Skip to content

gabimoncha/oh-my-t3 / skills/omt-setup

omt-setup

Verified

Set up, configure, update, or explain omt in T3 Code. Use when the user asks to set up, configure, or update omt, change models, roles, or profiles, pick the harnesses (Claude Code, Codex), or review new defaults, or asks for help, what omt can do, what to do next, which skill to use, or what a command means. Asks before each write.

Setup and help

You answer questions about omt and configure it through the omt CLI. Show the text output of each command.

Write rules

  1. Show the command, or the planned edit, before you run it.
  2. When the command has --dry-run, run that first and show the output.
  3. Ask. Pass --yes, or make the edit, only after the user agrees.
  4. After the write, run the check of the step.

Steps

Run step 1 first. For a question, run step 3. For a change, run step 2, then the steps that it names. For a plain "set up omt", run steps 2, 4, 5, and 6, then check for Anthropic's optional frontend-design skill and recommend it for UI work when it is missing. End with step 9 when you proposed a write.

1. Find the CLI and your mode

Follow The CLI. Call t3_thread_configure with no arguments and read interactionMode. In plan mode, run only read commands, list the writes for the user, and ask the user to switch to default mode.

Done when: omt --version prints a version, and you know your mode.

2. Read the doctor

Run omt doctor --project --dry-run, or omt doctor --dry-run for a global install. Map each line that is not PASS: harnesses to step 4, defaults to step 5, worktree to step 6, config to step 7, and skills with outdated, hooks, or update with is out to step 8. For another line, give the user its printed fix.

Done when: each line that is not PASS has a step or a fix, and the user knows your proposed steps.

3. Answer a question

This step writes nothing. Sort the question, and run its commands first:

  • Which skill to use, or what omt can do: the workflow list of the router.
  • What a command means: omt help <command>.
  • A config question: omt config show, omt config explain, omt profile list, omt profile show <name>, omt roles, or omt detect.
  • What to do next: the doctor command of step 2, and omt update --check for a new version.

Answer first, in one to three sentences. For a skill, give its composer form, such as $omt-ulw <task>. For a change, name the key and the file, and offer the write steps. Link one doc as https://github.com/gabimoncha/oh-my-t3/blob/main/docs/<file>.md: install, configuration, commands, workflows, walkthrough, or design.

Done when: every fact in the answer comes from a command that you ran in this turn, the router, or the linked doc.

4. Fit the roles to the harnesses

Pass --project, or --user for a global install, to each command of this step. With the T3 tools, save the full JSON result of orchestrator_capabilities to a temp file, and also pass --capabilities <file>.

  1. Run omt detect, and read the driver and effect lines.
  2. When it recommends a change, run omt detect --write --dry-run, and ask.
  3. After the user agrees, run omt detect --write --yes.
  4. The check: omt detect, then omt roles.

Without the T3 tools, say that detection sees only the T3 settings file and PATH. When omt detect finds no driver, ask the user to sign in to Claude Code or Codex and enable it in T3 Settings, then run this step again.

Done when: the check recommends no further change, and omt roles shows each role as the effect lines said.

5. Review current defaults

Run the doctor command of step 2, and explain each <path>: <from> -> <to> line. A line marked (overridden by <file>) has no effect. On the first review of an older install, show the current role choices, since no baseline exists. After the user accepts, run the same command with --yes in place of --dry-run. Declining keeps the last accepted defaults.

Done when: the user accepted or declined, and omt config validate exits 0 after acceptance.

6. Set up worktrees

Follow worktrees.md on the first run in a project, or for a worktree line that is not PASS.

Done when: you finished each step of worktrees.md that applies.

7. Change a profile or a role

  1. Run omt config path. Pick the project file for one project, or the user file for all. When it is absent, propose omt init or omt config init.
  2. Read the file and omt config show --defaults. Write the JSONC edit with its key path by the configuration guide. A new role needs description, access, and chain. A chain replaces the old chain, and null removes a role.
  3. Show the edit and ask. After yes, make it, and keep every other byte of the file.
  4. The check: omt config validate, then omt profile show <name>.

For a browser editor, run omt config edit --web in the background, with --capabilities <file> as in step 4, and give the user its URL. The page writes the files, so run only item 4, after the user is done.

Done when: omt config validate exits 0, and omt profile show <name> shows the new values.

8. Update omt

  1. Run omt update --dry-run, and show its update command, pin, hint, or package switch.
  2. When you ask, say that --yes also accepts the current defaults. When the user wants to see a defaults diff that the dry run cannot show, run omt update interactively instead.
  3. After the user agrees, run omt update --yes. Tell the user each kept <skill> line: that skill keeps the user's edits. When the npm registry is out of reach, follow Update omt.
  4. The check: the doctor command of step 2. When a skill changed, tell the user to start new threads, because a running session keeps the old skills.

Done when: the doctor prints PASS for skills, and the user knows to start new threads.

9. Report

The report contains each changed file and key, each write command with its check result, each declined step, and each open WARN or FAIL line.

Done when: each write of this run is in the report with its check result.

reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 470kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall