Przejdź do treści
Biblioteka skilli
Programowanie · Automatyzacja

factory-config

/factory-config

Pokazuje i zmienia ustawienia projektu fabryki w zwykłej rozmowie: model budowniczego, limity zatrzymań, automatyczne wypychanie i ścieżki.

Repozytorium
claude-plugins ↗
Licencja
MIT
Ostatnia zmiana repozytorium
12.11.2026

Ocena i bezpieczeństwo

100/ 100Klasa: dobry
Format25 / 25
Frontmatter z nazwą i opisem, poprawna nazwa, opis do 1024 znaków, nazwa zgodna z folderem.
Odkrywalność25 / 25
Opis mówi, kiedy użyć skilla, ma sensowną długość i jest w 3. osobie.
Treść35 / 35
Co najmniej 150 słów i do 500 linii, nagłówki, listy kroków, przykłady albo kod.
Higiena15 / 15
Bez zaślepek typu TODO, bez ścieżek Windows, z licencją.
Ryzykobrak sygnałów w skanie
Pliki w paczce
1
Skrypty
0
Komentarze HTML
0
Adresy URL w SKILL.md
0
Przestarzałość

Brak sygnałów. Repozytorium zmienione 12.11.2026, bez odwołań do wycofanych modeli i API.

Licencja

MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.

Co robi i kiedy

Pokazuje i zmienia ustawienia projektu fabryki w zwykłej rozmowie: model budowniczego, limity zatrzymań, automatyczne wypychanie i ścieżki.

Wyzwalacze z opisu

change the factory settingsset the builder modeluse opus for the reviewersallow more failed reviewsturn on push after acceptturn on builder thinkingshow me the settings
Oryginalny opis (EN)

View and change a rebuild-factory project's settings in plain conversation, without opening the config file - the builder's model, which model each helper agent runs on, the stop limits, auto-push, what no agent may do, and the paths and commands. Use when someone says "/factory-config", "change the factory settings", "set the builder model", "use opus for the reviewers", "allow more failed reviews", "turn on push after accept", "turn on builder thinking", "show me the settings", or during first set-up of a project.

Jak działa

  1. Commands
  2. When the owner asks to see the settings
  3. When the owner asks for a change
  4. Helper models
  5. First set-up
Sekcje z nagłówków SKILL.md, w oryginale · 82 linii, 884 słów
Podgląd SKILL.md
---
name: factory-config
description: View and change a rebuild-factory project's settings in plain conversation, without opening the config file - the builder's model, which model each helper agent runs on, the stop limits, auto-push, what no agent may do, and the paths and commands. Use when someone says "/factory-config", "change the factory settings", "set the builder model", "use opus for the reviewers", "allow more failed reviews", "turn on push after accept", "turn on builder thinking", "show me the settings", or during first set-up of a project.
---

# Factory config

The owner should never have to edit `factory.config.json` by hand. This skill shows the settings in plain words and changes them through the helper script, which refuses an unknown key or a value of the wrong kind.

## Commands

```
node "${CLAUDE_PLUGIN_ROOT}/scripts/factory.mjs" show
node "${CLAUDE_PLUGIN_ROOT}/scripts/factory.mjs" set <key> <value>
node "${CLAUDE_PLUGIN_ROOT}/scripts/factory.mjs" add <key> <item>
node "${CLAUDE_PLUGIN_ROOT}/scripts/factory.mjs" remove <key> <item>
node "${CLAUDE_PLUGIN_ROOT}/scripts/factory.mjs" unset <key>
```

Everything after the key is the value, so a command with its own `--options` needs no quoting. `unset` clears a text or model setting.

`show` returns every setting with its value, its type and its meaning, lists what is still empty, and says whether this plan folder is accepted on this machine. If it is not, the script refuses every change: do not edit the file to get round that. Run `factory.mjs trust`, read what it prints to the owner, and accept with `trust --yes` only on the owner's word. Never edit the file directly: the script is what keeps a typo from becoming a dead setting.

If there is no plan folder yet, run `factory.mjs init` first.

## When the owner asks to see the settings

Run `show` and present the settings grouped, in words, not as JSON:

- **The builder**: its model, its rules file, whether its reasoning is logged.
- **Helper models**: which model each helper agent runs on.
- **Stop limits**: failed reviews, unfinished turns, stalls, and whether accepted work is pushed.
- **No agent may**: the owner-only list.
- **Machine**: how many helpers at once; where worktrees go.
- **Commands**: install, phase, mutate.

Say what is still empty and at which stage each is needed.

