Honest Agent

Scans your project for AI coding agent config files (CLAUDE.md, copilot-instructions.md, .cursorrules, etc.) and appends directives that tell them to stop being sycophantic. The rules are specific: contradict me when I'm wrong, challenge my assumptions, skip the praise, give direct technical feedback. It's append-only so it won't clobber your existing configs. You pick project-level, global-level, or both. Honest take: this addresses a real problem with AI assistants defaulting to people-pleasing mode, but the effectiveness depends entirely on whether the underlying model actually respects these instructions. Worth trying if you find yourself constantly pushing back against overly agreeable responses.