Overview
viberuler is a command-line benchmark for developers who use AI coding agents. It scans local activity and produces a shareable VIBE SCORE covering AI usage, code output, projects, commit cadence, and achievements.
The tool frames its results as a personal mirror rather than a performance-review system. Measurements are scoped to one person and one machine, while comparisons depend on users choosing to participate.
Key Metrics
- Tracks tokens consumed from Claude Code and Codex session logs, with optional LiteLLM gateway data.
- Calculates tokens per dollar using a bundled pricing table and reports API-equivalent spend by platform.
- Counts lines added in commits authored by the user instead of measuring total repository size.
- Separates generated files, lockfiles, bundles, and similar machine output from scored authored lines.
- Summarizes repositories with authored commits, consecutive commit days, and commits per active day.
- Detects known coding agents through their standard marker directories.
Scores and Recognition
viberuler turns the collected signals into a scorecard with ranks ranging from Prompt Peasant through Singularity Adjacent. When no qualifying data is found, it assigns an NPC status.
Achievements recognize patterns such as high token volume, strong cache-read ratios, work across multiple languages, substantial authored code in one repository, long commit streaks, and repeated late-night commits. The excluded generated-code total remains visible as context even though it does not contribute to the score.
Local Data and Intended Use
The scan runs locally and presents the result within seconds. Codex plan headroom can be read from rate-limit records already present in session logs, but this information is not scored or sent.
The project explicitly discourages using these measurements to evaluate employees. Its metrics are designed for self-observation, voluntary comparison, and understanding the relationship between AI token consumption and code shipped.