setup
/setup
Instaluje, konfiguruje, aktualizuje lub usuwa t3-orchestration w projekcie w T3 Code, gdy użytkownik o to prosi.
- Autor
- sanztheo ↗
- Repozytorium
- t3-orchestration ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 28.10.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 28.10.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Instaluje, konfiguruje, aktualizuje lub usuwa t3-orchestration w projekcie w T3 Code, gdy użytkownik o to prosi.
Oryginalny opis (EN)
In T3 Code only, install, configure, update or remove t3-orchestration for a project. Use when the user asks to set up orchestration on a project, add another project, update the plugin, or uninstall it.
Jak działa
- 1. Load the plugin in T3
- 2. Collect the decisions
- 3. Create the coordinator
- 4. Write the files
- 5. Create the memory thread (optional)
- 6. Check it end to end
- Update
- Uninstall
Podgląd SKILL.md
---
name: setup
description: In T3 Code only, install, configure, update or remove t3-orchestration for a project. Use when the user asks to set up orchestration on a project, add another project, update the plugin, or uninstall it.
---
# Set up t3-orchestration
Run this from a Claude thread in T3 Code, inside the project to orchestrate. Show the user each
change before making it, and ask only for the decisions listed below. The plugin directory is
`${CLAUDE_PLUGIN_ROOT}`. If that text was not replaced by a path, you were asked to read this file
directly: use the directory that contains `skills/setup/SKILL.md` instead.
Requirements: T3 Code with its `t3-code` MCP server exposing `t3_thread_configure`,
`t3_thread_read`, `t3_thread_launch`, `t3_thread_send`, `t3_project_list` and `schedule_task`;
Claude Code with mods (tested on 2.1.312); Python 3.9+ for memory. If a tool is missing, stop and
report it instead of working around it.
All user state lives in `~/.t3-orchestration/`, never in the plugin directory, so updates never
touch it:
```text
~/.t3-orchestration/
├── config.json # every orchestrated project
└── projects/<name>/
├── coordinator.md # injected into the coordinator's prompts
├── worker.md # injected into its workers' prompts
└── library/ # optional long-term memory
```
## 1. Load the plugin in T3
Skip this step when this skill was invoked as `/t3-orchestration:setup`: the plugin is loaded.
Otherwise the T3 Claude provider must start Claude Code with `--plugin-dir <plugin directory>`.
Prefer letting the user add it to the launch arguments of their Claude provider instance in T3
Code's settings. If they ask you to do it: T3 stores it in `~/.t3/userdata/settings.json`, under
`providerInstances.<id>.config.launchArgs` of an instance whose `driver` is `claudeAgent`. Back up
the file, append the flag to the existing arguments without changing anything else, and show the
diff. Only threads started afterwards load the plugin: ask the user to open a new Claude thread in
the project and run `/t3-orchestration:setup` there.
The plugin is inert in threads it does not recognise, so loading it on a provider used by other
projects is safe.
## 2. Collect the decisions
1. Identify this thread with `t3_thread_configure`, then its project with `t3_thread_read`.
Confirm the T3 project with the user (`t3_project_list` if it is not this one).
2. Ask for:
- a project `name`: lowercase letters, digits and dashes;
- the project rules both roles must follow: repositories, guardrails (production, secrets,
personal data), delivery process, review policy, language of replies. Keep them short: they
are injected into every prompt;
- whether to enable long-term memory, and how often it runs (default every 10 minutes).
3. If `~/.t3-orchestration/config.json` already lists this `projectId` or `name`, stop and ask
whether to replace it, instead of creating a second coordinator.
## 3. Create the coordinator
Create it idle, without a first message, so that its first turn already receives its reminder:
`t3_thread_launch` with the `projectId`, the title `Orchestration` and no `message`. Keep the
returned `threadId` and `link`. Suggest the user pins it.
## 4. Write the files
1. Create `~/.t3-orchestration/projects/<name>/`.
2. Fill `${CLAUDE_PLUGIN_ROOT}/templates/coordinator.md` and `worker.md`: replace
`{{PROJECT_NAME}}`, `{{COORDINATOR_THREAD_ID}}`, `{{PLUGIN_ROOT}}` (the plugin directory) and `{{PROJECT_RULES}}` (a short bullet list).
Write them as `coordinator.md` and `worker.md` in that directory.
3. Multiple Claude accounts (optional): each account is its own T3 Claude instance with its own
`homePath` (a separate `CLAUDE_CONFIG_DIR`, logged in with
`CLAUDE_CONFIG_DIR=<dir> claude auth login`) and the same `--plugin-dir`. Run
`npm install --prefix "${CLAUDE_PLUGIN_ROOT}"` once, then check
`node "${CLAUDE_PLUGIN_ROOT}/scripts/accounts.mjs"` lists every account.
