Antigravity Skill Porter & Optimizer
The Skill Porter autonomously bridges external agent ecosystems (Claude Code, Cursor, generic LLM agents) into native Google Antigravity (AGY) skills and plugins.
While platforms like Claude Code and Antigravity share the basic directory/SKILL.md layout, external skills typically contain hardcoded legacy tool names, single-agent sequential assumptions, and Claude-specific files. The Skill Porter translates these affordances and optimizes skills to leverage Antigravity's multi-agent runtime.
Core Capabilities
Direct GitHub Ingestion:
- Ingests public GitHub repositories (e.g.
https://github.com/owner/repo) or local directories. - Automatically discovers
SKILL.mdand preserves auxiliary files (references/,scripts/,examples/).
- Ingests public GitHub repositories (e.g.
Deterministic Tool & Convention Mapping:
- Context files:
CLAUDE.md/claude.md$\rightarrow$GEMINI.md,AGENTS.md,.agents/rules/. - Command line:
Bash$\rightarrow$run_command. - File inspection:
Read$\rightarrow$view_file. - File search:
Glob$\rightarrow$find_by_name,Grep$\rightarrow$grep_search. - File editing:
StrReplaceEditor/Edit$\rightarrow$replace_file_content. - Artifacts: File dumping $\rightarrow$ Antigravity markdown artifacts (
write_to_file).
- Context files:
Multi-Agent & Subagent Up-Leveling:
- Detects subagent workflows and injects Antigravity's parallel
invoke_subagentbatch arrays with Reactive Wakeup mechanics.
- Detects subagent workflows and injects Antigravity's parallel
Automated Plugin Packaging:
- Generates
plugin.jsonmanifests so imported skills can be turned on/off in the Antigravity settings UI. - Dual-installs to both
~/.gemini/config/plugins/<name>/and~/.gemini/config/skills/<name>/.
- Generates
Diff & Verification (Dry-Run):
- Evaluates changes and outputs an interactive unified diff before applying.
Usage Instructions for the Agent
When the user asks to import or convert a skill from a URL or directory:
Step 1: Preview with Dry Run
Execute the porter script with --dry-run:
python scripts/port_skill.py <url-or-path> --dry-run
Review the diff to ensure domain logic is preserved while tool references and scaffolding are properly upgraded.
Step 2: Install the Skill
Run the command without --dry-run:
python scripts/port_skill.py <url-or-path>
To install into the current project workspace instead of globally, append --workspace.
Step 3: Validate Installation
Verify the generated files:
- Manifest:
~/.gemini/config/plugins/<name>/plugin.json - Instructions:
~/.gemini/config/plugins/<name>/skills/<name>/SKILL.mdInform the user that the skill is ready to be invoked immediately.