tstack: Theo Browne's Open-Source Claude Code Skills, Explained
Theo Browne’s tstack bundles nine Claude Code skills for daily work—inside the PR-merge pipeline, the curation debate, and how to vet third-party skills safely.
Posted by
Related reading
Composio Instant: The Agent Tool Wallet, Revisited After Its Product Hunt Resurgence
Composio Instant gives agents a prepaid wallet and 1,500+ MCP tools—we revisit the Product Hunt resurgence, wallet-vs-virtual-card trade-offs, and spending guardrails.
Eleven Realtime vs OpenAI Realtime API: Inside ElevenLabs' All-in-One Speech-to-Speech Bet
ElevenLabs’ Eleven Realtime pitches one speech-to-speech model on an OpenAI-compatible API—what that means for latency, voice quality, and vendor switching.
T3 Code: The Open-Source Coding Agent Control Plane, and What Its Latest Update Changes
T3 Code turns scattered coding agents into one control plane—what the late-October update adds for remote iOS simulators, connect speed, and shared threads.
What skills are, in one paragraph
In Claude Code, a skill is a folder of instructions—and optionally scripts—that the agent loads on demand, conventionally a SKILL.md plus supporting files. Skills shape how the agent behaves; MCP servers add what it can connect to. Both are cheap to write and trivially easy to accumulate, which is exactly why curated collections get attention: most developers do not want to author their own review checklist, they want a decent one that works on day one. Our Claude Code MCP and tasks hub covers the complementary protocol side.
What tstack ships
tstack, published by Theo Browne (t3dotgg) on GitHub around October 24, 2026, is a curated collection of roughly nine skills aimed at daily workflows—review habits, debugging routines, and the PR-merge pipeline that drew the most attention. The repo reportedly picked up about 769 stars within days, and the exact skill list is still evolving, so read the README rather than secondhand threads before deciding what to adopt.
The PR-merge pipeline, step by step
- Run the test suite and linters, and stop the pipeline on the first red build.
- Group the working tree into conventional commits that each stand alone.
- Rewrite the PR description against the actual diff, not the original prompt.
- Apply reviewer feedback in follow-up passes, keeping the history reviewable.
The exact steps live in the repo and in X discussion, so treat them as subject to change. The value is less any single step than enforcing the same sequence every time—humans skip steps under deadline pressure, pipelines do not.
The curation debate
Reception split along a familiar line. Fans argued the defaults encode hard-won judgment and save hours of setup. Critics—including a plagiarism-controversy thread on X—questioned overlap with existing skill collections and whether the packaging credits earlier authors enough. Both observations can be true at once: curation is a contribution, and attribution still matters in open source. The practical takeaway is to judge each file on its own content rather than on the star count, and to keep your own local diff of anything you adopt so upstream changes never arrive unreviewed.
How to vet third-party skills safely
A skill is instructions injected into your agent’s context plus optional shell scripts that run on your machine. Before installing: read every file, pin to a specific commit, and diff updates instead of pulling main. Refuse any skill that phones home, asks the agent to paste environment variables somewhere, or chains shell commands you have not read. Layer enforcement underneath— tool-call guards for policy and secret redaction for leaked credentials. For a deeper method on evaluating agent harnesses themselves, see our Harness2 recursive harness breakdown.