Batch-generate a week of strategic tweets in ~15 minutes. Power-law-optimized content for indie devs building in public.
A single-file Claude Code / Cursor / Codex skill that turns your Sunday planning session into a full week of ready-to-post tweets — with automatic performance tracking that gets smarter over time.
You know you should tweet. You know build-in-public works. But every day you stare at the compose box thinking "what do I even post?"
So you either don't post, or you post something generic that gets 3 likes.
One skill file. One command on Sunday. 18 tweets for the week.
The skill uses power law content theory (80% volume tweets + 20% high-effort "power bets") combined with proven hook formulas from marketing research to generate tweets that are strategically allocated across your week.
It also tracks your performance over time and auto-adjusts — if your data tweets consistently outperform your hot takes, next week's batch shifts accordingly.
# Claude Code
mkdir -p .claude/skills/weekly-tweet-engine
cp SKILL.md .claude/skills/weekly-tweet-engine/SKILL.md
# Cursor / Codex
mkdir -p .agents/skills/weekly-tweet-engine
cp SKILL.md .agents/skills/weekly-tweet-engine/SKILL.mdPaste this into Claude Code or Cursor:
Read the skill at .claude/skills/weekly-tweet-engine/SKILL.md.
Then ask me a few questions to create my tweet profile at tweets/profile.md.
Keep it short — just my projects, voice, and schedule.
The skill asks ~8 questions and creates a 40-line profile. Takes 2 minutes.
Every Sunday, paste this:
Read tweets/profile.md and the skill at .claude/skills/weekly-tweet-engine/SKILL.md.
Run the weekly review first (ask me about last week's performance).
Then generate my tweets for this week:
- This week I [what happened]
- Trends I spotted: [any trends, or "none, find some"]
- Focus project: [which project gets the power bets]
Output: A tweets/week-YYYY-MM-DD.md file with your entire week of tweets, organized by day, tagged by type and hook formula.
| Type | ~% | What it is |
|---|---|---|
| Quick tweets | 60% | BIP updates, tips, reactions — your "lottery tickets" |
| Engagement | 15% | Questions, polls, hot takes that drive replies |
| Power bets | 20% | Threads, case studies, data stories — your outlier candidates |
| Promo | 5% | Product mentions always wrapped in value |
### Wednesday AM — power-bet (thread)
> The "10 Game" is going viral on TikTok right now.
> 2 players count 1-10, replacing numbers with words each round.
> There's no app for it.
> So I'm building one this weekend.
> Here's the concept 🧵
>
> #buildinpublic #indiedev
Hook: Curiosity + Story · Pillar: Viral trends · Project: 1$forYourNewAppFor power bet threads: full thread structure + visual suggestions + double-down plan if it performs.
The skill has 7 hook families built in:
| Hook | Best for | Example opener |
|---|---|---|
| Curiosity | Power bets, threads | "I found a niche with 2M searches and 3 apps..." |
| Story | BIP updates, threads | "Last Tuesday I spotted a trend at 11pm..." |
| Value | Quick tweets, educational | "How to get 1,000 downloads without ads:" |
| Contrarian | Engagement, hot takes | "Unpopular opinion: 6 months on your app is wrong" |
| Data | Power bets, credibility | "I analyzed 500 App Store listings..." |
| Question | Engagement drivers | "Do you spend more time on code or metadata?" |
| Statement | Quick tweets, punchy | "Your subtitle is your most underrated asset." |
Plus a banned words list to avoid AI slop: "game-changer", "buckle up", "let that sink in", "unlock", "masterclass", etc.
The skill creates and maintains tweets/log.md automatically.
Every Sunday before generating new tweets, it asks you 4 quick questions:
- Best tweet this week? (paste or describe)
- Rough numbers? (impressions/likes, or "did well"/"flopped")
- Follower count now?
- Anything that surprised you?
It logs the answers, analyzes patterns, and uses them to adjust next week's batch:
- 3+ weeks of data → auto-adjusts pillar percentages
- Identifies your winning hook type
- Flags stale topics
- Suggests experiments ("You haven't tried a Question hook in 3 weeks")
You never edit log.md manually. It maintains itself.
When a tweet performs unusually well, the skill includes a ready-made escalation plan:
Within 4h → Reply with additional context
Within 24h → Follow-up thread going deeper
Within 24h → Cross-post to Reddit (skill suggests specific subreddits)
Within 48h → Pin the follow-up
Within 1 week → Turn into blog post / README / product feature
Within 1 week → Start a numbered series
Trigger it anytime with:
My tweet about [topic] got [X impressions]. Run the outlier protocol.
After a few weeks of use, your project looks like this:
tweets/
├── profile.md ← your voice, projects, schedule (auto-created)
├── log.md ← performance tracking (auto-managed)
├── week-2026-03-24.md ← batch for this week
├── week-2026-03-31.md ← batch for next week
└── ...
One skill file in .claude/skills/. Everything else is generated.
Read tweets/profile.md. Write 3 tweets about [topic/trend/milestone].
Use the hook formulas from the tweet engine skill.
Read tweets/profile.md. My tweet about [topic] got [X impressions / Y likes].
Run the outlier protocol: generate follow-up thread + Reddit cross-post + series plan.
Read tweets/profile.md. I'm now focusing on [new project] instead of [old project].
Update my profile and adjust the content pillars.
- coreyhaines31/marketingskills — The
social-contentandcopywritingskills complement this one. This skill handles the generation pipeline; those handle platform strategy and copy polish. - rgranet/designlint — If you're tweeting about shipping UI, DesignLint makes sure it doesn't look like AI slop before you screenshot it.
This skill is built on two frameworks:
Power law content strategy — In content, 1% of posts generate ~85% of total reach. You can't predict which tweet will be the outlier, so you need volume (lottery tickets) + quality (power bets) to let the math work. The 80/20 split is calibrated for this.
Corey Haines' social-content framework — Hook formulas, content pillars, engagement routines, and repurposing workflows adapted from the marketingskills repo.
The combination: you post enough to buy lottery tickets, but each ticket is optimized with proven hooks and backed by your personal performance data.
| Agent | Status |
|---|---|
| Claude Code | ✅ Tested |
| Cursor | ✅ Compatible |
| Codex | ✅ Compatible |
| Gemini CLI | 🔄 Should work (untested) |
Found a hook formula that works? A better way to structure the weekly review? PRs welcome.
MIT
Built by Ruben Granet · Part of the 1$forYourNewApp ecosystem