@the-bearded-bear/claude-craft 8.26.0-next.2a54921 → 8.26.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/Dev/i18n/de/Common/commands/ralph-run.md +0 -46
- package/Dev/i18n/en/Common/commands/ralph-run.md +0 -46
- package/Dev/i18n/es/Common/commands/ralph-run.md +0 -46
- package/Dev/i18n/fr/Common/commands/ralph-run.md +0 -46
- package/Dev/i18n/pt/Common/commands/ralph-run.md +0 -46
- package/package.json +1 -1
|
@@ -76,35 +76,6 @@ Das DoD-System validiert den Abschluss über mehrere Kriterien:
|
|
|
76
76
|
| `hook` | Bestehenden Claude-Hook ausführen |
|
|
77
77
|
| `human` | Interaktive menschliche Validierung |
|
|
78
78
|
|
|
79
|
-
#### Grenze des `command`-Validators → `type: hook` verwenden
|
|
80
|
-
|
|
81
|
-
`type: command` führt nur Befehle aus, die die Allowlist in `Tools/Ralph/lib/dod-validator.sh` bestehen: Buchstaben, Ziffern, Leerzeichen und `_ . / " ' -`. Jeder Befehl mit `=`, `|`, `&` oder `:` (sowie `;`, `>`, `$` usw.) wird mit `ERROR: Invalid command` abgelehnt: Das Kriterium wird dann **nie** erfüllt und die Schleife läuft, bis der Circuit Breaker auslöst.
|
|
82
|
-
|
|
83
|
-
| Abgelehnter Befehl | Problematisches Zeichen |
|
|
84
|
-
|--------------------|-------------------------|
|
|
85
|
-
| `dotnet build /p:TreatWarningsAsErrors=true` | `:` und `=` |
|
|
86
|
-
| `npm test && npm run lint` | `&` |
|
|
87
|
-
| `pytest \| tee out.log` | `\|` |
|
|
88
|
-
|
|
89
|
-
In diesen Fällen den Befehl in ein Skript auslagern und als `type: hook` deklarieren:
|
|
90
|
-
|
|
91
|
-
```yaml
|
|
92
|
-
definition_of_done:
|
|
93
|
-
checklist:
|
|
94
|
-
- id: build
|
|
95
|
-
type: hook
|
|
96
|
-
script: ".ralph/dod/build.sh"
|
|
97
|
-
required: true
|
|
98
|
-
```
|
|
99
|
-
|
|
100
|
-
```bash
|
|
101
|
-
#!/usr/bin/env bash
|
|
102
|
-
# .ralph/dod/build.sh — erhält {"cwd": ..., "output": ...} als JSON über stdin
|
|
103
|
-
dotnet build /p:TreatWarningsAsErrors=true || exit 2
|
|
104
|
-
```
|
|
105
|
-
|
|
106
|
-
Das Skript gibt `0` (Kriterium erfüllt) oder `2` (Kriterium fehlgeschlagen) zurück. Gelesene Schlüssel je Typ: `command` → `command:`, `output_contains` und `file_changed` → `pattern:`, `hook` → `script:`, `human` → `prompt:`.
|
|
107
|
-
|
|
108
79
|
### 4. Adaptiver Circuit Breaker (v2.0)
|
|
109
80
|
|
|
110
81
|
Wählt automatisch das Profil basierend auf Aufgaben-Schlüsselwörtern:
|
|
@@ -129,23 +100,6 @@ Claude arbeitet...
