agentleFS
DeepusleepyDeepusleepy/t3-tots/skills/t3-pm-thread/SKILL.md

t3-pm-thread

Skill110 starsChanged 17 days agoMITView on GitHub

Set up and run a project manager (PM) thread in T3 that plans the work, pushes back, starts worker and reviewer threads, and checks results before calling anything done. Use for "make a PM thread", "run <project> for me", "manage this project in T3", "set up workers and reviewers", or when a goal spans several PRs or threads.

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 1 is distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read what exists: the project's README and docs, open PRs (t3ctl prs --project '<P>'), recent threads (t3ctl threads --project '<P>' --main …

    line 20
  2. Very specific:

    Write it to <tools folder>/out/prompts/ (t3-tots-rules "Sending into threads") and show it to the user.

    line 25
  3. Very specific:

    Start one t3-watch routine for the PM thread with t3ctl watch.

    line 43
  4. Very specific:

    When the PM goes idle, read <pm folder>/questions.md in the repo.

    line 48
  5. Distinctive:

    When a worker finishes, check that the PM picked up the result.

    line 50

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Follow the t3-tots-rules skill (skills/t3-tots-rules/SKILL.md). Steps that use the box, ListMachines, routines, cards or SendToUser are Grok Bot only; other agents follow t3-tots-rules "Grok Bot only". Starting the PM thread sends a prompt into T3, so it needs the user's yes to the brief. After that, the PM thread runs the project and you keep it moving.

Roles

  • The user decides what to build and approves anything risky or public.
  • The PM thread plans, pushes back on weak ideas, splits the work, starts worker and reviewer threads, and checks their results.
  • Worker threads do one task each, each in its own worktree and branch, so parallel workers never share a checkout.
  • Reviewer threads read a worker's diff and report findings. They don't edit code.
  • You watch every thread, pass the user's decisions in, and decide routine things under t3-tots-rules "Decisions". You don't make design choices for the PM.

1. Gather the brief

  1. Read what exists: the project's README and docs, open PRs (t3ctl prs --project '<P>'), recent threads (t3ctl threads --project '<P>' --main --limit 15), and its tracker if there is one (t3-project-tracker).
  2. Collect the user's own words about the goal, constraints and decisions already made. Don't fill gaps with your own choices; list them as open questions for the PM.
  3. Take the PM's model, the worker and reviewer models, the review rules and the PR rules (merge method, branch protection, what PR text may say) from the user's memory. If memory has none, ask once and save the answer.

2. Write the PM prompt

Write it to <tools folder>/out/prompts/ (t3-tots-rules "Sending into threads") and show it to the user. Use this outline:

You are the project manager for <project>. Goal: <the user's words>.
Context: <constraints, decisions so far, links, what already exists>.
Open questions: <gaps the user hasn't decided>.
How to work:
- Start with a plan of small tasks, each with a "done when" check. Push back where the plan or the ask looks wrong.
- Start one worker thread per task, each in its own new worktree on its own branch (never the shared project folder), and give it the task, the context it needs and its check.
- For each worker result, start reviewers on <reviewer models>. Reviewers report findings and don't edit. Post each reviewer's summary as a comment on the worker's PR. Fix and re-review until no high-severity findings remain, at most <N> rounds.
- PR rules: <the user's PR rules>.
- Verify before calling a task done: run the tests and the app, and show the output.
- Keep three files in <pm folder> in the repo and update them after each step: state.md (the plan and each task's status), decisions.md (dated decisions and who made them) and questions.md (open questions for me, newest first).
- Ask me only about product direction, money, publishing, merges to main, or anything listed here: <user's keep-for-me list>.
- Never merge, publish, deploy or force push without my yes.

3. Start it

  1. On the user's yes: t3ctl new --project '<P>' --model <m> --effort <e> --title '<Project> PM' --prompt-file <file>.
  2. Start one t3-watch routine for the PM thread with t3ctl watch. Add each worker and reviewer thread to the same watch as the PM starts them. t3ctl subtasks --thread '<PM>' lists the ones it started as sub-tasks, and t3ctl threads --project '<P>' --limit 20 the ones it started as separate threads.
  3. Tell the user in two lines: the PM thread's title and id, and what you'll bring to them.

4. Keep it moving

  • Pass the user's answers into the PM thread with t3ctl send, not into workers, so the PM keeps one picture. Keep these messages short, as t3-tots-rules "Sending into threads" says.
  • When the PM goes idle, read <pm folder>/questions.md in the repo. Bring new questions to the user, and send the answers back to the PM.
  • Handle questions, approvals, stalls and limits as in t3-watch "Handle an event".
  • When a worker finishes, check that the PM picked up the result. If the PM is idle with work left, send one short nudge naming the finished thread.
  • At each milestone, send the user a short update: what's done and checked, what's next, and what needs them.

5. Finish

The project is done when state.md shows every task done and checked, and the user has seen the result. Offer t3-save-learnings for the PM thread, and update the tracker.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.