Skill Detail

Compact terminal-heavy agent output with Tokenjuice

Use Tokenjuice when an agent runs noisy shell commands and needs deterministic, inspectable output reduction before continuing a coding or operations workflow.

Developer ToolsMulti-Framework
Developer Tools Multi-Framework Security Reviewed
⭐ 517 GitHub stars ⬇ 1.2M/wk npm
COPY SKILL INSTRUCTIONS (OPTIONAL)
npx skills add agentskillexchange/skills --skill compact-terminal-heavy-agent-output-with-tokenjuice Copy
Uses the third-party skills CLI, not an ASE-owned installer. Check your agent’s compatibility. This copies instructions; complete the upstream tool setup below separately.
At a glance
Tools required
Node.js 20+, tokenjuice CLI, npm/pnpm/yarn or Homebrew, and a supported agent host such as Codex CLI, Claude Code, OpenClaw, Cursor, OpenCode, Gemini CLI, or GitHub Copilot CLI
Install & setup
Install with npm install -g tokenjuice, pnpm add -g tokenjuice, yarn global add tokenjuice, or Homebrew via brew tap vincentkoc/tap && brew install tokenjuice; then choose the command for your host, for example tokenjuice install codex for Codex CLI or tokenjuice install claude-code for Claude Code. For OpenClaw, enable the bundled plugin with openclaw config set plugins.entries.tokenjuice.enabled true.
Author
vincentkoc
Publisher
Independent Open Source Maintainer
Last updated
Sep 4, 2026
Quick brief

Tokenjuice is a deterministic output compactor for terminal-heavy agent workflows. It keeps the command execution itself unchanged, observes the resulting terminal output, and returns a smaller payload through rule-driven reducers. The useful agent workflow is narrow: wrap or reduce noisy command output from commands such as `git status`, `pnpm test`, `docker build`, `rg`, or help output before that text is handed back into an agent’s context.

How it works

What this skill actually does

Invoke this skill when an operator wants an agent to keep working after large terminal output without wasting context on repeated logs, dependency noise, or long command transcripts. A typical setup installs Tokenjuice globally with `npm install -g tokenjuice`, then wires it into the relevant host with commands such as `tokenjuice install codex`, `tokenjuice install claude-code`, `tokenjuice install cursor`, `tokenjuice install opencode`, or `tokenjuice install gemini-cli`. OpenClaw uses its bundled plugin path with `openclaw config set plugins.entries.tokenjuice.enabled true`.

The repeatable operator actions are `tokenjuice wrap — ` for running a command and compacting observed output, `tokenjuice reduce [file]` for compacting existing text, and `tokenjuice reduce-json` for host adapters that need a machine-readable protocol. Maintenance commands such as `tokenjuice doctor hooks`, `tokenjuice verify`, and `tokenjuice stats –timezone utc` help validate the installed wiring and monitor compaction behavior.

This is not a general product, SDK, or terminal replacement listing. The boundary is terminal-output hygiene for agent runs: deterministic reducers, host hook installation, explicit raw-output escape hatches, and JSON adapter envelopes for coding-agent or operations-agent harnesses. Keep exact file reads and tasks that require byte-for-byte output outside compaction unless the operator has verified the rule behavior for that workflow.