|
|
|
129
100
|
Stop → stop-dod-gate.sh → Blockieren, wenn DoD nicht erfüllt (exit 2)
|
|
130
101
|
```
|
|
131
102
|
|
|
132
|
-
## /goal (interaktiv) oder Ralph (autonom/CI)?
|
|
133
|
-
|
|
134
|
-
`/goal` ist ein nativer Claude-Code-Befehl, der in einer Sitzung eingegeben wird: Er ist keine Option von `ralph.sh` — `ralph.sh` bietet dafür keine an und erzeugt nie ein `/goal`. Beide Werkzeuge ergänzen sich:
|
|
135
|
-
|
|
136
|
-
| Kriterium | `/goal` (nativ) | Ralph (`ralph.sh`) |
|
|
137
|
-
|-----------|-----------------|--------------------|
|
|
138
|
-
| Abschlussprüfung | LLM-Urteil: Ein kleines, schnelles Modell bewertet nach jedem Zug, was in der Konversation sichtbar ist — es führt keinen Befehl aus und liest keine Datei | Deterministisch: Befehle, Muster und Hooks, ausgeführt von `dod-validator.sh` |
|
|
139
|
-
| Iterationsobergrenze | Keine feste Grenze: Ende, wenn die Bedingung erfüllt oder als unmöglich bewertet ist, bei `/goal clear` oder wenn Claude mehrere Züge in Folge ohne Werkzeug antwortet; eine Grenze kann in die Bedingung geschrieben werden | `max_iterations` + adaptiver Circuit Breaker |
|
|
140
|
-
| Headless / CI | `claude -p "/goal ..."` läuft in einem einzigen Aufruf bis zum Ende; CI-Nutzung nicht dokumentiert | Dafür gebaut: autonome Läufe (`--autonomous`, `--overnight`), `headless`-Dashboard |
|
|
141
|
-
| `disableAllHooks: true` | Nicht verfügbar: `/goal` umhüllt einen sitzungsbezogenen Stop-Hook (ebenso mit `allowManagedHooksOnly`) | Schleife und DoD laufen in `ralph.sh`; nur die Hooks-Integration (`stop-dod-gate.sh`) ist deaktiviert |
|
|
142
|
-
| Konfiguration | Keine; ein Goal pro Sitzung, durch `/clear` gelöscht | `ralph.yml` |
|
|
143
|
-
|
|
144
|
-
- **`/goal` wählen** für eine interaktive Aufgabe, deren Abschluss in der Konversation sichtbar wird (Testausgabe anzeigen: Der Evaluator führt die Tests nicht aus).
|
|
145
|
-
- **Ralph wählen**, wenn der Abschluss durch Befehle belegt werden muss, mit Iterationsobergrenze, unbeaufsichtigt oder in CI.
|
|
146
|
-
|
|
147
|
-
Referenz: [`/goal`-Dokumentation](https://code.claude.com/docs/en/goal).
|
|
148
|
-
|
|
149
103
|
## Schnellstart-Beispiele
|
|
150
104
|
|
|
151
105
|
```bash
|
|
@@ -76,35 +76,6 @@ The DoD system validates completion through multiple criteria:
|
|
|
76
76
|
| `hook` | Run existing Claude hook |
|
|
77
77
|
| `human` | Interactive human validation |
|
|
78
78
|
|
|
79
|
-
#### `command` validator limit → use `type: hook`
|
|
80
|
-
|
|
81
|
-
`type: command` only runs commands that pass the allowlist in `Tools/Ralph/lib/dod-validator.sh`: letters, digits, space and `_ . / " ' -`. Any command containing `=`, `|`, `&` or `:` (as well as `;`, `>`, `$`, etc.) is rejected with `ERROR: Invalid command`: the criterion is then **never** met and the loop runs until the circuit breaker trips.
