# Theo said "I give up" and released his 9 agent skills, gaining 2000 stars in two days · GitHub open-source project

> Source: https://stringers.io/en/github/gh_t3dotgg__tstack.html (Stringers / 什么值得做, generated 2026-11-08)
> Stringers answers one question every day: what small product can one or two people build in a few months and actually make money from. Numbers come from 15 databases we collect daily, each with source and date.
> For agents: these verdicts and the underlying databases (verified revenue, App Store / Google Play charts, ad libraries, GitHub, Product Hunt…) are queryable live over MCP — tools today / hot / products / directions / direction / data. Setup: https://github.com/komako-workshop/stringers (the user logs in at stringers.io → "配置 agent" to get a personal URL). Site index: https://stringers.io/llms.txt

> Theo teased in a livestream that "32 times, 27 of them auto-merged," dangling it in front of viewers for over two months. 10-26 then posted it as a 230k-view video and a 500k-view tweet. Almost all the stars came from his own audience.

[GitHub open-source project](https://stringers.io/en/github/index.md) · written 11-6

t3dotgg/tstack · t3dotgg

# Theo said "I give up" and released his 9 agent skills, gaining 2000 stars in two days

**What it is** The 9 agent skills YouTuber Theo uses every day. Say "full send" and the agent proves the change itself, has another model review it, opens a PR, watches CI, and merges.

**Why it blew up** Theo had been teasing "32 times, 27 auto-merged" on stream for over two months, then dropped it with a 230k-view video and a 500k-view tweet on 10-26. Most of the stars came from his own audience.

**C: Copy the approach, not the stack.** **What you can take from it** Don't build an "XX-stack". Take three things from it: cut out one narrow stage of a big package that nobody has done well, ship it publicly and publish the results before open-sourcing it, and give each skill a trigger phrase people can say out loud. Then push it through your own content and big accounts that repost.

C grade: the launch ran on Theo's own followers, which other people don't have. Personal skill packs at the top of the category already have 100K to 300K stars, while similar repos without a famous author sat at 861 stars after over a month. The playbook is worth copying, but copying the package isn't.Word in small circles

Data (updated daily) · Repo created 2026-10-24 · 3,231 stars · 7 days +706

GitHub · stars

Data 2026-10-25 → 2026-11-08 (15 days, 15 observations). Points before 10-28 come from deeply / archive backfill, not every day has one

t3dotgg/tstack is a set of Claude Code / Codex skills published by Theo Browne, the YouTuber and author of T3 Chat and T3 Code. It covers one job: after the code is written, it makes the agent prove the change works, gets a review from a different model, opens the PR, watches CI, and merges it itself. It reached 3074 stars and 224 forks by day 13 of the repo, and 11-06. Almost all of those 3000 stars came from one of Theo's videos and one of his tweets. Compared with the "famous person's personal skill pack" group, it's a small account: only 2% of garrytan/gstack. Below: what it is, where the stars came from, and what you can take from it if you don't have Theo's audience.

### "I caved"

2026 Oct, 26 UTC, around 7:16 in the morning, Theo posted a YouTube video called "These 3 skills changed my life" and mentioned in the description "Oh also I did make tstack", with the repo link. 26 minutes later he posted on X:

> "So uh...I guess it's time to announce tstack? I caved. I am publishing my skills now."

"I caved" has a backstory. In August he posted "My AGENTS.md & SKILLS.md Breakdown (Don't copy them)", which got 128k views, and the title literally told everyone not to copy them. In Oct 14 he ran a 5-hour livestream on how his workflow has changed since, and gave one number: 32 full sends, 27 of which were merged without anyone looking. Viewers had been kept waiting for over two months, so this post is basically "fine, here you go."

As of 11-06, the tweet had 498K views, 207 replies, 187 reposts, 2.6K likes, and 2.3K bookmarks (matched to the X page in order, inferred). Bookmarks were nearly as high as likes, meaning many people saved it to install later. The video had 11-06 views by 232 thousand.

### What it is

An MIT-licensed repo with 9 skills. Each skill's name is the phrase you say to the agent (from the README):

| Say | What it does |
|---|---|
| "full send" | Takes the current changes all the way to a merged PR with no one watching: proof, review, PR, monitoring, merge |
| "full stop" | Same as above, but stops at a PR ready to merge and waits for you to press the button |
| "council" | Gets reviews from three models from different vendors and loops until they agree |
| "sol loop" | Runs a review loop with GPT Sol until both agents agree |
| "take over" | Picks up someone else's half-finished PR, branch, or agent conversation and keeps the original author's credit |
| "babysit" | Watches a PR through CI and review bots and fixes the real problems |

There are also file-pr (writes a short PR that explains the problem and doesn't list every change), modernize (rebases old PRs and decides whether they should still exist), and leaving-pr-comments (the format for when an agent leaves comments for you). Install is a git clone plus a cp into ~/.claude/skills or ~/.codex/skills.

