Crate mcp_execution_skill 

Skill generation for MCP progressive loading.

This crate provides functionality to generate Claude Code skill files (SKILL.md) from generated progressive loading TypeScript files.

Architecture

The skill generation flow:

1. parser - Reads the _meta.json sidecar emitted by codegen
2. context - Builds structured context from parsed tools
3. template - Renders Handlebars template with context

Examples

use mcp_execution_skill::{scan_tools_directory, build_skill_context, ScanError};
use std::path::Path;

let result = scan_tools_directory(Path::new("~/.claude/servers/github")).await?;
let context = build_skill_context("github", &result.tools, None);

Re-exports

pub use types::GenerateSkillParams;

pub use types::GenerateSkillResult;

pub use types::SaveSkillParams;

pub use types::SaveSkillResult;

pub use types::SkillCategory;

pub use types::SkillMetadata;

pub use types::SkillTool;

pub use types::ToolExample;

pub use types::validate_server_id;

Modules

types

Type definitions for skill generation.

Structs

ParsedParameter

A parsed parameter from a tool’s metadata.

ParsedToolFile

Parsed metadata from a server’s generated tool set.

ScanResult

Result of scan_tools_directory: the parsed tools plus any non-fatal drift warnings.

Enums

ScanError

Errors that can occur while scanning a server directory for its _meta.json sidecar.

TemplateError

Errors that can occur during template rendering.

Constants

MAX_FILE_SIZE

Maximum sidecar file size to read in bytes (1MB).

MAX_TOOL_FILES

Maximum number of tools accepted from a single sidecar (denial-of-service protection).

Functions

build_skill_context

Build skill generation context from parsed tools.

extract_skill_metadata

Extract skill metadata from SKILL.md content.

render_generation_prompt

Render the skill generation prompt.

render_skill_md

Render SKILL.md content directly from skill context.

scan_tools_directory

Scan a server directory and read its _meta.json sidecar.