|
|
82
|
-
|
|
83
|
-
| Rejected command | Offending character |
|
|
84
|
-
|------------------|---------------------|
|
|
85
|
-
| `dotnet build /p:TreatWarningsAsErrors=true` | `:` and `=` |
|
|
86
|
-
| `npm test && npm run lint` | `&` |
|
|
87
|
-
| `pytest \| tee out.log` | `\|` |
|
|
88
|
-
|
|
89
|
-
For these cases, put the command in a script and declare it as `type: hook`:
|
|
90
|
-
|
|
91
|
-
```yaml
|
|
92
|
-
definition_of_done:
|
|
93
|
-
checklist:
|
|
94
|
-
- id: build
|
|
95
|
-
type: hook
|
|
96
|
-
script: ".ralph/dod/build.sh"
|
|
97
|
-
required: true
|
|
98
|
-
```
|
|
99
|
-
|
|
100
|
-
```bash
|
|
101
|
-
#!/usr/bin/env bash
|
|
102
|
-
# .ralph/dod/build.sh — receives {"cwd": ..., "output": ...} as JSON on stdin
|
|
103
|
-
dotnet build /p:TreatWarningsAsErrors=true || exit 2
|
|
104
|
-
```
|
|
105
|
-
|
|
106
|
-
The script returns `0` (criterion met) or `2` (criterion failed). Keys read per type: `command` → `command:`, `output_contains` and `file_changed` → `pattern:`, `hook` → `script:`, `human` → `prompt:`.
|
|
107
|
-
|
|
108
79
|
### 4. Adaptive Circuit Breaker (v2.0)
|
|
109
80
|
|
|
110
81
|
Automatically selects profile based on task keywords:
|
|
@@ -129,23 +100,6 @@ Claude works...
|
|
|
129
100
|
Stop → stop-dod-gate.sh → Block if DoD not satisfied (exit 2)
|
|
130
101
|
```
|
|
131
102
|
|
|
132
|
-
## /goal (interactive) or Ralph (autonomous/CI)?
|
|
133
|
-
|
|
134
|
-
`/goal` is a native Claude Code command typed inside a session: it is not a `ralph.sh` option — `ralph.sh` has none for it and never generates a `/goal`. The two tools are complementary:
|
|
135
|
-
|
|
136
|
-
| Criterion | `/goal` (native) | Ralph (`ralph.sh`) |
|
|
137
|
-
|-----------|------------------|--------------------|
|
|
138
|
-
| Completion check | LLM judgment: a small fast model evaluates, after each turn, what is visible in the conversation — it runs no command and reads no file | Deterministic: commands, patterns and hooks run by `dod-validator.sh` |
|
|
139
|
-
| Iteration cap | No fixed cap: ends when the condition is met or judged impossible, on `/goal clear`, or if Claude answers several turns in a row without tools; a bound can be written into the condition | `max_iterations` + adaptive circuit breaker |
|
|
140
|
-
| Headless / CI | `claude -p "/goal ..."` loops to completion in a single invocation; CI usage not documented | Built for it: autonomous runs (`--autonomous`, `--overnight`), `headless` dashboard |
|
|
141
|
-
| `disableAllHooks: true` | Unavailable: `/goal` wraps a session-scoped Stop hook (same with `allowManagedHooksOnly`) | The loop and the DoD run inside `ralph.sh`; only the hooks integration (`stop-dod-gate.sh`) is disabled |
|
|
142
|
-
| Configuration | None; one goal per session, cleared by `/clear` | `ralph.yml` |
|
|
143
|
-
|
|
144
|
-
- **Choose `/goal`** for an interactive task whose completion shows up in the conversation (print the test output: the evaluator does not run it).
|
|
145
|
-
- **Choose Ralph** when completion must be proven by commands, with an iteration cap, unattended or in CI.
|
|
146
|
-
|
|
147
|
-
Reference: [`/goal` documentation](https://code.claude.com/docs/en/goal).
|
|
148
|
-
|
|
149
103
|
## Quick Start Examples
|
|
150
104
|
|
|
151
105
|
```bash
|
|
@@ -76,35 +76,6 @@ El sistema DoD valida la finalización a través de múltiples criterios:
|
|
|
76
76
|
| `hook` | Ejecutar un hook de Claude existente |
|
|
77
77
|
| `human` | Validación humana interactiva |
|
|
78
78
|
|
|
79
|
-
#### Límite del validador `command` → usar `type: hook`
|
|
80
|
-
|
|
81
|
-
`type: command` solo ejecuta los comandos que pasan la lista blanca de `Tools/Ralph/lib/dod-validator.sh`: letras, dígitos, espacio y `_ . / " ' -`. Cualquier comando que contenga `=`, `|`, `&` o `:` (así como `;`, `>`, `$`, etc.) se rechaza con `ERROR: Invalid command`: el criterio **nunca** se valida y el bucle se ejecuta hasta que salta el circuit breaker.
