agent-interface
Agent Interface
Design and check CLI command and MCP tool names against a noun-verb standard, and how a CLI behaves for agents: output, errors, paging, prompts and safe retries.
Install
Claude Code
claude plugin marketplace add hsb3/agent-interface
claude plugin install agent-interface@agent-interface
Codex
codex plugin marketplace add hsb3/agent-interface
codex plugin add agent-interface@agent-interface
Antigravity
agy plugin install https://github.com/hsb3/agent-interface
Start a new session after installing. Antigravity support is untested; see supported hosts.
Try it
- Names. Ask your agent “check my CLI’s command names” and name the CLI,
for example
mytool. In Claude Code you can also run/agent-interface:menu-design; in Codex, pick the Menu Design skill. The agent lists the CLI’s commands, runs the check, and replies with the number of violations and advisories and the names to change. - Behavior. Ask “check how
mytoolbehaves for agents”. In Claude Code you can also run/agent-interface:cli-behavior; in Codex, pick the CLI Behavior skill. The agent reads the CLI’s source, records a result for each rule and command, and writes a plan that brings the CLI to the standard, with any optional deviations for you to accept or decline.
Skills
| Skill | Use it to… |
|---|---|
 Menu Design | Design and check CLI command and MCP tool names against a noun-verb standard, and plan renames that list affected callers. |
 CLI Behavior | Check and design how a CLI behaves for agents: machine output, errors, paging, prompts and safe retries, with a plan to full compliance. |