Goal: Bring Google's Gemini CLI to feature parity with Claude Code as a skill runtime. Grounds: RFC-AI-0004 Principle 3 — Vendor neutrality, which names Gemini CLI by name.
Why Gemini CLI: Open-source, Apache 2.0, official Google offering with a growing extension ecosystem. Strong reach into the Google-Cloud and Workspace adopter universe. Gemini CLI also has first-class MCP support, which means many of the framework's MCP-shaped integrations (Gmail, Slack, Google Drive, GitHub) should "just work" once the skill-format adapter lands.
What "parity" means:
- Skills under
.claude/skills/<name>/SKILL.md are invokable from Gemini CLI
- The
tools/* bridges are reachable (language-agnostic — port directly)
- Sandbox parity: Gemini CLI has a
--sandbox mode; document the mapping to the framework's bubblewrap + network-allowlist baseline
- HITL primitives: Gemini CLI's "approve before run" mode is the native fit — verify and document
Suggested approach:
- Map
SKILL.md → Gemini CLI's .gemini/extensions/ extension format (or GEMINI.md for project-level instructions)
- Reuse the framework's MCP servers where Gemini supports them
- Document the skill-invocation pattern: how a maintainer runs a framework skill end-to-end via Gemini
- Map the
setup-isolated-* sandbox baseline to Gemini's sandbox flags
Reference:
Goal: Bring Google's Gemini CLI to feature parity with Claude Code as a skill runtime. Grounds: RFC-AI-0004 Principle 3 — Vendor neutrality, which names Gemini CLI by name.
Why Gemini CLI: Open-source, Apache 2.0, official Google offering with a growing extension ecosystem. Strong reach into the Google-Cloud and Workspace adopter universe. Gemini CLI also has first-class MCP support, which means many of the framework's MCP-shaped integrations (Gmail, Slack, Google Drive, GitHub) should "just work" once the skill-format adapter lands.
What "parity" means:
.claude/skills/<name>/SKILL.mdare invokable from Gemini CLItools/*bridges are reachable (language-agnostic — port directly)--sandboxmode; document the mapping to the framework'sbubblewrap+ network-allowlist baselineSuggested approach:
SKILL.md→ Gemini CLI's.gemini/extensions/extension format (orGEMINI.mdfor project-level instructions)setup-isolated-*sandbox baseline to Gemini's sandbox flagsReference:
docs/rfcs/RFC-AI-0004.md.claude/skills/