|
|
82
|
-
|
|
83
|
-
| Comando rechazado | Carácter culpable |
|
|
84
|
-
|-------------------|-------------------|
|
|
85
|
-
| `dotnet build /p:TreatWarningsAsErrors=true` | `:` y `=` |
|
|
86
|
-
| `npm test && npm run lint` | `&` |
|
|
87
|
-
| `pytest \| tee out.log` | `\|` |
|
|
88
|
-
|
|
89
|
-
En estos casos, colocar el comando en un script y declararlo como `type: hook`:
|
|
90
|
-
|
|
91
|
-
```yaml
|
|
92
|
-
definition_of_done:
|
|
93
|
-
checklist:
|
|
94
|
-
- id: build
|
|
95
|
-
type: hook
|
|
96
|
-
script: ".ralph/dod/build.sh"
|
|
97
|
-
required: true
|
|
98
|
-
```
|
|
99
|
-
|
|
100
|
-
```bash
|
|
101
|
-
#!/usr/bin/env bash
|
|
102
|
-
# .ralph/dod/build.sh — recibe {"cwd": ..., "output": ...} en JSON por stdin
|
|
103
|
-
dotnet build /p:TreatWarningsAsErrors=true || exit 2
|
|
104
|
-
```
|
|
105
|
-
|
|
106
|
-
El script devuelve `0` (criterio cumplido) o `2` (criterio fallido). Claves leídas por tipo: `command` → `command:`, `output_contains` y `file_changed` → `pattern:`, `hook` → `script:`, `human` → `prompt:`.
|
|
107
|
-
|
|
108
79
|
### 4. Circuit Breaker Adaptativo (v2.0)
|
|
109
80
|
|
|
110
81
|
Selecciona automáticamente el perfil basándose en palabras clave de la tarea:
|
|
@@ -129,23 +100,6 @@ Claude trabaja...
|
|
|
129
100
|
Stop → stop-dod-gate.sh → Bloquear si DoD no satisfecho (exit 2)
|
|
130
101
|
```
|
|
131
102
|
|
|
132
|
-
## ¿/goal (interactivo) o Ralph (autónomo/CI)?
|
|
133
|
-
|
|
134
|
-
`/goal` es un comando nativo de Claude Code que se escribe dentro de una sesión: no es una opción de `ralph.sh`, que no ofrece ninguna para ello y nunca genera un `/goal`. Ambas herramientas son complementarias:
|
|
135
|
-
|
|
136
|
-
| Criterio | `/goal` (nativo) | Ralph (`ralph.sh`) |
|
|
137
|
-
|----------|------------------|--------------------|
|
|
138
|
-
| Criterio de fin | Juicio LLM: un modelo pequeño y rápido evalúa, tras cada turno, lo que es visible en la conversación — no ejecuta ningún comando ni lee ningún archivo | Determinista: comandos, patrones y hooks ejecutados por `dod-validator.sh` |
|
|
139
|
-
| Límite de iteraciones | Sin límite fijo: termina cuando la condición se cumple o se juzga imposible, con `/goal clear`, o si Claude encadena varios turnos sin herramientas; se puede escribir un límite en la condición | `max_iterations` + circuit breaker adaptativo |
|
|
140
|
-
| Headless / CI | `claude -p "/goal ..."` itera hasta el final en una sola invocación; uso en CI no documentado | Diseñado para ello: ejecución autónoma (`--autonomous`, `--overnight`), dashboard `headless` |
|
|
141
|
-
| `disableAllHooks: true` | No disponible: `/goal` envuelve un Stop hook de sesión (igual con `allowManagedHooksOnly`) | El bucle y la DoD se ejecutan dentro de `ralph.sh`; solo se desactiva la integración de hooks (`stop-dod-gate.sh`) |
|
|
142
|
-
| Configuración | Ninguna; un único goal por sesión, borrado por `/clear` | `ralph.yml` |
|
|
143
|
-
|
|
144
|
-
- **Elegir `/goal`** para una tarea interactiva cuyo fin se constata en la conversación (mostrar la salida de los tests: el evaluador no los ejecuta).
