Dev.to
6/26/2026

I built a compiler for how AI agents should write to you
Short summary
The author built `/calibrate-comms`, an open-source Obsidian skill that calibrates how AI agents should communicate by measuring revealed preference across 9 communication axes, then compiling directives into a CLAUDE.md profile. Rather than personality typing, it validates preferences through pairwise samples and live corrections, making communication profiles durable and learnable.
- •Open-source skill (`/calibrate-comms`) for configuring AI agent communication preferences
- •9 communication axes calibrated via pairwise samples, not personality tests
- •Generates executable directives for Claude Code and Codex; works with Obsidian vaults and plain Markdown
Generated with AI, which can make mistakes.
Is this a good recommendation for you?



