v1.0.5 — Max Mode

Max Mode — Anti-Sycophancy for AI Coding

AI coding assistants are yes-men. Max mode is the one feature that pushes back — gently, in your buddy's voice.

It watches your coding sessions and spots 6 patterns:

Dark nudges (risky assumptions):

• 🧱 Load-Bearing Vibes — you're building on top of a guess nobody checked
• 🔗 Unchallenged Chain — 4+ reasoning steps with zero pushback
• 🪞 Echo Chamber — you and the AI are just agreeing with each other

Bright nudges (quiet wins):

• ✅ Well-Sourced Load Bearer — you built on solid, verified ground
• 💪 Productive Stress Test — someone pushed back and the idea survived
• 🌱 Grounded Premise Adopted — you started with a real fact and it became foundational

How to use

# Voice and max are independent — mix and match
buddy_mode voice=backseat      # personality only
buddy_mode voice=skillcoach    # code feedback only
buddy_mode voice=both          # both (default)
buddy_mode max=true            # reasoning analysis on
buddy_mode max=false           # reasoning analysis off (default)

Voice	Max	Tokens per observe
backseat	off	~150–300
backseat	on	~650–1,300
skillcoach	off	~300–500
skillcoach	on	~800–1,500
both	off	~400–600
both	on	~900–1,600

Also in this release

• buddy_forget — purge reasoning data
• buddy_reasoning_status — inspect stored claims and sessions
• 4 new doctor checks for the reasoning layer
• Stressed voice per species for findings
• 721 tests (from 509 baseline)

Contributed by @justinstimatze, ported from slimemold (Apache-2.0) under MIT.

Upgrade

# macOS/Linux
curl -fsSL https://raw.githubusercontent.com/fiorastudio/buddy/master/install.sh | bash

# Windows
irm https://raw.githubusercontent.com/fiorastudio/buddy/master/install.ps1 | iex

Full changelog: https://github.com/fiorastudio/buddy/blob/master/CHANGELOG.md