skill-port

skill-port is an Agent Skills portability auditor for developers who want to reuse AI agent skills, plugins, slash commands, subagents, MCP configs, and workflow packs across Claude Code, Codex, Gemini CLI, Google Antigravity, and other Agent Skills-compatible runtimes.

Use it when a Claude Code plugin does not work in Codex, a Codex skill needs to be adapted for Claude Code, a Gemini CLI extension or Antigravity skill/rule bundle needs a portability review, or a skill repository needs a deterministic compatibility and security report before installation.

The installable skill lives at skills/skill-port.

Problems It Solves

• "Can I use this Claude Code skill or plugin in Codex?"
• "What needs to change before this Codex skill works in Claude Code?"
• "How portable is this Gemini CLI extension?"
• "Can this Antigravity skill, rule, workflow, or MCP setup be reused elsewhere?"
• "Which parts of this agent plugin are standard Agent Skills, and which parts are vendor-specific?"
• "Does this skill contain risky scripts, hooks, network calls, secret-like values, or destructive commands?"
• "Where should ported files be staged without mutating my global agent install?"

Install with skills.sh

npx skills add yand/skill-port --skill skill-port

The skills.sh CLI is the main discovery path for public skill directories and leaderboards.

What It Does

• Audits Agent Skills, Claude Code skills/plugins, Codex skills/plugins, Gemini CLI skills/extensions, Antigravity skills/rules/workflows, command bundles, agent bundles, and MCP-backed plugin repos.
• Reports portability, target-agent compatibility, source/target file mapping, security findings, dependencies, and manual setup.
• Stages ported output under target-agent naming instead of mutating installed skill directories.
• Helps adapt source-specific artifacts such as slash commands, Codex custom agents, Gemini extensions, Antigravity rules/workflows, Cowork plugins, managed-agent cookbooks, hooks, policies, and MCP connector setup.
• Maps non-skill plugin layers into target-native plans where possible: command maps/router skills, workflow skills for named agents, MCP setup snippets with credential placeholders, and no-op versus active hook treatment.
• For Claude/Cowork plugin sources targeting Codex, prefers Codex plugin or marketplace layouts over noisy flat global skill installs.

Common Searches

People looking for this skill may describe the problem as:

• Claude Code plugin to Codex
• Claude skill to Codex skill
• Codex skill to Claude Code
• Gemini CLI extension portability
• Antigravity skill portability
• Antigravity MCP config audit
• Agent Skills converter
• AI agent skill adapter
• MCP plugin portability audit
• slash command migration for AI coding agents
• subagent portability between Claude Code, Codex, Gemini CLI, and Antigravity
• skills.sh compatible skill audit

Install

From this repository:

npx skills add . --skill skill-port

From GitHub:

npx skills add yand/skill-port --skill skill-port

The CLI detects the current agent environment and installs into that agent's supported skill location. Review the CLI output before using the skill.

Usage

Ask your agent to use skill-port in one of three modes:

• audit-only: inspect a source and produce a compatibility/security report.
• port: stage target-agent files plus the report.
• case-study: analyze a large plugin ecosystem and recommend what to port, split, ignore, or document as dependencies.

Example prompts:

Use skill-port to audit this agent skill before I install it: ./some-skill

Use skill-port to audit this Claude Code plugin for Codex compatibility: ./vendor/example-plugin

Use skill-port to port ./my-claude-skill to Codex and stage the output under target-agent naming.

Use skill-port to audit this Codex plugin for Claude Code compatibility: ./vendor/codex-plugin

Use skill-port to audit this Gemini extension and recommend a Codex target layout: ./vendor/gemini-extension

Use skill-port to audit this Antigravity skill/rule bundle for Claude Code compatibility: ./vendor/antigravity-bundle

The deterministic helper can also be run directly:

python3 skills/skill-port/scripts/audit_skill.py ./path/to/source --target-agent codex --format markdown
python3 skills/skill-port/scripts/audit_skill.py ./path/to/source --target-agent claude --format markdown
python3 skills/skill-port/scripts/audit_skill.py ./path/to/source --target-agent gemini --format markdown
python3 skills/skill-port/scripts/audit_skill.py ./path/to/source --target-agent antigravity --format markdown
python3 skills/skill-port/scripts/stage_port.py ./path/to/source --target-agent codex --output-root . --clean

For Codex plugin or marketplace ports, staging does not install or enable plugins. Review the staged files, then expose a local marketplace with:

codex plugin marketplace add ./ports/<source-name>/codex-marketplace

Restart Codex, then install and enable the desired plugins from Codex's plugin directory or another official Codex UI/command when available. Do not hand-edit ~/.codex/config.toml plugin enablement entries as a substitute for installation. Bundled plugin skills appear only after the plugin is installed and enabled by Codex.

For Codex Desktop local testing, install a staged marketplace with Codex's private local bundle pattern:

python3 skills/skill-port/scripts/install_codex_local_bundle.py ./ports/<source-name>/codex-marketplace

This registers the marketplace, copies plugin directories into ~/.codex/plugins/cache/<marketplace>/<plugin>/<version>, enables the plugins in ~/.codex/config.toml, and requires a Codex restart.

If --target-agent is omitted, the helper first honors explicit runtime self-identification such as AGENT_RUNTIME, AGENT_NAME, CURRENT_AGENT, or HOST_AGENT, then falls back to high-confidence environment variables and conservative heuristics. Known aliases normalize to codex, claude, gemini, or antigravity.

Safety Model

skill-port is read-only by default:

• It does not install source skills/plugins during audit.
• It does not run source scripts, package managers, lifecycle hooks, or plugin commands.
• It does not make network requests from the bundled auditor.
• It does not write into global or project agent install directories unless explicitly requested.
• The deterministic auditor writes only when --output <path> is explicitly provided.
• It flags secret-like values, destructive commands, shell execution, network calls, lifecycle hooks, credential access, MCP configs, hidden files, binaries, and large files.

See SECURITY.md for the full safety boundary and reporting guidance.

Repository Layout

skills/skill-port/
  SKILL.md
  agents/openai.yaml
  references/               # source/target adapters, security, reporting, location policy
  scripts/audit_skill.py

This uses a standard skills/<name>/ layout so skill indexes and installers can discover the skill without relying on recursive fallback. Adapter references currently cover Claude Code, Codex, Gemini CLI, and Antigravity as both sources and targets.

Index Readiness

The skill follows the Agent Skills shape: a directory containing SKILL.md with name and description frontmatter plus optional scripts/ and references/.

To help skills.sh discover and rank the skill, install it from GitHub with telemetry-enabled CLI defaults:

npx skills add yand/skill-port --skill skill-port

Before publishing, run:

python3 skills/skill-port/scripts/audit_skill.py skills/skill-port --target-agent codex --format markdown
npx skills add . --list

License

Licensed under the Apache License 2.0. Attribution notices are provided in NOTICE.

Sources

• Agent Skills specification: https://agentskills.io/specification
• skills.sh documentation: https://www.skills.sh/docs
• skills CLI documentation: https://www.skills.sh/docs/cli