to comment.
2. **LINK.** The URL the DM sends people to.
3. **THE FREE THING.** What they're sending, in one short phrase, the way they'd say it out loud.
4. **WHICH POST.** One specific post, or every post.
Defaults unless they say otherwise: follow gate on (Instagram), no email gate, one link button, DM trigger off.
### Step 2: Check the keyword before building anything
This is the step that quietly breaks launches.
1. **Existing automations.** Run `blotato_list_automations`. If an active automation on the same account already uses this keyword on every post, the two will fight over the same comments. The fix is scoping, not a new keyword: scope the new one to its post, and flag the old any-post one so they can scope it or pause it.
2. **Everyday words.** Matching is on "contains", so assume ordinary comments that include the word will fire it. On an any-post automation, warn plainly about short or common words (AI, ME, GO, LINK). Scoped to one post, a common word is usually fine, because anyone saying it on that post probably wants the thing.
3. **Case.** Pass the keyword in caps, lowercase and title case (for example `CONTENT`, `content`, `Content`). It costs nothing and removes a guess.
4. **Already filmed?** If the video already says the keyword, the keyword stays. Scope instead.
### Step 3: Find the post (only if scoping)
Run `blotato_list_posts` with platform `instagram` and status `published`. Match on the caption or the postUrl, and confirm the match with the user.
If the post isn't there, it was posted natively. Explain the Blotato post id rule and give them the two options: publish the reel through Blotato (the `/schedule` skill can), or run the automation on every post and accept the collision risk. Their call.
### Step 4: Show the build, then wait
Run `blotato_list_accounts` and name the account this is about to attach to. For Facebook, it also needs the `pageId` from that account's subaccounts, and it can't have a follow gate.
Then show the whole thing:
```
Account: [platform, @handle]
Name: KEYWORD - comment to DM
Trigger: comment-received
Keywords: KEYWORD, keyword, Keyword
Post: [the post, or every post]
Follow gate: ON
Message: "Yes! Follow me first and I'll send it straight over.
Tap the button below once you're following
(or just reply "done" if you don't see a button)."
Button: "I'm following"
DM message: "Here it is, THE FREE THING. Tap below and it's yours."
Link button: "Send it over" -> LINK
Active on create: yes
```
Write the DM like a person talking: what it is, then "tap below". Keep it short. Nobody reads a long DM.
**Wait for an explicit go. Don't create anything before they confirm.**
### Step 5: Create it, then hand back proof
Create it with `isActive: true`. Report the automation id, the account, the keywords, the scope, and that it's active.
Then tell them the thing everyone forgets: **your own comments never trigger it.** Test from a second Instagram account. People assume it's broken when it's working fine.
---
## Job 2: Check
1. **Find it.** `blotato_list_automations` with metrics `triggered`, `completed`, `failed`, `issues`. Match on name or keyword. If nothing matches, list them all and ask which one.
2. **Is it active?** An inactive draft never fires. That's the most common "it's broken."
3. **Triggered is 0?** Work through this in order: the automation is active; its trigger is active; it's scoped to the right post id; they didn't test from their own account; the keyword they typed matches.
4. **Triggered is high, completed is low?** Run `blotato_list_automation_runs` with status `waiting` and `expired`. **Waiting** means sitting at a gate (up to 30 days for a follow, 72 hours for an email). **Expired** means they never answered. **Superseded** means the same person triggered it again and the newer run took over. None of these are failures. Report them before calling anything broken.
5. **Failed above 0?** Run `blotato_list_automation_runs` with status `failed`, then `blotato_list_automation_logs` with that run's `flowRunId`. Quote the error and give the exact fix. `issues` is the failures from the last 7 days, which tells you whether the problem is current or old.
6. **Answer in plain English:**
- Is it live right now?
- How many times has it fired?
- How many people actually got the DM?
- How many are waiting at a gate, and how many gave up?
- If anything really failed: the reason, and the exact fix.
---
## Job 3: Upgrade to capture email
### Step 1: Get 2 things
1. **WHICH ONE.** The automation name or its keyword.
2. **WEBHOOK URL.** Their endpoint: Zapier, Make, n8n, or their own.
