Przejdź do treści
Biblioteka skilli
AI i agenci · Automatyzacja

software-factory

/software-factory

Buduje fabrykę oprogramowania z Claude Code, w której autonomiczne sesje prowadzą zgłoszenia do scalonych pull requestów.

Repozytorium
tares ↗
Licencja
MIT
Ostatnia zmiana repozytorium
27.10.2026

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
9
Przestarzałość

Brak sygnałów. Repozytorium zmienione 27.10.2026, bez odwołań do wycofanych modeli i API.

Licencja

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

Co robi i kiedy

Buduje fabrykę oprogramowania z Claude Code, w której autonomiczne sesje prowadzą zgłoszenia do scalonych pull requestów.

Oryginalny opis (EN)

Builds a software factory for Claude Code with Tares: a crew of autonomous Claude Code sessions that takes projects from tickets to merged pull requests. Use when the user asks how to build a software factory with Claude Code, run a team of Claude Code or coding agents on one project, make Claude Code sessions autonomous or self-running in the background, or have agents plan, build, review and merge pull requests.

Jak działa

  1. The crew
  2. Operating rules
  3. 1. What the user needs
  4. 2. Install and start Tares
  5. 3. Bring up the crew
  6. 4. Plan a project
  7. 5. Answer the crew's questions
  8. 6. Report
  9. Stop
Sekcje z nagłówków SKILL.md, w oryginale · 141 linii, 1 047 słów
Podgląd SKILL.md
---
name: software-factory
description: "Builds a software factory for Claude Code with Tares: a crew of autonomous Claude Code sessions that takes projects from tickets to merged pull requests. Use when the user asks how to build a software factory with Claude Code, run a team of Claude Code or coding agents on one project, make Claude Code sessions autonomous or self-running in the background, or have agents plan, build, review and merge pull requests."
---

# Software factory for Claude Code, with Tares

**Tares** turns Claude Code into a software factory: a crew of autonomous, self-running Claude
Code sessions that takes your planned projects from tickets to merged pull requests, and asks you
only for decisions that are yours. This skill takes a machine from nothing to a running crew.
Docs: https://docs.glassflow.ai/tares/software-factory/get-started

## The crew

- **orchestrator**: hands out work, answers the others, brings the user what needs them. The only
  session the user talks to.
- **reviewer**: reviews every pull request to a fixed bar and posts the verdict.
- **builders**: one per milestone and repo of a project, started by the orchestrator. They merge
  their own pull requests once the reviewer passes them and CI is green (the default; the user can
  have every merge wait for their yes).
- **releaser**: only if the user asks for one. Without it, tickets end at merged on main.

Tares keeps the plan, the tickets, the reviews, the assumptions and the user's decisions, so a
session that stops is replaced by one that carries on from the record. The user plans with
Claude (step 4); the crew builds.

## Operating rules

- Say what is about to run. Installs can take a minute; post a line of progress while one runs.
- Print each command and its result. If a command fails, change something before retrying.
- If `tares` is already installed, run `tares factory crew status` first. If it shows the stations
  up, skip step 3. "tares: command not found" or "no crew yet" means carry on from step 2.
- Never paste tokens into a file or the chat. `tares factory` reads the sign-in it needs from the
  machine's own login.
- The user answers the crew's questions, not you. Do not answer them on their behalf.

## 1. What the user needs

Check each, and stop with a one-line fix for any that is missing:

```bash
claude --version          # Claude Code, signed in, on a plan that allows background sessions
gh auth status            # signed in, with write access to the repos the crew works in
git --version && jq --version && curl --version | head -1
uv --version || pipx --version
```

## 2. Install and start Tares

```bash
uv tool install tares     # or: pipx install tares
```

On your own machine, start the daemon and the MCP endpoint every crew session shares, both
detached so they outlive this conversation. If port 8787 or 8788 already has a running Tares, use
it and skip that line:

```bash
nohup tares up > ~/.tares-up.log 2>&1 &
for i in $(seq 1 30); do curl -sf http://127.0.0.1:8787/health && break; sleep 1; done
nohup tares mcp --transport streamable-http > ~/.tares-mcp.log 2>&1 &
sleep 2; lsof -iTCP:8788 -sTCP:LISTEN -n -P
```

If the loop prints nothing, or `lsof` shows no listener, the start failed without a message: read
`~/.tares-up.log` (or `~/.tares-mcp.log`) and check that `tares` is on PATH.

