skill-porter

Skill Porter for Google Antigravity

When to Use

Use when adapting a locally obtained Claude Code, Cursor, Codex, or generic agent

skill bundle for Google Antigravity. The utility preserves support files and

previews limited tool-name substitutions; it does not prove client compatibility.

Prerequisites

• Python 3.9+; standard library only.
• A reviewed local skill directory containing SKILL.md, that file itself, or a

repository with skills/<lowercase-hyphenated-id>/SKILL.md directories.

• Verify the upstream identity, pinned revision, license and complete bundle first.

Obtain remote material separately through your reviewed download/clone workflow.

• Use stable local directories you control. Do not run against a tree being

modified by another process or user. Inspect scripts without executing them.

Workflow

1. From this skill directory, preview the local bundle:

   python3 scripts/port_skill.py --source "/absolute/path/my-skill" --dry-run

1. Review the diff. Only exact backtick-quoted tool identifiers such as View,

Edit, and Bash change. Frontmatter and support bytes remain intact. Verify

each target tool and its argument semantics in your actual client.

1. Copy to a fresh staging destination, then inspect before activating:

   python3 scripts/port_skill.py --source "/absolute/path/my-skill" --dest "/absolute/path/staging"

Existing skill destinations are rejected; no global paths are written.

1. When the user requests workspace installation, verify the current project:

   python3 scripts/port_skill.py --source "/absolute/path/my-skill" --workspace --dry-run
   python3 scripts/port_skill.py --source "/absolute/path/my-skill" --workspace

This writes only .agents/skills/<id> under the current working directory.

Confirm that discovery path is supported by the intended host first.

1. Review context-file references, tool argument shapes, client configuration,

dependencies, licensing and multi-agent ordering manually. Validate the adapted

skill and test actual client invocation before claiming compatibility.

Examples

For a multi-skill repository, use its local root as the source:

python3 scripts/port_skill.py --source "/absolute/path/reviewed-repository" --dry-run
python3 scripts/port_skill.py --source "/absolute/path/reviewed-repository" --dest "/absolute/path/fresh-output"

A source-only invocation defaults to preview. Remote URLs are rejected; obtain

and inspect a pinned local checkout first. Run bundled tests from this directory:

PYTHONDONTWRITEBYTECODE=1 python3 -m unittest scripts/test_port_skill.py -v

Limitations

• Conservative text adaptation only: no AST conversion, semantic optimization,

automatic parallelization, artifact generation, or compatibility certification.

• Context references and support scripts retain their original bytes and may need

manual adaptation. Binary support files are copied intact, never interpreted.

• Symbolic links and non-regular files are rejected. Input is bounded to 1,000 files

and 20 MiB. This is not a sandbox against concurrent hostile filesystem changes;

use only stable directories you control.

• Disk or filesystem failure may leave a partial new destination. Inspect it and

choose a fresh output for retries. There is no overwrite or rollback mode.

• No downloads, credentials, network calls, global installation, plugin manifest

generation/registration or external MCP setup. Verify tool availability in the

actual host, whose version and capabilities may differ.

Source and license

Adapted from Pranav-Nexus/antigravity-skill-porter. The upstream MIT notice is

preserved in [LICENSE](LICENSE).