Price terms as published

Each line is the vendor’s own wording, split from the price capture checked on 2026-10-28. Nothing here is rounded, converted or re-worded — the full sentences stay in the review below. Lines that carry only a price continue the plan named on the line above, in the vendor’s own order.

2 price terms for tstack
#	Plan / term	Checked
1	Free (MIT)	2026-10-28
2	requires GitHub CLI and your own model access	2026-10-28

One-liner: tstack is Theo Browne’s personal set of agent skills, cleaned up and published — the load-bearing part is a chain of skills that carries a change from draft to merged PR while you watch.

Evidence level: GitHub API and the repository README, both read 2026-10-28. No hands-on run of the skills; every behavioural claim below is the project’s own description.

What is in the collection

The repository collected 2,218 stars and 137 forks between its creation on 2026-10-24 and our check four days later, which makes it the fastest-moving skills collection in this week’s sweep. It is MIT licensed, mostly Python, and it works in Claude Code and Codex — with a caveat the README states plainly: some skills use T3 Code features when those are available and fall back to native tools when they are not.

Skill	Trigger	What it does
`full-send`	“full send”	Takes a change to a merged PR with nobody watching: proves it, reviews it, files it, babysits it, merges it
`full-stop`	“full stop”	Same pipeline, but stops when the PR is ready for you to merge
`council`	“council”	Gets review from three models on different providers and loops until they agree
`sol-loop`	“sol loop”	Reviews a change with GPT Sol in a loop until both agents agree
`file-pr`	“file a PR”	Files a short PR that leads with the problem, not an inventory of changes
`takeover`	“take over”	Finishes someone else’s PR, branch or agent thread, and keeps their credit
`babysit-pr`	“babysit”	Watches a PR through CI and review bots, fixes the real findings
`modernize`	“modernize”	Rebases an old PR and decides whether it should still exist
`leaving-pr-comments`	automatic	Formats comments an agent posts on your behalf

The interesting design decision is the dependency graph, which the README spells out rather than leaving to trial and error. full-send needs sol-loop, file-pr and babysit-pr; full-stop needs full-send; takeover needs modernize, file-pr, babysit-pr and leaving-pr-comments; and the three PR-shaped skills all need leaving-pr-comments. Copy one skill on its own and it will quietly do less than its name promises.

Install and first run

Two lines, then a dependency check:

git clone https://github.com/t3dotgg/tstack
cp -R tstack/skills/* ~/.claude/skills/   # Claude Code
cp -R tstack/skills/* ~/.codex/skills/    # Codex

Requirements, as stated: an authenticated GitHub CLI (gh), python3 for the pr-wait script inside babysit-pr, and — this is the one that trips people — council and sol-loop need subagents running on other providers’ models. T3 Code does that with delegate_task; elsewhere you need a harness that can reach those models. A Claude Code user with no second provider configured gets the PR plumbing but not the cross-provider review.

The README closes with a habit worth copying: keep personal settings — which models sit on your council — in CLAUDE.md or AGENTS.md rather than inside the skill, so skills can be updated without losing your configuration.

What it costs

The code is MIT: free to use, modify and redistribute. The real cost is downstream and comes in three parts. A GitHub CLI session with rights to file and merge in the repos you point it at — and full-send merging without supervision is exactly as consequential as it sounds. Model access for the review loops, which on council means paying three providers rather than one. And the time to read what each skill actually does before you say “full send” in a repository that matters.

The blast radius is the thing to price, not the licence. A skill that merges PRs unattended needs branch protection and CI that you trust more than the agent.

How it compares with other agent-skill collections

Most published skill collections are prompting conveniences: a folder of instructions that make an agent write better commits or tests. tstack is a workflow collection — the unit is a task with a definition of done (a merged PR, a rebase that survives review), and the skills compose into a chain with real dependencies between them. That makes it more useful out of the box and more dangerous to install piecemeal.

The council and sol-loop idea is the genuinely distinctive part: instead of one model reviewing its own output, several models from different providers are made to agree before the change proceeds. It is a different answer to “who reviews the agent” than the one everybody else ships, and it only works if you actually have those models wired up.

If your workflow is a single provider and a human reviewing every diff, you will use file-pr and babysit-pr and skip the rest — which is a perfectly reasonable way to adopt it.

Verdict

Adopt the two or three skills that match how you already work — babysit-pr and file-pr are useful even without the rest — and read the dependency list before copying a single folder. The star count reflects Theo’s audience as much as the code, so judge it on the graph and the requirements, not the counter. What it does well is rare: it takes the last mile of agent work, the part where a diff has to survive CI and review, and packages it as something an agent can be told to do.