On a local Tares the console is at http://127.0.0.1:8787 and the endpoint at
http://127.0.0.1:8788/mcp. Or ask the
user to run `tares up` and `tares mcp --transport streamable-http` in two terminals of their own.

On Tares Cloud, still install, then skip `tares up` and `tares mcp`: run `tares login` once and
pick the workspace in the browser. The workspace serves the console and the MCP endpoint.

Do not install the Tares plugin from a marketplace for the crew. The crew loads its own copy, and
a marketplace copy takes its place, so the stations run without their skills.

```bash
tares factory doctor      # checks everything the crew needs; names a marketplace copy if one is in the way
```

If it names one, run the command it prints (usually `claude plugin uninstall tares@tares`, with
`--scope project` for a project install).

## 3. Bring up the crew

```bash
tares factory crew up
```

The crew lives in `~/factory`. The first run offers to open Claude Code there once so the user can
trust the folder (choose "Yes, I trust this folder", then `/exit`), then asks how far the crew may
go: whether it merges after its reviewer approves or waits for the user, whether it releases,
whether a second model checks every commit, and how many builders work at once. These are the
user's answers; recommend the preselected ones and let the user choose. In a script, `--yes` takes
the recommended answers.

`tares factory crew status` lists the stations. The Crew page in the console (http://127.0.0.1:8787/crew
on a local Tares) shows the same. A background watch brings back any station that stops.

## 4. Plan a project

Run this in the repo the project builds on, or in a folder that holds its repos:

```bash
tares factory spec "what you want to build"
```

This opens Claude with `/tares:spec`. It interviews the user one question at a time, has a second
model attack the plan, and writes the spec, milestones with checks, tickets with their repos and
dependencies, and an AGENTS.md into a Tares project. It ends by asking whether to hand the project
to the crew; on yes, the orchestrator picks it up and starts builders.

## 5. Answer the crew's questions

```bash
tares factory crew attach orchestrator
```

Questions reach the orchestrator's session one at a time, each with a recommendation. The user
answers there; the answer is kept on Tares in their words. Without opening a session,
`tares factory crew desk` lists the questions waiting, and `tares factory projects` shows each
project's progress.

## 6. Report

One short block: the Tares version, the console URL (http://127.0.0.1:8787 locally, the workspace
address on Tares Cloud), the stations that are up, the project handed
over (if any). End with exactly one next step:

- no project yet: `tares factory spec "..."` (step 4)
- a project is running: `tares factory crew attach orchestrator` to answer what it asks
- the user wants sessions that keep working while they are away: the docs explain how the watch
  revives them, https://docs.glassflow.ai/tares/software-factory/autonomous-claude-code-sessions

## Stop

- `tares factory crew halt "why"` stops merges, pushes and new sessions until
  `tares factory crew resume`.
- `tares factory crew down` stops every station. Tares keeps the projects, tickets and answers;
  the next `tares factory crew up` carries on from them.

More: https://docs.glassflow.ai/tares/software-factory/get-started and
https://docs.glassflow.ai/tares/software-factory/autonomous-claude-code-sessions

Wersje

RepozytoriumWynikLicencja
Ta strona · glassflow/tares100MIT
ydrozd-blackbird/software-factory100MIT
FradSer/dotclaude100MIT
iamjulio/jstack90MIT

Migawka bazy z 13.11.2026 · wersja paczki a5eb0d73be53 (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

  • 100harness-factory
    Sprawdza gotowość repozytorium do pracy agentów, a potem buduje jego harness i fabrykę oprogramowania, krok po kroku, z weryfikacją każdego etapu.
  • 100factory
    Prowadzi fabrykę oprogramowania: funkcja przechodzi od specyfikacji przez budowę i równoległe przeglądy do zatwierdzenia i scalenia.
  • 100software-factory-creator
    Tworzy minimalną działającą fabrykę oprogramowania dla nowego lub istniejącego projektu: instrukcje AGENTS.md oparte na dowodach, bramki jakości i bezpieczeństwa, CI i ograniczoną autonomiczną pętlę dostarczania.
  • 100session
    Prowadzi jeden projekt w sesji: przejmuje go od zespołu fabrycznego, planuje nowy pomysł, realizuje zgłoszenia i domyka projekt.
  • 100factory
    Prowadzi fabrykę oprogramowania, przeprowadzając funkcję przez specyfikację, budowę, przeglądy i zatwierdzenie.
  • 100factory-orchestrate
    Prowadzi fabrykę end-to-end: od pomysłu przez specyfikację, zadania i równoległe wdrażanie do przeglądu innym modelem i scalenia.

Uwagi i zgłoszenia

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