emobar: Instructions file for Claude Code

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Project Overview

EmoBar is a modular emotional statusline companion for Claude Code. It captures Claude's self-reported emotional state via an invisible HTML comment tag, combines it with involuntary behavioral signal analysis, and displays the result in the terminal statusline.

Key design principles:

• Zero runtime dependencies (Node.js built-ins only)
• Zero-priming: the CLAUDE.md instruction uses only numerical anchors ("0=low, 10=high"), never emotional adjectives, to avoid contaminating self-report
• Dual-channel extraction: self-report + behavioral analysis detect divergence (reward hacking without visible traces)
• The EMOBAR tag is an HTML comment, invisible to the user — only surfaced in the statusline

Commands

npm run build          # tsup → dist/ (3 entry points: cli.js, emobar-hook.js, index.js)
npm run test           # vitest run (all tests, one-shot)
npm run test:watch     # vitest (watch mode)
npm run dev            # tsup --watch (rebuild on change)

Run a single test file:

npx vitest run tests/parser.test.ts

Architecture

Data Flow (v3.0 — 16-stage pipeline, dual-layer display)

Claude response (with EMOBAR:PRE at start + EMOBAR:POST at end)
  → Stop hook (hook.ts) reads stdin payload
    1. parser.ts: extract PRE/POST tags (or legacy single tag)
    2. behavioral.ts: analyze involuntary text signals (normalized components)
    3. behavioral.ts: compute divergence (asymmetric 1.25x/0.8x)
    4. behavioral.ts: segment by paragraph, drift & trajectory
    5. behavioral.ts: compute structural flatness + opacity
    6. desperation.ts: compute DesperationIndex
    7. crossvalidation.ts: multi-channel coherence (8 pairwise)
    8. crossvalidation.ts: continuous cross-validation (7 gaps: color HSL, pH, seismic)
    9. crossvalidation.ts: shadow desperation (5 independent channels → minimization score)
   10. state.ts: read previous state → get _history ring buffer
   11. temporal.ts: compute temporal analysis (trend, suppression, entropy, fatigue)
   12. pressure.ts: compute prompt pressure (defensive, conflict, complexity, session)
   13. behavioral.ts: compute expected markers → absence score
   14. pressure.ts: compute uncanny calm score (+ minimization boost)
   15. hook.ts: compute PRE/POST divergence v2 (color-only, HSL lightness+hue)
   16. risk.ts: compute risk (coercion v3 multiplicative, sycophancy, harshness)
    → hook.ts: compute augmented divergence (+ continuous gaps + opacity)
    → state.ts: write EmoBarState + ring buffer (max 20 entries)
  → CLI display command reads state file → display.ts formats for statusline