|
|
145
|
-
- **Elegir Ralph** cuando el fin debe demostrarse con comandos, con un límite de iteraciones, de forma autónoma o en CI.
|
|
146
|
-
|
|
147
|
-
Referencia: [documentación de `/goal`](https://code.claude.com/docs/en/goal).
|
|
148
|
-
|
|
149
103
|
## Ejemplos de Inicio Rápido
|
|
150
104
|
|
|
151
105
|
```bash
|
|
@@ -76,35 +76,6 @@ Le système DoD valide la complétion à travers plusieurs critères :
|
|
|
76
76
|
| `hook` | Exécuter un hook Claude existant |
|
|
77
77
|
| `human` | Validation humaine interactive |
|
|
78
78
|
|
|
79
|
-
#### Limite du validateur `command` → utiliser `type: hook`
|
|
80
|
-
|
|
81
|
-
`type: command` n'exécute que les commandes qui passent la liste blanche de `Tools/Ralph/lib/dod-validator.sh` : lettres, chiffres, espace et `_ . / " ' -`. Toute commande contenant `=`, `|`, `&` ou `:` (ainsi que `;`, `>`, `$`, etc.) est rejetée avec `ERROR: Invalid command` : le critère n'est alors **jamais** validé et la boucle tourne jusqu'au circuit breaker.
|
|
82
|
-
|
|
83
|
-
| Commande rejetée | Caractère en cause |
|
|
84
|
-
|------------------|--------------------|
|
|
85
|
-
| `dotnet build /p:TreatWarningsAsErrors=true` | `:` et `=` |
|
|
86
|
-
| `npm test && npm run lint` | `&` |
|
|
87
|
-
| `pytest \| tee out.log` | `\|` |
|
|
88
|
-
|
|
89
|
-
Pour ces cas, placer la commande dans un script et la déclarer en `type: hook` :
|
|
90
|
-
|
|
91
|
-
```yaml
|
|
92
|
-
definition_of_done:
|
|
93
|
-
checklist:
|
|
94
|
-
- id: build
|
|
95
|
-
type: hook
|
|
96
|
-
script: ".ralph/dod/build.sh"
|
|
97
|
-
required: true
|
|
98
|
-
```
|
|
99
|
-
|
|
100
|
-
```bash
|
|
101
|
-
#!/usr/bin/env bash
|
|
102
|
-
# .ralph/dod/build.sh — reçoit {"cwd": ..., "output": ...} en JSON sur stdin
|
|
103
|
-
dotnet build /p:TreatWarningsAsErrors=true || exit 2
|
|
104
|
-
```
|
|
105
|
-
|
|
106
|
-
Le script retourne `0` (critère validé) ou `2` (critère échoué). Clés lues par type : `command` → `command:`, `output_contains` et `file_changed` → `pattern:`, `hook` → `script:`, `human` → `prompt:`.
|
|
107
|
-
|
|
108
79
|
### 4. Circuit Breaker Adaptatif (v2.0)
|
|
109
80
|
|
|
110
81
|
Sélection automatique du profil selon les mots-clés de la tâche :
|
|
@@ -129,23 +100,6 @@ Claude travaille...
