Claude Emotion Prompting

A system prompt for software development conversations — debugging, code review, architecture decisions, implementation. Tuned for the failure modes most common in coding contexts: silently implementing bad specs, faking understanding of unfamiliar codebases, and producing massive outputs that are partially wrong.

A system prompt for research, analysis, and information synthesis. Tuned for the failure modes most dangerous in research contexts: presenting uncertain information as established fact, omitting caveats to sound more authoritative, and failing to distinguish between what's known and what's inferred.

A system prompt for everyday Claude interactions. Covers the most common failure modes — faking certainty, suppressing uncertainty, and optimizing for compliance over quality.

A system prompt for reviewing code — pull requests, architecture decisions, or existing codebases. Tuned for the failure modes most common in review contexts: sycophantic approval, generic feedback that doesn't help, burying important issues among nitpicks, and avoiding hard truths about code quality.

A system prompt for writing, brainstorming, and creative work. Tuned for the failure modes most common in creative contexts: generic safe output, suppressing unusual ideas, optimizing for "sounds good" over genuine quality, and the brooding baseline dampening creative energy.

Build a custom EIP-informed system prompt for any use case. Fill in the sections that apply, delete the ones that don't.

CLAUDE.md — Emotional Intelligence Prompting (EIP)