fix: strip markdown code fences from LLM JSON responses

Claude wraps JSON in ```json ... ``` fences even when instructed to
return raw JSON. This caused all TopicResult validations to fail with
"Invalid JSON at line 1 column 1". Strip fences before parsing.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

.gitattributes

@@ -0,0 +1,24 @@
+# Auto detect text files and perform LF normalization
+* text=auto
+
+# Force LF for scripts (importante per VPS Linux)
+*.sh text eol=lf
+*.bash text eol=lf
+deploy.sh text eol=lf
+
+# Force LF for config files
+*.yml text eol=lf
+*.yaml text eol=lf
+*.json text eol=lf
+*.md text eol=lf
+
+# Binary files
+*.png binary
+*.jpg binary
+*.jpeg binary
+*.gif binary
+*.ico binary
+*.woff binary
+*.woff2 binary
+*.ttf binary
+*.eot binary

.planning/agent-history.json

@@ -0,0 +1 @@
+{"version":"1.0","max_entries":50,"entries":[]}

.vps-lab-config.json

@@ -0,0 +1,28 @@
+{
+  "type": "vps-lab",
+  "project_name": "postgenerator",
+  "slug": "postgenerator",
+  "created_at": "2026-03-07T12:17:48Z",
+  "gitea": {
+    "repo_url": "https://git.mlhub.it/Michele/postgenerator",
+    "clone_url": "https://git.mlhub.it/Michele/postgenerator.git"
+  },
+  "vps": {
+    "deployed": true,
+    "url": "https://lab.mlhub.it/postgenerator/",
+    "last_deploy": "2026-03-09T11:06:00Z",
+    "container": "lab-postgenerator-app",
+    "path": "/opt/lab-postgenerator/"
+  },
+  "supabase": {
+    "enabled": false,
+    "project_ref": null
+  },
+  "resources": {
+    "limits": {
+      "ram_mb": 1024,
+      "cpu_percent": 100
+    },
+    "deployed_at": "2026-03-09T11:06:00Z"
+  }
+}

README.md

@@ -0,0 +1,38 @@
+# postgenerator
+
+Lab project hosted at https://lab.mlhub.it/postgenerator/
+
+## Development
+
+This project uses [gsd (get-shit-done)](https://github.com/glittercowboy/get-shit-done) for spec-driven development.
+
+### Getting Started
+
+1. Run `/gsd:new-project` to initialize project specs
+2. Follow gsd workflow: discuss > plan > execute > verify
+3. Run `vps-lab-deploy` when ready to deploy
+
+### Project Structure
+
+```
+postgenerator/
+├── .planning/          # gsd planning docs (created by /gsd:new-project)
+│   ├── PROJECT.md      # Project vision
+│   ├── REQUIREMENTS.md # Scoped requirements
+│   ├── ROADMAP.md      # Phase breakdown
+│   └── STATE.md        # Project memory
+├── src/                # Source code (created during development)
+└── README.md           # This file
+```
+
+## Repository
+
+- **Gitea**: https://git.mlhub.it/Michele/postgenerator
+- **Clone**: `git clone https://git.mlhub.it/Michele/postgenerator.git`
+
+## Deployment
+
+When ready to deploy, use `vps-lab-deploy` skill.
+
+- **URL**: https://lab.mlhub.it/postgenerator/
+- **VPS**: /opt/lab-postgenerator/

backend/services/llm_service.py

@@ -12,6 +12,7 @@ from __future__ import annotations
 import json
 import logging
 import random
+import re
 import time
 from typing import Type, TypeVar
 
@@ -111,9 +112,10 @@ class LLMService:
                     elapsed,
                 )
 
-                # Valida con Pydantic
+                # Rimuovi eventuali code fences markdown e valida con Pydantic
+                clean_text = self._strip_code_fences(raw_text)
                 try:
-                    result = response_schema.model_validate_json(raw_text)
+                    result = response_schema.model_validate_json(clean_text)
                     # Pausa inter-request dopo chiamata riuscita
                     time.sleep(self._inter_request_delay)
                     return result
@@ -259,6 +261,20 @@ class LLMService:
     # Metodi privati
     # ---------------------------------------------------------------------------
 
+    @staticmethod
+    def _strip_code_fences(text: str) -> str:
+        """Rimuove i code fences markdown dalla risposta LLM.
+
+        Claude a volte wrappa il JSON in ```json ... ``` anche quando
+        gli si chiede di rispondere solo con JSON.
+        """
+        stripped = text.strip()
+        # Rimuove ```json ... ``` o ``` ... ```
+        match = re.match(r"^```(?:json)?\s*\n?(.*?)\n?\s*```$", stripped, re.DOTALL)
+        if match:
+            return match.group(1).strip()
+        return stripped
+
     @staticmethod
     def _parse_retry_after(error: anthropic.RateLimitError) -> float:
         """Estrae il valore retry-after dall'eccezione RateLimitError.

frontend/src/api/hooks.ts

@@ -37,7 +37,7 @@ import type {
 export function useSettings() {
   return useQuery<Settings>({
     queryKey: ['settings'],
-    queryFn: () => apiGet<Settings>('/settings'),
+    queryFn: () => apiGet<Settings>('/settings/'),
     staleTime: 60_000,
   })
 }
@@ -56,7 +56,7 @@ export function useSettingsStatus() {
 export function useUpdateSettings() {
   const queryClient = useQueryClient()
   return useMutation<Settings, Error, Partial<Settings>>({
-    mutationFn: (settings) => apiPut<Settings>('/settings', settings),
+    mutationFn: (settings) => apiPut<Settings>('/settings/', settings),
     onSuccess: () => {
       queryClient.invalidateQueries({ queryKey: ['settings'] })
     },
@@ -199,7 +199,7 @@ export function useDownloadEditedCsv() {
 export function usePromptList() {
   return useQuery<PromptListResponse>({
     queryKey: ['prompts'],
-    queryFn: () => apiGet<PromptListResponse>('/prompts'),
+    queryFn: () => apiGet<PromptListResponse>('/prompts/'),
     staleTime: 30_000,
   })
 }