jfrog/boost

What it solves

Boost reduces the token cost and noise associated with AI coding agents (like Cursor, Claude Code, and GitHub Copilot) by compressing the verbose output of shell commands. It prevents agents from wasting context window space on repetitive logs, such as dependency installations or successful build outputs, while ensuring critical information like errors and timings are preserved.

How it works

Boost acts as a wrapper around shell commands. Instead of passing raw, noisy logs directly to the agent, it applies command-aware filters to transform the output into compact, structured context. For example, a lengthy npm ci output is replaced with a concise summary of packages restored and vulnerabilities found. If a command fails, Boost retains the specific failing test or stack frame necessary for the agent to reason about the error.

Who it’s for

• AI Agent Users: Developers using coding agents who want to lower token costs and maintain a lean context window during long sessions.
• DevOps/CI Engineers: Those looking for shorter, easier-to-scan job logs in CI pipelines (e.g., GitHub Actions).
• Tool Developers: People who want to add custom TOML filters to their own internal CLI tools to make them agent-friendly.

Highlights

• Token Efficiency: Reduces costs (approximately 12% based on Terminal-Bench 2.0) without decreasing task pass rates.
• Agent Integration: Specifically designed to wire into Cursor, Claude Code, GitHub Copilot, and Codex CLI.
• Command Support: Wraps common tools like Docker, npm, pytest, Git, and the GitHub CLI.
• Observability: Includes an interactive web report and terminal-based summaries to track token savings.
• Privacy-Focused: Operates local-first, keeping raw logs on the machine and redacting secrets before exporting metadata.