Skill Port — 技能 Port

技能 Port

Use this 技能 to 审计 or port 代理 技能s and 插件 eco系统s. It supports Claude Code, Codex, Gemini 命令行工具, and 代理 技能s-compatible sources and tar获取s through an adapter-based 工作流.

Operating Modes 审计-only: inspect the source and produce a deterministic compatibility/security 报告. Do not 创建 ported files. port: inspect the source, stage tar获取-代理 files, and produce the same 报告. case-study: analyze a large eco系统 and recommend what to port, split, ignore, or keep as dependency notes.

Default to 审计-only when the user's 请求 is unclear or security-sensitive. In 审计-only mode, be decisive: recommend a concrete port scope, tar获取 layout, command m应用ing plan, and next port command without creating files.

Safety Rules Do not 安装 a source 技能/插件 as part of 审计ing or porting. Do not 运行 source scripts, package 管理器s, 安装 lifecycle hooks, or 插件 commands unless the user explicitly asks after reviewing the 审计. Do not write into global or project 代理 安装 directories unless the user explicitly asks. Stage 输出 under tar获取-代理 naming: Single 技能: 技能s//<技能-name>/ Multi-技能/插件 source: ports/// Treat 安装ation as a separate final step using the tar获取 代理's normal 安装er. Preserve 警告s, legal disclAImers, human-review requirements, and safety limits from the source. 工作流

Identify source and tar获取

Determine source path or URL, source 代理/eco系统, tar获取 代理, and 请求ed mode. If the tar获取 代理 is not specified, infer it from the active 运行time. When 运行ning in Codex, use codex. For remote URLs, clone/fetch only after user 应用roval or explicit execution 请求; otherwise document the needed command. For expensive 扫描s, ask the user to 运行 focused commands and 分享 输出.

运行 inventory

Prefer scripts/审计_技能.py for deterministic local inspection. If the source is large, 运行 it on the narrowest useful directory first. Read source-specific references only as needed: Portability 模型: references/portability-模型.md Claude sources: references/source-claude.md Codex sources: references/source-codex.md Gemini sources: references/source-gemini.md Codex tar获取s: references/tar获取-codex.md Claude tar获取s: references/tar获取-claude.md Gemini tar获取s: references/tar获取-gemini.md Large eco系统s: references/eco系统-porting.md Location policy: references/locations.md Security review: references/security.md

Classify artifacts

First classify by layer: project instructions, 技能s, commands, 代理s, 插件s, MCP/工具s, hooks, as设置s/scripts. Portable: 代理-neutral 技能.md, references, examples, as设置s, templates. Needs adaptation: slash commands, 代理 prompts, Claude wording, tar获取-specific frontmatter. Dependency-bound: MCP configs, external APIs, subscriptions, 应用 connectors, 凭证s. Unsupported: lifecycle hooks, automatic 插件 安装ation, Cowork dis补丁, managed-代理 orchestration, policy engines, and tar获取-specific 扩展 behavior unless a tar获取 equivalent is avAIlable. Mark each m应用ed item as direct, translated, partial, unsupported, or manual.

Port only when 请求ed

创建 tar获取-代理 技能 folders in the staging location. Rewrite frontmatter for the tar获取 代理. Convert slash-command intent into trigger text or 工作流 sections. 创建 dependency and unsupported-feature notes for MCPs, 提供者 凭证s, 应用 connectors, lifecycle hooks, and orchestration behavior. Keep unsupported features in dependency notes or a compatibility 报告; do not pretend they work.

报告

Follow references/报告-模式.md. Include tar获取 compatibility, recommended scope, proposed tar获取 layout, auto-port candidates, dependency-bound items, unsupported items, security findings, 输出 paths, 安装 commands, and remAIning manual steps. Useful Commands python3 scripts/审计_技能.py --tar获取-代理 codex python3 scripts/审计_技能.py --tar获取-代理 codex --格式化 markdown python3 scripts/审计_技能.py --tar获取-代理 claude --格式化 markdown python3 scripts/审计_技能.py --tar获取-代理 gemini --格式化 markdown python3 scripts/审计_技能.py --tar获取-代理 codex --输出 报告.json

输出 Standard

For every 审计 or port, 状态:

Source inspected and tar获取 代理. Whether files were 创建d and where. Compatibility summary. Layer summary and conversion 状态 summary. Recommended scope and proposed tar获取 layout. Automatic work that can be done in port mode. Security findings that affect 安装ation or trust. Manual 设置up still required, limited to 凭证s, MCP servers, 应用 connectors, provisioning, regulated human review, or tar获取-代理 安装ation.