The full-send SKILL.md is blunt: when the user says "full send", they are not watching this conversation, and you are responsible for the thing through PR merge. The phrase itself is the authorization. You can merge on your own, force-push your own branch, and run the tests needed for proof without asking. But changing production or deleting data still requires asking. Merging has gates: every change has proof, checks are all green, review bots have finished, and nobody has asked for changes.

### How the stars came

| Date | Stars that day | What happened |
|---|---|---|
| 10-24 | — | Repo created, first commit "publish everyday agent skills" |
| 10-25 / 10-26 | 22 / 47 | Not promoted yet |
| 10-26 | — | Video (232K views) and tweet (498K views) went out back to back |
| 10-27 | 1177(+1130) | Peak of the whole run; that day 0 xMarioNawfal (@RoundtableSpace) reposted it, 48K views |
| 10-28 | 2079(+902) | Afterglow |
| 10-29 ~ 11-04 | 2280 → 2673 | Daily +47 to +200, then fading |
| 11-05 / 11-06 | 2929 / 3074 | Went up again +256 / +145, cause not found |

(Daily GitHub star counts; video and tweet numbers taken from the pages, 11-06.)

In the first two days it added up to under 50 stars. Videos and one tweet did the work: over 48 hours it gained 2000 stars. Roughly 700k impressions turned into 2000 stars, about a 0.3% conversion rate (estimate). Then reposts came in: 0 a tweet from @MarioNawfal, a YouTube explainer with 676 views, a few Shorts and Instagram Reels, an install tutorial on explainx.ai, and a listing on skillleaderboard. No HN threads, no dedicated Reddit discussion, no GitHub Trending placement, no media coverage.

One detail shows people are actually using it: in 13 days there were a dozen-plus external issues and PRs covering Windows compatibility, an install path for the Claude Code plugin marketplace, a full-send-fast variant, and removing Gemini. Theo's own commits stopped at 10-28.

### Why this one, and why only 3000 stars

**The pain point they picked is real.** Most skill packs cover writing code and planning. tstack covers what happens after the code is written: proof, review, CI, merging, and picking up someone else's branch. In the second half of 2026, the people who were heavy agent users all hit the same wall: the agent writes code fast, and the PRs pile up with nobody merging them. 0 Under MarioNawfal's post, someone replied: "Takeover is the one I'd try first. Half-done branches are where teams quietly stall."

**The names are the trigger.** "full send", "council", "babysit": say them and the skill fires. They're easy to remember and easy to pass along in a tweet.

**But reach depended little on the content.** sno-ai/sno-station-skills was created in Oct 1 with almost the same pitch: cross-vendor review, evidence-backed tickets, and autonomous loops that run overnight. It had no famous author, and by 11-06 it had 861 stars, a bit over a quarter of tstack's.

**And it came late.** The "famous person publishes their skill directory" format had already been taken by others, who got most of the reach:

| Project | Launched | Stars | Compared with tstack |
|---|---|---|---|
| mattpocock/skills | 2026-02 | 316k | Also a famous person's personal directory covering the whole engineering workflow, but 9 months earlier |
| obra/superpowers | 2025-10 | 313k | A full methodology: planning, TDD, subagents |
| anthropics/skills | 2025-09 | 188k | Official examples and format standard |
| garrytan/gstack | 2026-03 | 140k | 23 "role" skills; the name tstack clearly echoes it |
| vercel-labs/agent-skills / openai/skills | 2025-12 / 2025-11 | 34k / 29k | Big-company official directory |
| EveryInc/compound-engineering-plugin | 2025-10 | 26k | Company-made, also includes review and multi-agent |
| sno-ai/sno-station-skills | 2026-10 | 861 | Closest pitch, no famous author |

(GitHub, 11-06.)

Theo has 576K YouTube subscribers, yet arriving seven to nine months late he got only 2% of gstack and 1% of mattpocock. This category was already crowded by October, and celebrity effect only bought a small tail.

### If it were you

**C: Learn from it, but following it doesn't pay.** tstack's 3000 stars came from Theo's audience, and you can't take that. The category it sits in, personal skill packs, is topped by repos with 100K to 300K stars, and late-arriving celebrity packs only get scraps.

**Who it fits** : developer YouTubers or X accounts that already have tens of thousands of followers and want to publish a workflow they already use to build reputation; or people who have gotten results from a specific agent workflow that others haven't, and who can back it up with numbers.