### Step 2: Say the part that matters, before building
Blotato saves the email to the contact, but **the webhook is the only thing that sends it on to their email list.** No working webhook means the email never reaches their list.
So before going live:
- Confirm the URL is publicly reachable. Private, loopback and link-local addresses get rejected.
- Use POST. A GET webhook sends no body, which means no email.
- Once it's live, check the first delivery in their tool's history (the Zapier, Make or n8n run log) before trusting it.
### Step 3: Add only 2 things
- **Email gate.** Message: "One thing first, what's the best email to send this to? I'll send it right after."
- **Webhook.** Method POST, URL = their webhook.
Keep everything else exactly as it is: same keyword, same scope, same follow gate, same DM, same button.
This automation is live, so the update publishes the moment it saves. **Show the full updated automation and wait for a go.**
### Step 4: Tell them what the person experiences
Comment, then the gate or gates, then the DM with the link. The webhook fires after the DM goes out. On the first test run, check the logs to see the exact gate order.
Say plainly that every gate costs some conversion. Two gates in front of a link is worth it when they want an email list. It isn't worth it for a pure follower push.
---
## What NOT to do
- **Don't create or edit anything before an explicit go.** Edits to an active automation go live instantly.
- **Don't skip the keyword check.** Two any-post automations on one keyword is the most common silent failure.
- **Don't put a link button and a plain-text URL in the same DM.** On desktop the URL stops being clickable, and those readers end up with no working link at all.
- **Don't call low completion broken** until you've counted waiting and expired runs.
- **Don't promise email capture without a working POST webhook.**
- **Don't change a keyword that's already been filmed.** Scope the automation instead.
- **Don't add a follow gate on Facebook.** The automation can't go live with one.
What you need
- Claude Code or the Claude desktop app
Claude Code runs all 5 and is where the one-prompt install works best. In the desktop app, use the Code tab. /analyze-competitors, /schedule, /trial-reels and /comment-to-dm also work in Cowork and claude.ai once uploaded. /edit is Claude Code on a Mac only.
- A paid Blotato plan
Blotato is what connects Claude to your social accounts. It powers /schedule, /trial-reels and /comment-to-dm. Claude can only talk to it on a paid plan, the free trial does not include that. Code
ZACH30takes 30% off. - An Instagram Business or Creator account
Personal accounts cannot connect. It takes 10 seconds to switch in the app under Settings, then Account type and tools.
- 1,000 followers for trial reels
Instagram only offers trial reels once you pass 1,000 followers. The other 4 codes work from day one.
- A Mac for /edit
About 5 GB free. The install prompt sets everything else up for you.
Use code ZACH30 for 30% off.
8 things that quietly break this
- 1. Blotato needs a paid plan for this
Claude talks to Blotato through its MCP connection, and that is not included in the free trial. Code ZACH30 takes 30% off. If Claude says 403 or "subscription", this is why.
- 2. Your own comments never trigger the DM
Test from a second Instagram account. Most people think the automation is broken when it is working fine.
- 3. One keyword, one automation per post
If the same keyword runs on two automations that watch every post, whichever fires first wins and the other silently never runs. Scope it to the post.
- 4. Posting from the Instagram app means no post scoping
To aim a comment-to-DM at one reel, Blotato needs that reel's Blotato post id, which only exists if you posted it through Blotato. That is one more reason to post with /schedule.
- 5. Instagram needs a Business or Creator account
Personal accounts cannot connect. Switch it in the Instagram app under Settings, then Account type and tools.
- 6. Trial reels need 1,000 followers
Under 1,000, Instagram does not offer trial reels at all. /trial-reels checks first and offers to schedule a normal reel instead.
- 7. /edit needs a Mac
It runs real video tools on your computer (ffmpeg, Whisper, Remotion), so it needs Claude Code on a Mac. It does not work on your phone, in the regular Claude chat app, or on Windows yet. The other 4 work anywhere Claude has Blotato connected.
- 8. If /schedule opens something else
Some versions of Claude Code have their own built-in /schedule for scheduling cloud agents. If you type it and Claude starts talking about routines, just say "schedule this video for tomorrow at 9am" and it picks the right skill.