using-arkhe-skills

Using arkhe Skills Across Platforms

arkhe plugins ship the same SKILL.md files across CLI harnesses. This bootstrap establishes how to access skills and how to translate platform tool names into equivalent operations on Claude Code, Antigravity CLI (agy), Gemini CLI, and Codex CLI.

How to Access Skills

• Claude Code: Use the Skill tool. When invoked, the skill content is loaded and presented to follow directly.
• Antigravity CLI (agy): Skill metadata (name + description) is auto-loaded into the system prompt at session start. When triggers fire, use view_file to load skills/<skill-name>/SKILL.md from the plugin directory, then follow its instructions.
• Gemini CLI (Legacy): Skill metadata is loaded into system prompt. Use read_file to load skills/<skill-name>/SKILL.md.
• Codex CLI: Skill index and trigger phrases are inlined in AGENTS.md at install time. Follow the skill body directly when a trigger phrase fires.

Platform Tool Mapping

arkhe SKILL.md files reference core agent primitives. On each platform, substitute the equivalent operation from this table.

Primitive / Claude tool	Antigravity CLI (agy) equivalent	Gemini CLI (Legacy) equivalent	Codex CLI equivalent
AskUserQuestion	ask_question (native interactive modal with chip UI options)	Ask a plain-text question with numbered choices and await reply.	Plain-text question with numbered choices.
TaskCreate / TaskUpdate	manage_task or inline TODO list rendered on each turn	Inline TODO list rendered on each turn	Inline TODO list rendered on each turn
EnterPlanMode / ExitPlanMode	Native --mode plan / accept-edits mode	Announce plan mode and gate writes on approval	Announce plan mode and gate writes on approval
Skill tool (Skill)	view_file on skills/<name>/SKILL.md	read_file on skills/<name>/SKILL.md	Skill content inlined in AGENTS.md
Agent tool (subagent_type)	invoke_subagent / define_subagent (native background/concurrent subagent execution)	Inline agent prompt into current session	Inline agent prompt into current session

Command Naming Across Platforms

Commands ship from the same plugins/<plugin>/commands/<name>.md source on every platform, but the slash-command syntax differs:

Platform	Form	Example
Claude Code	/<plugin>:<command> (plugin-namespaced)	/core:think
Antigravity CLI (agy) / Gemini CLI	/<command> (or agy plugin commands)	/think
Codex CLI	Trigger phrase in chat (no slash)	"think through whether to use Postgres or Redis"

When this skill or other arkhe docs refer to core:think, core:debug, etc., resolve them through this table — same command, platform-specific invocation.

Subagent-Heavy Commands

Six arkhe commands dispatch subagents on Claude Code and Antigravity CLI (agy). On legacy Gemini and Codex CLI, they run inline with the agent's prompt collapsed into the command body.

The affected commands:

• core:debug (with --deep)
• core:think
• core:research
• core:double-check (with --deep)
• spring-boot:spring-review
• spring-boot:verify-upgrade

Instruction Priority

1. User's explicit instructions (CLAUDE.md, GEMINI.md, AGENTS.md, direct requests) — highest priority
2. arkhe skills — override default system behavior where they conflict
3. Default system prompt — lowest priority

If CLAUDE.md, GEMINI.md, or AGENTS.md says "do not use TDD" and a skill says "always use TDD," follow the user's instructions. The user is in control.

Where Skills Live

• Canonical source: plugins/<plugin>/skills/<skill>/SKILL.md
• Claude Code: loaded via .claude-plugin/marketplace.json
• Antigravity CLI / Gemini CLI: loaded via plugins/<plugin>/plugin.json or .gemini-extensions/<plugin>/plugin.json (skills directory is a symlink to the canonical source)
• Codex CLI: loaded via .codex-marketplace/<plugin>/AGENTS.md (skill index synthesized at build time; full content via symlinked skills/)

Skill bodies are not duplicated across platforms. One file, cross-platform support.