4. With memory: copy `${CLAUDE_PLUGIN_ROOT}/templates/library/` to `library/`, replacing
`{{PROJECT_NAME}}`, then run
`python3 "${CLAUDE_PLUGIN_ROOT}/scripts/memory.py" --library "<library path>" status`.
5. Back up `config.json` if it exists, then add this project to `projects`, keeping the others
(see `${CLAUDE_PLUGIN_ROOT}/config.example.json`). `excludedThreadIds` holds this setup thread's
id; `memoryThreadId` stays `null` until step 5; `libraryPath` is absolute, or `null`.
## 5. Create the memory thread (optional)
A schedule runs in the thread that creates it, so the memory thread creates its own. Launch it
with `t3_thread_launch` in the same project, titled `Memory`, with this first message, filled in:
> You are the memory thread of `<name>`. Create your own schedule with `schedule_task`:
> `schedule: {type: "interval", everyMs: <interval in ms>}`, `bindToCurrentThread: true`, title
> `<name> memory pass`, prompt `Run one incremental pass with the t3-orchestration:memory skill.`
> Report the schedule id and next run time, then stop. Do not run a pass yet.
Then set `memoryThreadId` in `config.json` to the returned `threadId`. Read the memory thread to
confirm the schedule exists; report its id or the failure exactly.
## 6. Check it end to end
Give the user this check and the links:
1. In the coordinator, send any request. Its context contains `[T3O_COORDINATOR_V1]`.
2. Ask the coordinator to open a small worker thread. The worker's context contains
`[T3O_WORKER_V1]`, and when its turn ends the coordinator receives `[T3O_SIGNAL_V1]`.
3. In a thread the user creates manually, neither marker appears and no signal is sent.
Report what was created, with the exact links, and what remains unverified.
## Update
1. Note the current commit (`git -C "${CLAUDE_PLUGIN_ROOT}" rev-parse HEAD`), then
`git -C "${CLAUDE_PLUGIN_ROOT}" pull --ff-only`. Stop on any conflict or local change.
2. Show `git log` and the diff of `templates/` between the two commits. The user's reminder files
are theirs: propose the template changes, never overwrite them.
3. Run `npm install --prefix "${CLAUDE_PLUGIN_ROOT}"` if `package.json` changed.
4. New threads load the new version; ask the user to restart the threads already running.
## Uninstall
On the user's explicit request only: delete the project's memory schedule
(`list_scheduled_tasks`, then `delete_scheduled_task`) and remove the project from `config.json`.
Remove `--plugin-dir` from the T3 launch arguments only when no project is left: it serves every
configured project. Ask before deleting `~/.t3-orchestration/projects/<name>/`: it holds
the memory library. Threads are kept; archiving them is the user's choice.
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · sanztheo/t3-orchestration | 100 | MIT |
| HwangSlater/harness-fit | 100 | MIT |
| nesquikm/dev-process-toolkit | 100 | MIT |
| userKael/rastro | 100 | MIT |
| als-apg/osprey | 100 | BSD-3-Clause |
| dedujeong/gemba | 100 | Apache-2.0 |
| eigenwise/eigenwise-toolshed | 100 | MIT |
| awrshift/agent-memory-kit | 100 | MIT |
| lifecycle-innovations-limited/claude-ops | 100 | MIT |
Migawka bazy z 13.11.2026 · wersja paczki be7d4ea5f517 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100setting-up-software-writer-extension
Instaluje, konfiguruje lub migruje rozszerzenie wtyczki software-writer dla bieżącego projektu, tylko gdy użytkownik o to prosi. - 100cc-spex-init
Instaluje lub odświeża przepływy Spex oparte na specyfikacji w bieżącym projekcie ze źródła dołączonego do wtyczki cc-spex oraz konfiguruje rozszerzenia i ustawienia bezpieczeństwa. - 100manage-skills
Instaluje, usuwa i sprawdza skille z klona repozytorium kapo-skills i aktualizuje wspólny zestaw. - 90framework
Prowadzi rekord instalacji i wszystkie zmiany w projekcie: manifest oraz przepływy setup, instalacji, przywracania, odinstalowania i diagnostyki. - 100openspec-init
Konfiguruje OpenSpec w projekcie, gdy użytkownik prosi o inicjalizację lub chce dodać ten framework do repozytorium. - 90update
Aktualizuje wersję frontendowego harnessu w projekcie do najnowszej wersji wtyczki.