**Who it doesn't fit** : people with no distribution of their own who hope a "generic skill pack" will earn a few thousand stars. Compare sno-station-skills: similar content, no famous author, 861 stars after over a month. Another auto-merge skill pack would most likely do even worse.

**How to do it** : don't copy the "XX-stack" format. Three things are worth taking:

1. **Cut out one narrow stage** : don't build "the whole engineering team", only one stage that the big packages haven't handled well. The PR stage is already taken by tstack, sno-station, and compound-engineering, so look narrower and more vertical: database migration validation and rollback, mobile release and app review, dependency upgrades, production incident investigation.

2. **Use it publicly and show results before you release it** : Theo said "32 full sends, 27 auto-merged" on stream, then released it two months later. You can show real before-and-after comparisons on X or in videos first, and open-source once people ask.

3. **The name is the trigger** : give each skill a trigger phrase people can say out loud, and put a "say / what it does" table on the first screen of the README so someone can screenshot the whole thing and share it.

**How to get users** : your own videos and tweets are the main channel. Next, send the repo to English-language accounts like 0 xMarioNawfal that repost new open-source projects, and to skill directories like skillleaderboard and explainx, which write up tutorials afterward.

**When to move** : narrow-domain skills still have room, but official big-company directories and a few top packs keep adding to their lists. Estimate: before 2027 H1, the openings you can slip into will keep shrinking.

**What would show you misread it** : ① if a skill pack in a narrow domain gets reposted by an account with tens of thousands of followers and still doesn't reach 200 stars within a week, this kind of thing is no longer fresh; ② conversely, if a skill pack with no celebrity backing passes a few thousand stars on content alone, then "it depends on the author's reach" was too pessimistic, and the narrow-stage route is worth taking earlier.

[GitHub repo ↗](https://github.com/t3dotgg/tstack)

**Research notes: competitor table, acquisition evidence, full research report**

Competitors

| Name | Where | Launched | Scale | How it differs from this one |
|---|---|---|---|---|
| mattpocock/skills | Open source | 2026-02-03 | 316K stars | Also a famous person's personal skill directory covering the whole engineering workflow, but 9 months earlier; tstack only covers the back half of the PR process |
| obra/superpowers | Open source | 2025-10-09 | 313K stars | A full dev-methodology framework (planning / TDD / subagents), not one person's PR workflow |
| anthropics/skills | Big-company features | 2025-09-22 | 188K stars | Official examples and format standard, no personal workflow opinions |
| garrytan/gstack | Open source | 2026-03-11 | 140K stars | 23 role skills covering the whole sprint cycle; the name tstack clearly echoes it, with a narrower scope |
| EveryInc/compound-engineering-plugin | Open source | 2025-10-09 | 26K stars | Company-made compound engineering plugin, methodology-focused |
| openai/skills / vercel-labs/agent-skills | Big-company features | 2025-11 / 2025-12 | 29k / 34k stars | Official skills catalog, doesn't auto-merge PRs |
| sno-ai/sno-station-skills | Open source | 2026-10-01 | 861 stars | The pitch is similar (cross-vendor peer review, evidence-backed tickets), but no celebrity backing, and only 1/4 stars vs tstack's |

Where users come from

- **Verified** **The author's own YouTube** : 10-26 07:16 UTC video "These 3 skills changed my life"; the description links the repo; 232k views. The star spike lands the same day

- **Verified** **The author's own X** : 10-26 07:42 UTC tweet "I guess it's time to announce tstack? I caved", about 498k views, 2.6K / 2.3K engagements

- **Verified** **Earlier streams and video teasers** : on 10-14 he streamed the full-send approach; 32 times full-send, 27 auto-merged. In an August video he said "don't copy me"; this "I caved" answered viewer demand

- **Verified** **Aggregators and reposts** : explainx.ai tutorial, skillleaderboard, Alex Hitt explainer video (676 views), Shorts / IG Reel, Grok summary

- Inferred **HN / Reddit** : HN Algolia finds nothing; no dedicated Reddit threads

## t3dotgg/tstack: Theo's personal agent skill pack, scaled by one video and one tweet

> Channel: GitHub | data card 2026-11-06:3074 stars, 224 forks, repo age 13 days | research day 2026-11-06

### One line

tstack is a set of 9 Claude Code / Codex skills released by Theo Browne (@theo, 576K YouTube subscribers), YouTuber and CEO of T3 Code / T3 Chat. The core idea is letting the agent take a change all the way to a merged PR on its own (full-send / full-stop / sol-loop / council / babysit-pr / takeover...). The 3000 stars came almost entirely from one of Theo's videos and one tweet; the product itself has no built-in growth. Compared with other personal skill packs, it's the **small one** : the similar mattpocock/skills has 316K stars, garrytan/gstack has 140K, and tstack has only 1%–2% of those.