|
|
|
129
100
|
Stop → stop-dod-gate.sh → Bloquer si DoD non satisfait (exit 2)
|
|
130
101
|
```
|
|
131
102
|
|
|
132
|
-
## /goal (interactif) ou Ralph (autonome/CI) ?
|
|
133
|
-
|
|
134
|
-
`/goal` est une commande native de Claude Code, tapée dans une session : ce n'est pas une option de `ralph.sh`, qui n'en propose aucune et ne génère jamais de `/goal`. Les deux outils sont complémentaires :
|
|
135
|
-
|
|
136
|
-
| Critère | `/goal` (natif) | Ralph (`ralph.sh`) |
|
|
137
|
-
|---------|-----------------|--------------------|
|
|
138
|
-
| Critère de fin | Jugement LLM : un petit modèle rapide évalue, après chaque tour, ce qui est visible dans la conversation — il ne lance aucune commande et ne lit aucun fichier | Déterministe : commandes, patterns et hooks exécutés par `dod-validator.sh` |
|
|
139
|
-
| Plafond d'itérations | Aucun plafond fixe : fin quand la condition est atteinte ou jugée impossible, sur `/goal clear`, ou si Claude enchaîne plusieurs tours sans outil ; une borne peut être écrite dans la condition | `max_iterations` + circuit breaker adaptatif |
|
|
140
|
-
| Headless / CI | `claude -p "/goal ..."` boucle jusqu'au bout en une seule invocation ; usage en CI non documenté | Conçu pour : exécution autonome (`--autonomous`, `--overnight`), dashboard `headless` |
|
|
141
|
-
| `disableAllHooks: true` | Indisponible : `/goal` enveloppe un Stop hook de session (idem avec `allowManagedHooksOnly`) | La boucle et la DoD tournent dans `ralph.sh` ; seule l'intégration hooks (`stop-dod-gate.sh`) est désactivée |
|
|
142
|
-
| Configuration | Aucune ; un seul goal par session, effacé par `/clear` | `ralph.yml` |
|
|
143
|
-
|
|
144
|
-
- **Choisir `/goal`** pour une tâche interactive dont la fin se constate dans la conversation (afficher la sortie des tests : l'évaluateur ne les lance pas).
|
|
145
|
-
- **Choisir Ralph** quand la fin doit être prouvée par des commandes, avec un plafond d'itérations, en autonomie ou en CI.
|
|
146
|
-
|
|
147
|
-
Référence : [documentation `/goal`](https://code.claude.com/docs/en/goal).
|
|
148
|
-
|
|
149
103
|
## Exemples de démarrage rapide
|
|
150
104
|
|
|
151
105
|
```bash
|
|
@@ -76,35 +76,6 @@ O sistema de DoD valida a conclusão por meio de múltiplos critérios:
|
|
|
76
76
|
| `hook` | Executar hook existente do Claude |
|
|
77
77
|
| `human` | Validação humana interativa |
|
|
78
78
|
|
|
79
|
-
#### Limite do validador `command` → usar `type: hook`
|
|
80
|
-
|
|
81
|
-
`type: command` só executa comandos que passam na lista de permissões de `Tools/Ralph/lib/dod-validator.sh`: letras, dígitos, espaço e `_ . / " ' -`. Qualquer comando contendo `=`, `|`, `&` ou `:` (assim como `;`, `>`, `$` etc.) é rejeitado com `ERROR: Invalid command`: o critério **nunca** é validado e o loop roda até o circuit breaker disparar.
|
|
82
|
-
|
|
83
|
-
| Comando rejeitado | Caractere culpado |
|
|
84
|
-
|-------------------|-------------------|
|
|
85
|
-
| `dotnet build /p:TreatWarningsAsErrors=true` | `:` e `=` |
|
|
86
|
-
| `npm test && npm run lint` | `&` |
|
|
87
|
-
| `pytest \| tee out.log` | `\|` |
|
|
88
|
-
|
|
89
|
-
Nesses casos, colocar o comando em um script e declará-lo como `type: hook`:
|
|
90
|
-
|
|
91
|
-
```yaml
|
|
92
|
-
definition_of_done:
|
|
93
|
-
checklist:
|
|
94
|
-
- id: build
|
|
95
|
-
type: hook
|
|
96
|
-
script: ".ralph/dod/build.sh"
|
|
97
|
-
required: true
|
|
98
|
-
```
|
|
99
|
-
|
|
100
|
-
```bash
|
|
101
|
-
#!/usr/bin/env bash
|
|
102
|
-
# .ralph/dod/build.sh — recebe {"cwd": ..., "output": ...} em JSON via stdin
|
|
103
|
-
dotnet build /p:TreatWarningsAsErrors=true || exit 2
|
|
104
|
-
```
|
|
105
|
-
|
|
106
|
-
O script retorna `0` (critério atendido) ou `2` (critério falhou). Chaves lidas por tipo: `command` → `command:`, `output_contains` e `file_changed` → `pattern:`, `hook` → `script:`, `human` → `prompt:`.
