Creator pack No. 02 · Creator

Theo Browne

From a finished change to a merged pull request, unattended.

Taken from Theo Browne's public repository. Not affiliated with or endorsed by Theo Browne.

Who it's forDevelopers who work through GitHub pull requests and want an agent to carry a change from done on your machine to merged. The nine skills review the diff with other models, file the pull request, sit through CI and review bots, and then merge or stop at the gate. You need the GitHub CLI signed in, and access to models from other providers for the review skills.
npx skills add t3dotgg/tstack@11e37a74d0a6868c121cd0910ab2b3bd9873b92a --skill full-send full-stop sol-loop council file-pr babysit-pr takeover modernize leaving-pr-comments -g -y

Fetches 9 skills from 1 author repo, each pinned to the version we reviewed. Nothing is mirrored. Uses the open-source skills CLI; needs Node.js 22.20 or newer.

What's inside

9 skills. Each card links to its author's repository at the exact version we reviewed.

  1. Full send 1/9 Autonomous shipping On the words "full send" the agent finishes a change alone: proves it works, gets a second-model review, files the pull request, handles CI and review bots, then merges. @t3dotgg MIT

    Full send

    The centre of the set: one phrase hands over everything from proof to merge. Read what it permits before you say it.

    Community · Theo Browne MIT
    Source t3dotgg/tstack at 11e37a7
  2. Full stop 2/9 Autonomous shipping Does the same unattended work as full-send but never merges: it switches off auto-merge first and stops with the pull request open, checked and ready for you. @t3dotgg MIT

    Full stop

    The same run without the merge, for repositories where you want the last click yourself.

    Community · Theo Browne MIT
    Source t3dotgg/tstack at 11e37a7
  3. Sol loop 3/9 Second-model review Has one OpenAI GPT Sol reviewer check your diff, three rounds by default, with a ledger of findings and a brief against scope creep. It stops when both agree or the budget ends. @t3dotgg MIT

    Sol loop

    The review step full-send relies on: a second model reads the diff before anything is filed.

    Community · Theo Browne MIT
    Source t3dotgg/tstack at 11e37a7
  4. Council 4/9 Multi-model review Asks three strong models from three different providers to review the work in parallel, scores each round out of 10 and repeats after fixes until all three are satisfied. @t3dotgg MIT

    Council

    For changes that deserve more than one opinion: models from three providers review and all must agree.

    Community · Theo Browne MIT
    Source t3dotgg/tstack at 11e37a7
  5. File PR 5/9 Pull requests Opens a pull request whose title says why the change matters and whose description starts with the problem, in the repository's own title style, after a check for secrets. @t3dotgg MIT

    File PR

    Short pull requests that open with the problem. full-send and takeover file through it.

    Community · Theo Browne MIT
    Source t3dotgg/tstack at 11e37a7
  6. Babysit PR 6/9 PR monitoring Watches a pull request through CI and review bots with a bundled watcher script, fixes real findings, answers false positives with a written reason and keeps the branch current. @t3dotgg MIT

    Babysit PR

    Waits out CI and review bots so you do not have to, and fixes what is real.

    Community · Theo Browne MIT
    Source t3dotgg/tstack at 11e37a7
  7. Takeover 7/9 Continuing work Finishes someone else's pull request, branch or agent thread: imports their commits, adds fixes separately, keeps author and co-author credit and re-checks the original problem. @t3dotgg MIT

    Takeover

    Picks up a stalled pull request or another agent's thread and keeps the original author's credit.

    Community · Theo Browne MIT
    Source t3dotgg/tstack at 11e37a7
  8. Modernize 8/9 Stale pull requests Brings an old pull request up to date and first asks whether it should still exist: it answers MODERNIZE, REWORK or CLOSE with evidence before any rebase or push. @t3dotgg MIT

    Modernize

    Decides whether an old pull request is still worth having before it spends effort on a rebase.

    Community · Theo Browne MIT
    Source t3dotgg/tstack at 11e37a7
  9. Leaving PR comments 9/9 Comment etiquette Ready Formats every GitHub comment the agent posts for you: a note on top names the model and says it answers on your behalf, then a short reply that leads with the conclusion. @t3dotgg MIT

    Leaving PR comments

    Makes every comment the agent posts say that an agent wrote it. Four of the other skills call it.

    Works right away Community · Theo Browne MIT
    Source t3dotgg/tstack at 11e37a7

Good to know

What you bring

  • GitHub CLI (gh), signed in
  • The sol-loop, file-pr and babysit-pr skills from the same repository
  • Access to an OpenAI model as a subagent (for sol-loop)
  • The full-send skill and the skills it uses
  • Access to OpenAI's GPT Sol model (paid)
  • A harness that can run it as a subagent, such as T3 Code
  • Access to models from three providers (paid)
  • A harness that can run them as subagents, such as T3 Code
  • The leaving-pr-comments skill for any follow-up comment
  • python3 on macOS or Linux for the pr-wait script
  • The leaving-pr-comments skill
  • The modernize, file-pr, babysit-pr and leaving-pr-comments skills
  • The leaving-pr-comments skill for any comment

Pick one, not both

  • Dependent set: this is the whole repository, kept together because the skills call each other. full-send uses sol-loop, file-pr and babysit-pr; full-stop uses full-send; takeover uses modernize, file-pr, babysit-pr and leaving-pr-comments.
  • Merge or stop: full-send merges on its own and full-stop never does. Install both and choose per task by the phrase you use.
  • Review: sol-loop asks one OpenAI model and council asks three providers. Both need a harness that can run other providers as subagents; the skills name T3 Code for that. Without such access, leave council out and expect full-send to stop at its review step.
  • Platform: the watcher script in babysit-pr needs python3 and does not run on plain Windows.

Do not combine with

These collections steer the same part of your agent's work and get in each other's way.

Review slipCreator pack No. 02
  • Every skill file read in full
  • Pinned to a reviewed version
  • Author and licence on every card
  • Requirements and costs disclosed
  • No sponsored skills inside
Last reviewed
Nov 2026
Reviewed