开源AI工作流

How the Installer Works

The skills-setup command guides you through two interactive screens:

1. Select AI tools — choose which tools to install skills to
2. Select skills — choose which skills to install

The skill selection screen shows two sections: Builtin Skills (bundled with the package) and Local Skills (discovered in the current directory). See Installing Local Skills for details.

Select skills to install:
Up/Down move | Space toggle | a all/none | Enter confirm | q quit

  [x]  Select All / Deselect All
  ------------------------------------
  Builtin Skills (16)
  [x]  blog-writer
  [x]  code-review-staged
  [x]  figma
  ...
  Local Skills (2)  [my-skills/]
  [x]  my-custom-skill
  [x]  jira  (overrides builtin)

  18/18 selected

Tools that are not installed on your machine are shown dimmed with [-] markers and cannot be selected.

After installation, skills-check runs automatically for builtin skills that need external dependencies (CLI tools or API tokens). Local skills are excluded from this check — managing their dependencies is the user's responsibility.

Installing Local Skills

skills-setup discovers local skills automatically when you run it from a directory that contains skill subdirectories (any non-hidden subdirectory with a SKILL.md file). This makes it easy to install third-party or custom skill collections you've downloaded:

```bash

Run skills-setup from inside that directory

skills-setup ```

The installer scans the current directory's immediate subdirectories and shows any discovered skills in a separate Local Skills section below the builtin ones.

Conflict handling: If a local skill has the same name as a builtin skill, it is highlighted in a distinct color with an (overrides builtin) note. When both are installed, builtin skills are installed first and local skills are installed last — so the local version takes precedence.

Dependency checks: skills-check only runs for builtin skills after installation. For local skills, dependency management (installing required CLI tools, API keys, etc.) is left to the user.

Cleanup installed skills

skills-cleanup

The cleanup command guides you through two screens:

1. Select AI tools — choose which tool directories to scan (defaults to all detected)
2. Select skills to remove — a tree view showing each tool and its installed skills (defaults to none selected)

Select skills to remove:
Up/Down move | Space toggle | a all/none | Enter confirm | q quit

  [ ]  Select All / Deselect All
  ------------------------------------
  Copilot CLI  ~/.copilot/skills/  (0/3)
      [ ]  code-review-staged
      [ ]  figma
      [ ]  skill-creator
  Cursor  ~/.cursor/skills/  (0/2)
      [ ]  gh-operations
      [ ]  jira

  0/5 selected for removal

Project-specific setup

To install skills at the project level instead, manually copy them:

```bash cp -r mythril_agent_skills/skills/skill-name ./your-project/.github/skills/

Quick Start

Use as-is skills:

<details> <summary>Option A: Install via pip</summary>

If you just want to install and use existing skills, start here. Install it from PyPI — no need to clone the repository:

pip install mythril-agent-skills

To upgrade to the latest version:

pip install -U mythril-agent-skills

This gives you four commands:

Install skills:

skills-setup              # Interactive: select tools, then skills
skills-setup .cursor      # Direct target: skip tool selection

Remove skills:

skills-cleanup

Check dependencies:

skills-check                    # Interactive: select skills to check
skills-check gh-operations jira figma  # Check specific skills

Clean up cached temp files:

skills-clean-cache          # Interactive: list cache, confirm before deleting
skills-clean-cache --force  # Delete without confirmation
skills-clean-cache --repos  # Interactive: select repos to delete

The checker will: - Launch an interactive UI to select skills (when run without arguments) - Detect missing CLI tools (e.g. gh) and offer to install them automatically - Prompt for missing API keys/tokens and save them to your shell config file - Verify authentication status (e.g. gh auth status)

</details>

<details> <summary>Option B: Claude Code plugin marketplace</summary>

If you use Claude Code, you can install skills as a plugin marketplace — no pip required. See also: discover and install plugins.

Add the marketplace:

/plugin marketplace add jie-meng/mythril-agent-skills

Install all skills at once:

/plugin install all-skills@mythril-agent-skills

Or install individual skills as needed:

/plugin install figma@mythril-agent-skills
/plugin install jira@mythril-agent-skills
/plugin install github-code-review-pr@mythril-agent-skills

You can use /plugin discover to interactively find and install plugins:

<p align="center"> <img src="docs/assets/cc-plugin-usage.webp" width="600" alt="Add marketplace and install plugin"> </p>

<details> <summary>All available plugins</summary>

</details>

To update later:

/plugin marketplace update mythril-agent-skills

To uninstall a plugin:

/plugin uninstall figma@mythril-agent-skills

To remove the marketplace entirely (also uninstalls all its plugins):

/plugin marketplace remove mythril-agent-skills

Note: This option only installs skills into Claude Code — the CLI commands (skills-setup, skills-cleanup, skills-clean-cache) are not included. For full toolkit support (multi-tool installation, dependency checking, cache management), use Option A.

</details>

Or customize your own skills:

<details> <summary>Option C: GitHub fork or independent clone</summary>

If you want to customize skills and keep your own repository, you have two equivalent paths:

• GitHub fork (stay linked to upstream on github.com)
• Independent clone (fully detach from upstream, any platform)

Pick one setup below, then follow the shared usage steps after it.

Setup A — GitHub fork (linked to upstream):

```bash