Port AI skills to Google Antigravity tool syntax

Adapts a locally obtained Claude Code, Cursor, Codex, or generic agent skill bundle for Google Antigravity by rewriting only exact, backtick-quoted tool identifiers (such as View, Edit, and Bash) while leaving frontmatter and support-file bytes intact. It runs as a Python 3.9+, standard-library-only script (scripts/port_skill.py) against a reviewed local skill directory - either a single directory containing SKILL.md or a repository with multiple skills/<lowercase-hyphenated-id>/SKILL.md directories - and previews the tool-name substitutions with --dry-run before writing anything.

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

It refuses to write to an existing skill destination, writes no global paths unless explicitly asked to install into the current workspace (--workspace, which writes only .agents/skills/<id> under the current working directory), and explicitly does not prove client compatibility - it's a conservative text-adaptation utility, not a semantic converter or a certification tool.

Use it when porting a reviewed, locally obtained skill bundle (not a remote URL - those are rejected; a pinned local checkout must be obtained and inspected first) from Claude Code, Cursor, Codex, or a generic agent format into Google Antigravity's expected tool-name conventions.

Do not treat its output as proven-compatible: it performs no AST conversion, semantic optimization, automatic parallelization, artifact generation, or compatibility certification, and context-file references, tool argument shapes, client configuration, dependencies, licensing, and multi-agent ordering must still be reviewed manually and tested against the actual target client before relying on the port. It is also not a sandbox against concurrent hostile filesystem changes - it should only be run against stable directories the user controls, never a tree another process or user is modifying.

Input is a source path to a reviewed local skill directory or multi-skill repository, bounded to 1,000 files and 20 MiB; symbolic links and non-regular files are rejected. Output is either a dry-run diff preview (tool identifiers changed, everything else unchanged) or a copied bundle at a fresh, non-existing destination path - or, for workspace installation, files written only under .agents/skills/<id> in the current working directory. Binary support files are copied intact and never interpreted. A disk or filesystem failure mid-run can leave a partial destination with no overwrite or rollback mode; the fix is to inspect it and retry into a fresh output path.

A standalone Python standard-library script with no downloads, credentials, network calls, global installation, plugin-manifest generation/registration, or external MCP setup - tool availability still has to be verified in the actual host client, whose version and capabilities may differ. Adapted from Pranav-Nexus/antigravity-skill-porter under its preserved upstream MIT license notice; bundled tests run via python3 -m unittest scripts/test_port_skill.py -v.

Anyone adapting an existing agent skill bundle for Google Antigravity who wants a conservative, inspectable text substitution rather than an automated "compatible" claim - and who will still manually verify tool semantics and run a real client invocation before trusting the port.