### 1. Why it took off

#### Timeline (star counts from the data card daily, events from the web)

| Time (UTC) | Event | Stars |
|---|---|---|
| 2026-08-11 | Theo posted the video "My AGENTS.md & SKILLS.md Breakdown (Don't copy them)", 128k views, explicitly saying "don't copy mine" | — |
| 2026-10-14 | 5.6-hour stream "So my workflow has changed a lot..." (51k views) covering the soul loop / full send / takeover. He said 32 full-sends, 27 of which merged without anyone watching | — |
| 10-24 05:00 | Repo created, first commit "publish everyday agent skills" | — |
| 10-25 / 10-26 | Not promoted yet | 22 / 47 |
| **10-26 07:16** | Video "These 3 skills changed my life"; the description reads "Oh also I did make tstack... github.com/t3dotgg/tstack". As of 11-06 it has **232k views** |  |
| **10-26 07:42** | Tweet "So uh...I guess it's time to announce tstack? I caved. I am publishing my skills now." (with video), roughly **498k views, 2.6K reposts, 2.3K likes** ; his next reply posted the repo link and plugged his course mega |  |
| 10-27 | Single-day +1130 (highest of the whole period) | 1177 |
| 10-28 | +902 | 2079 |
| 10-29 ~ 11-04 | Daily +47 to +200, then natural decline | 2280 → 2673 |
| 11-05 / 11-06 | Another bump, +256 / +145; cause not found | 2929 / 3074 |

Sources: YouTube video pages (view counts and publish dates from the page JSON, fetched 2026-11-06); x.com/theo/status/2114623574371963360 (read via r.jina.ai); GitHub API commit history; data card daily.

#### Where users come from

- **Theo's own YouTube and X (confirmed)** : the spike day was the day the video and tweet went out; the two days before it total under 50 stars. One video (230k views) plus one tweet (500k views) produced about 2000 stars in the first 48 hours, a conversion of roughly 0.3%.

- **Secondhand reposts and aggregators (confirmed, small)** : someone made a YouTube explainer (Alex Hitt, 676 views), YouTube Shorts, and an Instagram Reel; explainx.ai wrote an install guide; skillleaderboard.com indexes it; Grok summarized it on X. These are the long tail that followed, not the cause of the spike.

- **HN / Reddit (not found)** : HN Algolia returns no tstack posts; the Reddit results are old threads criticizing Theo, none discussing tstack specifically.

- No ads, and no GitHub trending (the repo was under 60 days old, trending_days_any=0).

#### Why people starred / stayed

- **Viewers were primed first** : 10-14's stream already covered this workflow in detail and gave numbers (32 full-sends, 27 auto-merged). The August video had also said "don't copy me", so "I caved" became a moment answering viewer requests; people got what they had been waiting for.

- **The positioning is sharp** : most skill packs cover "writing code and planning". tstack covers the back half of the PR: proving the change works, cross-model review, watching CI and review bots, merging, and taking over others' PRs. That is the pain point for heavy agent users in the second half of 2026: agents write code fast, and the bottleneck is review and merge.

- **A real maintainer** : in 13 days there were a dozen-plus external issues / PRs (Windows compatibility, a Claude Code plugin marketplace version, a full-send-fast variant, removing Gemini, and more). Theo's own commits stopped at 10-28.

### 2. Competitors and differentiation

| Product | Type | Launched | Scale (GitHub API, 2026-11-06) | How tstack differs |
|---|---|---|---|---|
| mattpocock/skills | Open source, a celebrity's personal skill pack | 2026-02-03 | 316K stars | Also a celebrity's personal .agents directory, but covers the whole engineering flow, 9 months earlier |
| obra/superpowers | Open source, a skills framework and methodology | 2025-10-09 | 313K stars | A full dev methodology (planning, TDD, subagents); tstack covers only the back half of the PR |
| anthropics/skills | Big-company official | 2025-09-22 | 188K stars | Official examples and format standard, no personal workflow |
| garrytan/gstack | Open source, a celebrity's personal skill pack | 2026-03-11 | 140K stars | 23 "role" skills (CEO / design / QA...) covering the whole sprint cycle; the name tstack clearly echoes it |
| wshobson/agents | Open-source collection | 2025-07-24 | 42K stars | Big and broad, multi-platform |
| vercel-labs/agent-skills | Big-company official | 2025-12-08 | 34K stars | Leans Vercel / frontend |
| openai/skills | Big-company official | 2025-11-25 | 29K stars | Official Codex skills catalog |
| EveryInc/compound-engineering-plugin | Company-made plugin | 2025-10-09 | 26K stars | "Compound engineering" methodology with review and multi-agent support |
| sno-ai/sno-station-skills | Small-team open source | 2026-10-01 | 861 stars | The pitch is similar (cross-vendor peer review, evidence-backed tickets, overnight loops), but no celebrity backing, and only 1/4 stars vs tstack's |

