jfrog/boost

What it solves

Boost reduces the token cost and noise of AI coding agents by compressing the output of shell commands. It prevents agents from being overwhelmed by long, repetitive logs (such as dependency installations or build outputs) while ensuring that critical information—like errors, timings, and cache hits—is preserved for the agent to reason about.

How it works

Boost acts as a wrapper around existing shell commands. It applies command-aware filters to the output of tools like npm, docker, pytest, and git. Instead of passing the entire raw log to the agent, Boost transforms noisy logs into compact, structured summaries. If a command fails, Boost retains the specific failing test or stack frame necessary for debugging, while keeping successful runs quiet.

Who it’s for

• AI Coding Agent Users: Developers using tools like Cursor, Claude Code, GitHub Copilot, and Codex CLI who want to lower token usage and maintain cleaner context windows.
• CIs and DevOps Engineers: Those looking for shorter, easier-to-scan job logs in GitHub Actions or other runners.
• Tool Developers: Users who can add custom TOML filters to compress the output of internal or proprietary CLI tools.

Highlights

• Command-Aware Compression: Preserves signal (errors, failures) while removing noise from common tools like npm ci or docker build.
• **- Agent Integration: Native support for wiring into popular AI coding assistants.
• **- Cost Reduction: Demonstrated ~12% lower cost with identical task pass rates according to Terminal-Bench 2.0.
• **- Observability: Includes an interactive web report and TUI for tracking token savings and performance.
• **- OpenTelemetry Support: Enabled for exporting metadata and usage statistics.