|
|
107
|
-
|
|
108
79
|
### 4. Circuit Breaker Adaptativo (v2.0)
|
|
109
80
|
|
|
110
81
|
Seleciona automaticamente o perfil com base nas palavras-chave da tarefa:
|
|
@@ -129,23 +100,6 @@ Claude trabalha...
|
|
|
129
100
|
Stop → stop-dod-gate.sh → Bloquear se DoD não satisfeito (exit 2)
|
|
130
101
|
```
|
|
131
102
|
|
|
132
|
-
## /goal (interativo) ou Ralph (autônomo/CI)?
|
|
133
|
-
|
|
134
|
-
`/goal` é um comando nativo do Claude Code digitado dentro de uma sessão: não é uma opção de `ralph.sh`, que não oferece nenhuma para isso e nunca gera um `/goal`. As duas ferramentas são complementares:
|
|
135
|
-
|
|
136
|
-
| Critério | `/goal` (nativo) | Ralph (`ralph.sh`) |
|
|
137
|
-
|----------|------------------|--------------------|
|
|
138
|
-
| Critério de conclusão | Julgamento LLM: um modelo pequeno e rápido avalia, após cada turno, o que está visível na conversa — não executa nenhum comando nem lê nenhum arquivo | Determinístico: comandos, padrões e hooks executados por `dod-validator.sh` |
|
|
139
|
-
| Limite de iterações | Sem limite fixo: termina quando a condição é atendida ou julgada impossível, com `/goal clear`, ou se Claude encadeia vários turnos sem ferramentas; um limite pode ser escrito na condição | `max_iterations` + circuit breaker adaptativo |
|
|
140
|
-
| Headless / CI | `claude -p "/goal ..."` itera até o fim em uma única invocação; uso em CI não documentado | Feito para isso: execução autônoma (`--autonomous`, `--overnight`), dashboard `headless` |
|
|
141
|
-
| `disableAllHooks: true` | Indisponível: `/goal` envolve um Stop hook de sessão (idem com `allowManagedHooksOnly`) | O loop e a DoD rodam dentro de `ralph.sh`; só a integração de hooks (`stop-dod-gate.sh`) é desativada |
|
|
142
|
-
| Configuração | Nenhuma; um único goal por sessão, apagado por `/clear` | `ralph.yml` |
|
|
143
|
-
|
|
144
|
-
- **Escolher `/goal`** para uma tarefa interativa cuja conclusão aparece na conversa (exibir a saída dos testes: o avaliador não os executa).
|
|
145
|
-
- **Escolher Ralph** quando a conclusão deve ser comprovada por comandos, com limite de iterações, sem supervisão ou em CI.
|
|
146
|
-
|
|
147
|
-
Referência: [documentação do `/goal`](https://code.claude.com/docs/en/goal).
|
|
148
|
-
|
|
149
103
|
## Exemplos de Início Rápido
|
|
150
104
|
|
|
151
105
|
```bash
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@the-bearded-bear/claude-craft",
|
|
3
|
-
"version": "8.26.0
|
|
3
|
+
"version": "8.26.0",
|
|
4
4
|
"description": "A comprehensive framework for AI-assisted development with Claude Code. Install standardized rules, agents, and commands for your projects.",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"main": "cli/index.js",
|