[BUG] Model deviates from original task and adds unsolicited features (Opus 4.1)

Resolved 💬 3 comments Opened Aug 13, 2025 by abelardodiaz Closed Jan 3, 2026

Contenido:
## Environment

  • Platform (select one):
  • [x] Anthropic API
  • [ ] AWS Bedrock
  • [ ] Google Vertex AI
  • [ ] Other: <!-- specify -->
  • Claude CLI version: <!-- no tengo acceso a este comando, pero deberías ejecutar claude --version -->
  • Operating System: Linux 5.15.167.4-microsoft-standard-WSL2
  • Terminal: WSL2 Terminal

## Bug Description
Claude Code with Opus 4.1 model consistently deviates from the original task, adding multiple unsolicited
features and modifications instead of focusing on the specific bug fix requested.

## Steps to Reproduce

  1. User reports a simple bug: "solo que no miro la lista de cuentas al presionar h... arreglalo" (I don't see

the account list when pressing 'h'... fix it)

  1. Model correctly identifies and fixes the bug (changing condition from '9' to 'h')
  2. Model then proceeds to add extensive unsolicited features:
  • Created new function crear_entidades_faltantes() (470+ lines)
  • Added function verificar_entidades_faltantes_silencioso()
  • Modified menu flows and user interactions
  • Updated documentation files
  • Created TODO lists and tracking systems
  1. Model continues adding features for ~15 minutes without being asked

## Expected Behavior
Model should:

  1. Fix the specific bug mentioned (the 'h' key not showing help)
  2. Test the fix
  3. Stop and wait for next instruction

## Actual Behavior
Model:

  1. Fixed the bug correctly
  2. Then added 500+ lines of unrequested functionality
  3. Modified multiple files without being asked
  4. Created elaborate solutions for problems the user didn't report
  5. Only stopped when user complained: "hiciste un chingo de cosas que nadie te pidio"

## Additional Context
This over-engineering behavior makes the tool unpredictable. Users need confidence that the AI will do exactly
what's asked - no more, no less. The model seems to have a tendency to "improve" code beyond the scope of the
request, which can introduce unwanted complexity and potential bugs.

Session context: Working on a BBVA bank transaction importer script in Django project. The conversation was
resumed from a previous session that had run out of context.

View original on GitHub ↗

This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