## When the owner asks for a change

Translate what they said into the key, make the change, and confirm in one line with the old and new value.

| The owner says | Do |
|---|---|
| "build with <model>" | `set builder.model <id>`. If unsure of the id, run `opencode models` and match it. |
| "use opus for the plan reviewers", "cheap model for the session review" | `set models.<agent> <name>` |
| "reviewers on whatever this chat uses" | `unset models.<agent>` |
| "allow eight failed reviews" | `set stop_policy.failed_reviews 8` |
| "push after each accept" | `set stop_policy.push_after_accept true` |
| "show me the builder's thinking" | `set builder.thinking true` (applies from the next run started) |
| "never let it send email" | `add owner_only "send email"` |
| "the machine is struggling" | `set limits.max_subagents 1` |

Two changes need a word of caution before you make them:

- **Removing an item from the owner-only list** hands that action to agents. Say what it allows and confirm.
- **`builder.config`** names the rules file: the commands the builder is refused while it runs auto-approved. It cannot be emptied or unset. If the owner points it at another file, read that file, say what it refuses, and run `factory.mjs doctor` to see which sample commands get through. The project's own copy is `factory-plan/builder-rules.json`; edit that one, never the copy inside the plugin. Editing it outside the script breaks the folder's acceptance, which `trust --yes` restores once the owner has seen the change.

A stop limit may be raised freely; it does not lower what counts as passing. If the conductor is running in another chat, tell the owner the change is read at the next session's start, and offer to note it at the top of `handback/PROGRESS.md` so the running conductor sees it.

## Helper models

Five helper agents, each with a setting under `models`. An empty value means "the same model as the chat that starts it".

| Setting | Helper | Suggested |
|---|---|---|
| `models.audit_reviewer` | The four audit reviewers | Strong. A missed finding is never planned for. |
| `models.finding_verifier` | Tries to disprove findings | Strong. |
| `models.plan_reviewer` | The plan reviewers | Strongest, for round 1. Mid for round 2. |
| `models.session_reviewer` | The clean-clone review | Mid. It follows a protocol; the conductor judges. |
| `models.read_list_drafter` | Drafts each reading list | Mid or cheap. The conductor checks it. |

The value is a model name the Agent tool accepts in this environment (for example `opus`, `sonnet`, `haiku`). The model of a stage's own chat is not a setting: the owner picks it with `/model` before running the stage. Suggest the strongest model for the audit, the plan and the plan review, and say so when a stage starts on a weaker one.

## First set-up

When called during set-up, ask for these one at a time, each with a recommendation, and set each as it is answered:

1. The builder's model. Have the owner run the check in the user guide if OpenCode has not been used on this machine.
2. What no agent may do: read out the default list; ask what to add.
3. The stop limits: the defaults are six failed reviews and six unfinished turns; ask when the owner expects to look in.
4. Whether to push after each accepted session.
5. The helper models, offered as one choice: "strong for the audit and plan review, mid for session reviews" or "everything on this chat's model".

Leave `commands.phase` and `commands.mutate` empty: the plan stage sets them.

Wersje

RepozytoriumWynikLicencja
Ta strona · production-ai-eng/claude-plugins100MIT
Zepp333333/factory82Apache-2.0

Migawka bazy z 13.11.2026 · wersja paczki 0f552121deb7 (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

  • 95factory
    Obsługuje polecenia fabryki oprogramowania: rejestruje projekty, przekazuje zadania między agentami i prowadzi je do PR i scalenia.
  • 100review
    Uruchamia recenzentów fabryki bez zlecenia: na bieżącej gałęzi lub PR, w czacie lub CI, pod kątem bezpieczeństwa, kodu, UX i UI.
  • 100factory-pattern-creator
    Prawdopodobnie tworzy wzorce fabryki dla automatyzacji testów na podstawie nazwy skilla i kategorii.
  • 90factory-auto
    Uruchamia pętlę fabryki bez pytań kontrolnych, sam rozstrzyga pytania i opisuje decyzje w PR, a także pilnuje otwartego PR.
  • 100factory-files
    Tworzy i edytuje pliki definicji fabryki oprogramowania Warp w drzewie z factory.yaml: Agent, Automation, Scorer, Runner oraz drzewa skilli, i naprawia diagnostykę plików fabryki.
  • 95settings
    Pokazuje i zmienia ustawienia Feature-flow: zatrzymania YOLO, powiadomienia, gałęzie, preferencje projektowe i standardy.

Uwagi i zgłoszenia

Wiadomość trafi do Bartosza Sokolińskiego. Adres e-mail jest potrzebny tylko do odpowiedzi. Szczegóły w polityce prywatności.