Why this matters

SkillTool is the bridge between command-style skills and the normal tool runtime. It is where “use a skill” becomes an actual execution path.

Big picture first

The important pieces are:

1. SkillTool/prompt.ts teaches the model when invoking a skill is mandatory.
2. getAllCommands(...) collects both local skills and MCP-loaded prompt skills.
3. executeForkedSkill(...) turns one skill invocation into a forked agent run.

tools/SkillTool/SkillTool.ts also anchors the exported tool name as SKILL_TOOL_NAME.

Important:
- Available skills are listed in system-reminder messages in the conversation
- When a skill matches the user's request, this is a BLOCKING REQUIREMENT: invoke the relevant Skill tool BEFORE generating any other response about the task

The command bridge

async function getAllCommands(context: ToolUseContext): Promise<Command[]> {
  const mcpSkills = context
    .getAppState()
    .mcp.commands.filter(
      cmd => cmd.type === 'prompt' && cmd.loadedFrom === 'mcp',
    )
  if (mcpSkills.length === 0) return getCommands(getProjectRoot())
  const localCommands = await getCommands(getProjectRoot())
  return uniqBy([...localCommands, ...mcpSkills], 'name')
}

That helper is why the chapter is called a bridge. SkillTool is not only about bundled skills on disk. It also merges in MCP-loaded prompt skills.

A skill often becomes a forked agent run

async function executeForkedSkill(
  command: Command & { type: 'prompt' },
  commandName: string,
  args: string | undefined,
  context: ToolUseContext,
  canUseTool: CanUseToolFn,
  parentMessage: AssistantMessage,
  onProgress?: ToolCallProgress<Progress>,
): Promise<ToolResult<Output>> {

    for await (const message of runAgent({
      agentDefinition,
      promptMessages,
      toolUseContext: {
        ...context,
        getAppState: modifiedGetAppState,
      },
      canUseTool,
      isAsync: false,
      querySource: 'agent:custom',

So the mental model is: SkillTool receives a skill request, loads the command context, and often turns it into a forked runAgent(...) execution.