When I used kimi-k2-thinking, the JSON args seems to be polluted, which might happen with other thinking models too.
Maybe use a try/except to use default values when there is a problem on line 44 of explore.jinx ?
Line 44 : max_iters = int(context.get('max_iterations') or '15')
🤖 picard | 👥 npcsh | kimi-k2-thinking
> [pasted: 22 lines, 4716 chars]
[iter 1] 0 msgs, last roles: []
[iter 1] prompt: Crée une application « terminal » en mode texte, mais stylée, avec ergonomie à l’état de l’art pour une appli texte, de ...
[iter 1] model text: Je vais analyser cette demande pour créer une application terminal stylée pour la création de tokens sur Solana. Permettez-moi de commencer par explor...
[iter 1] tools called: explore
⚡ explore[JINX-ERROR] Error executing step 'explore_codebase' in jinx 'explore': ValueError: invalid literal for int() with base 10: "}newcommand{'7'}"
✓
[iter 1] explore args: {'path': '.', 'task': "Explorer la structure du projet actuel pour comprendre l'architecture existante, les fichiers de configuration, et les dépendan
[iter 1] explore result: Error executing step 'explore_codebase' in jinx 'explore': ValueError: invalid literal for int() with base 10: "}newcommand{'7'}"
⚡ explore:
Error executing step 'explore_codebase' in jinx 'explore': ValueError: invalid literal for int() with base 10:
"}newcommand{'7'}"
[iter 2] 4 msgs, last roles: ['system', 'user', 'assistant', 'tool']
[iter 2] prompt: Continue. Call stop when done.
⣾ picard thinking...... [1:15] (type to queue, ESC to cancel)```
When I used kimi-k2-thinking, the JSON args seems to be polluted, which might happen with other thinking models too.
Maybe use a try/except to use default values when there is a problem on line 44 of explore.jinx ?
Line 44 : max_iters = int(context.get('max_iterations') or '15')