qwen3:8b sometimes writes a tool call as plain text instead of making it

Cerrado
#532 0 comentarios 0 reacciones 0 asignados Ver en GitHub

Los mantenedores suelen responder en 1 día

@itsskofficial ya está trabajando en esto.

Desde el 31/10/2026.

  • #538 de @itsskofficial — fusionado

Evaluación

Dificultad
4/5
Tiempo estimado
1-2 días
Aptitud para principiantes
48/100
Tipo de issue
Error
Claridad
Bien especificado
Estado de actividad
Activo
Stack tecnológico
python

Línea de trabajo

Start by reading sentient/agent/loop.py and searching sentient/llm/ for existing text-to-tool-call parsing. Reproduce the issue with qwen3:8b and inspect the written-out memory_remember call, including any <tool_call> wrapper. Done when it runs through the same safety checks or gets one nudge, and raw call text is never shown to the user.

Escrito por el modelo de indexación a partir del texto del issue.

Descripción

area: engine bug priority: P2 size: S

Seen while testing #531 (#526) on ollama_chat/qwen3:8b. One chat reply contained the memory_remember call written out as text, the tool name and its JSON arguments, rather than an actual tool call. Nothing was saved, and the user saw the raw call in the reply.

To look at

  • Whether the agent loop already recovers tool calls written as text for small models (search sentient/agent/loop.py and sentient/llm/ for any text-to-tool-call parsing), and why it didn't catch this one: its shape, or a <tool_call> wrapper qwen uses.
  • If not, add a tolerant parser: when a reply is only a tool call written out (a known tool name plus a JSON object), run it as a call. Use the same safety checks as a real call, so the risk rules still apply. Otherwise nudge the model once to make the call properly.
  • Never show the raw call text to the user.

Done when: in a real run on qwen3:8b, a written-out memory_remember either runs as a real call or gets one nudge, and the user never sees the raw text.

Lenguaje dominante
Python
Estrellas
716
Forks
132
Merge medio
15 h 24 min
PR fusionados (30 d)
211

Preparar el entorno

Primeros pasos

  1. Lee el issue completo y luego la guía de contribución del proyecto.
  2. Comenta en el issue que vas a ocuparte — evita que dos personas hagan lo mismo.
  3. Haz un fork del repositorio y trabaja en una rama.
  4. Abre un pull request que haga referencia al número del issue.

Más de existence-master/Sentient

Todos los issues de existence-master/Sentient

Issues similares

Más issues de Python

Recibe los nuevos issues en tu correo

Un resumen breve de issues de GitHub para principiantes.