**Differentiation** : (1) it covers only the PR lifecycle, end to end from "prove → review → open PR → watch CI → merge", and writes the permission for the agent to merge itself into the skill, with merge gates (proof covers every change, checks all green, review bots finished, no one requested changes); (2) cross-vendor model peer review (council loops three models until they agree; sol-loop uses GPT Sol as reviewer); (3) integration with his own T3 Code (delegate_task), which also drives traffic to his product.

**Whether differentiation caused the growth** : largely not. The growth followed Theo's release schedule entirely. Content-wise it is close to sno-station-skills, which has only 861 stars. Differentiation at most helped retain some heavy users and brought outside PRs. The ceiling (3000 stars) is far below gstack and mattpocock, which suggests the "celebrity personal skill pack" niche was saturated by 2026 October: even with celebrity effect, arriving 7–9 months later got 1%–2% of the volume.

### 3. Visibility and copycats

- **Visibility: circulating in small circles.** Theo's tweet and video drew roughly 700k impressions combined, all inside his own audience. No media coverage, no HN / Reddit discussion threads; only SEO aggregators like explainx and skillleaderboard, plus scattered reposts.

- **Copycats: almost none direct.** There are some small "personal xxx-stack skill" repos (e.g. GeneralVoritzio/mort-stack with 2 stars, teelyjc/skills with 1 stars), but nothing shows they copied tstack; they look more like following gstack's naming trend. full-send-fast emerged in the community as a variant proposed in a PR.

### 4. What to borrow

- **Use it publicly first, share the numbers, then release it** : he showed "32 full-sends, 27 auto-merged" in a stream and video, released once viewers asked, and used "I caved" to create a moment of being pulled in.

- **Cut a narrow stage in a crowded category** : not "a whole engineering team", only "nobody handles it after the PR merges". The skill names are the trigger phrases (saying "full send" triggers it), which makes them easy to remember and spread.

- **You need your own distribution channel** : star counts for this kind of repo track the author's follower count almost directly. Without an audience, the same content gets only a few hundred stars (compare sno-station-skills).

### 5. Still unclear

- What drove the 11-05 bump of +256 is unverified (Theo posted a PewDiePie video that day; whether it mentioned tstack is unchecked; it could also be an aggregator or a secondary repost).

- The local X corpus (x-radar) has no relevant tweets, so KOL reposts were not verified. The tweet engagement numbers come from the page as read via Jina; which number is reposts vs likes is inferred from page order.

- Actual install counts are unknown (clone or npx skills data is not visible).

### Sources

- [github.com/t3dotgg/tstack](https://github.com/t3dotgg/tstack) (README, commits, issues / PRs, GitHub API 2026-11-06)

- [x.com/theo/status/2114623574371963360](https://x.com/theo/status/2114623574371963360), [x.com/theo/status/2114623745486958945](https://x.com/theo/status/2114623745486958945)

- [www.youtube.com/watch?v=5Bmp9cz1JN8](https://www.youtube.com/watch?v=5Bmp9cz1JN8) (232k views), [www.youtube.com/watch?v=SY_woOjzlW4](https://www.youtube.com/watch?v=SY_woOjzlW4), [www.youtube.com/watch?v=e1snsuY4lTI](https://www.youtube.com/watch?v=e1snsuY4lTI)

- [finance.biggo.com/podcast/8e41e2f0b5a7a374](https://finance.biggo.com/podcast/8e41e2f0b5a7a374), [finance.biggo.com/podcast/edd2a2fc77a07445](https://finance.biggo.com/podcast/edd2a2fc77a07445) (video summaries)

- [explainx.ai/blog/tstack-t3dotgg-agent-skills-full-send-counc…](https://explainx.ai/blog/tstack-t3dotgg-agent-skills-full-send-council-claude-code-codex-guide-2026), [www.skillleaderboard.com/skill/t3dotgg-tstack](https://www.skillleaderboard.com/skill/t3dotgg-tstack)

- Competitor stars: GitHub API 2026-11-06; same-period skill library: github repo_profile(2026-11-06)

This article was written by AI from public data and the data we collect daily. Numbers are sourced, and inferences are labeled.