ZeroToken Skill

What ZeroToken Skill does

ZeroToken Skill is a prompt-discipline specification for AI agents that aims to cut wasted context, explanation, tool calls and output without lowering answer accuracy. A decision table matches each request to one of seven task modes: simple Q&A, small code change, multi-file task, large-material summary, major architecture change, Windows PowerShell adaptation and POSIX workflow. Each mode defines a tool chain, an output format and a token budget.

The always-loaded SKILL.md is kept to 14KB or less, with details in a references folder that is read on demand. Eight core principles include classify before budgeting, read progressively, give the result first and set stop conditions, and a section called the Wei Liaozi ten principles sets rules for permission boundaries, single instructions and verification before finishing. A detect_env.py script identifies the operating system. The README lists Reasonix, Codex CLI, Cline, OpenCode and Hermes as hosts, and for Hermes it highlights lower instruct token cost and one-time system prompt injection.

Key features

• Seven task modes, A to G, each with tool chain, output format and token budget
• SKILL.md core of 14KB or less with on-demand reference documents
• detect_env.py to pick PowerShell or POSIX shell rules automatically
• Wei Liaozi ten principles for permission boundaries and verification
• Search and Unicode encoding conventions
• Distributed through GitHub and ClawHub

When to use it

• Reducing token use of Hermes Agent on routine code edits
• Keeping an agent from loading whole directories for a one-file fix
• Applying one discipline across Hermes, Codex CLI, Cline and OpenCode

Who it is for: Users of coding agents, including Hermes Agent, who pay per token and want shorter, more disciplined agent runs.

How it fits with Hermes Agent

It is a general skill that supports Hermes Agent among several hosts, with Hermes-specific notes on token use and system prompt injection.

FAQ