<< All versions
Skill v1.0.0
currentAutomated scan100/100gabimoncha/oh-my-t3/omt
──Details
PublishedOctober 31, 2026 at 03:51 PM
Content Hashsha256:09064ad17ba62c41...
Git SHA6b59f03791be
──Files
Files (1 file, 7.2 KB)
SKILL.md7.2 KBactive
SKILL.md · 73 lines · 7.2 KB
name: omt description: oh-my-t3code router. Lists the omt workflows for T3 Code, when to use each, the omt CLI commands, and the shared references. disable-model-invocation: true metadata: version: "1.0.0"
oh-my-t3
omt runs workflows in this T3 thread: the main thread explores, plans, and builds, and role-routed subagents started with the T3 delegate_task tool review and test the work. Pick the workflow that matches the request, then load that skill.
Workflows
| Skill | Use it when | |
|---|---|---|
omt-ulw | The user wants something built or fixed end to end ("ulw", "ultrawork"). The main thread explores, asks, plans, and builds; one review child and one QA child check the work. The default. | |
omt-plan | The user wants a reviewed plan before any code changes, or omt-ulw needs one for a HEAVY task. | |
omt-grill | The user wants a goal grilled or stress-tested, its edge cases found, or a brainstorm before a plan, or omt-plan offers a grill for a vague goal. | |
omt-design | The user wants designs, mockups, screens, or a design pass, or asked for a design pass before a plan's UI units. | |
omt-code-review | The user wants a review of uncommitted changes, a branch, a commit, or a PR. | |
omt-research | The user wants a cited answer from this repository, docs, or the web, with no code changes. | |
omt-swarm | The user asks for separate threads: a plan built as parallel tracks in their own worktrees, then reviewed, merged, tested, and shipped as one PR. | |
omt-help | The user asks for help, what omt can do, what to do next, how to do something, which skill to use, or what a command means, or wants a tour. It only reads, and hands changes to omt-setup. | |
omt-setup | The user wants omt set up, configured, or updated: harnesses, the defaults, models, roles, profiles, or worktree setup. It asks before each write. |
For a pull request review, load omt-code-review with omt-code-review --profile pr-review <number>.
Before each commit, push, PR, or merge, a skill runs omt git check and follows the Git actions rules.
Each skill name works as a composer mention: $omt-ulw becomes /omt-ulw in Claude and Cursor, /skill:omt-ulw in Pi, and stays $omt-ulw in Codex.
CLI cheat sheet
To find the CLI (omt, .omt/bin/omt, or the manual steps), follow The CLI.
| Command | Prints | |||
|---|---|---|---|---|
omt init | creates .omt/config.jsonc, .omt/plans/, .omt/.gitignore; adds missing .gitignore lines; prints the worktree setup state | |||
omt worktree check | the current checkout's package manager, linker, t3.json setup script, and T3 activation note (worktrees.md) | |||
omt worktree setup [--dry-run] [--yes] | writes a missing t3.json, .t3/omt-worktree-setup.mjs, and for pnpm pnpm-workspace.yaml, for Bun or pnpm; never rewrites a file | |||
omt roles | every role with its target, access, and mode | |||
| `omt route <role> [--against <provider\ | driver\ | auto>]` | delegateTask and fallbacks for one role, as JSON; --against names the worker that a cross-family role reviews (roles.md) | |
| `omt brief <role> --task-file - --title "<role> <unit>: <title>" --scope <p> --done <c> --verify <cmd> --plan <slug> --unit <id> --round <n> --attempt <n> [--against <provider\ | driver\ | auto>] --save --json` | { delegateTask, fallbacks, ledger }; the full brief is in a file and delegateTask.task names it | |
omt profile list | available workflow profiles | |||
omt profile show [<name>] [--plan <slug>] --json | the effective profile, source, workflow, limits, roles, and models | |||
omt plan new <slug> [--light] [--profile <name>] | path of a new plan file under .omt/plans/; --light leaves out the F2 oracle item | |||
| `omt plan profile <slug> <name\ | none>` | switches the plan's profile; none removes the plan override | ||
| `omt plan mark <slug> <id> <done\ | open\ | blocked>` | ticks one plan item, for example omt plan mark auth U3 done | |
omt plan status [<slug>] | unit counts, the next open unit, the ready units, and a warning: line for each dependency that names no unit or names its own unit. It also prints one warning for the first dependency cycle only. | |||
omt ledger add --plan <slug> --unit <id> --role <r> --round <n> --attempt <n> --status <s> ... | appends one ledger entry; a terminal entry closes the matching dispatched entry (results.md) | |||
omt ledger stats --plan <slug> | dispatched and active counts, max parallel, and can dispatch: yes or no | |||
omt ledger request-id --plan <slug> --unit <id> --role <r> --round <n> --attempt <n> | the clientRequestId | |||
omt config show | the merged config before profile selection, including limits | |||
omt config explain [--profile <name>] [--json] | the layer that sets each value (the defaults, the user or project config, or a profile and the file that declares it) and the lower layers that it overrides | |||
| `omt detect [--capabilities <file>] [--json] [--write] [--user\ | --project] [--dry-run] [--yes]` | one line per driver (claudeAgent, codex) with its T3 status and executable, then the recommended harnesses value and its effect; --write adds, changes, or removes the harnesses key | ||
omt update [--check] [--dry-run] [--yes] [--skills-only] [--project] [--json] | updates the CLI with the command of its channel, then syncs the skills in user scope and in each recorded project; keeps an edited skill and prints kept <skill> (<reason>); --check writes nothing; shows defaults changes and asks before accepting them; --yes also accepts current defaults | |||
omt lock run <name> [--wait <seconds>] -- <command> | runs a command under a shared build lock and returns its exit code | |||
omt lock list | each lock's state and the holder of a held lock |
Pass delegateTask to delegate_task unchanged. Record the dispatch with the printed ledger.dispatched line after replacing <taskId>. Use delegation.md for retries and plan-mode dispatch.
Shared references
- delegation.md: the T3 rules.
delegate_taskfields, request ids, limits, waiting,task_status, releasing children, failure codes, top-level threads. - roles.md: the default roles, when to pick each, manual resolution, effort options, fallback, custom roles.
- brief.md: the brief layout, how to write the task text, and a full example.
- results.md: how to read each result block and verify a claim.
- frontend-design.md: when design and UI work load Anthropic's optional
frontend-designskill, and how setup recommends it. - worktrees.md: the T3 worktree setup script. Check it on the first run in a project, keep a compatible setup, and activate a new one only from the base branch.
references/roles/<role>.md: one persona per role, for example execute.md.references/results/<kind>.md: one result contract per kind, for example work.md.