skill-mcp defines 5 read-only tools with consistent naming (skill_* prefix) and basic schemas. All tools have descriptions (10-60 chars) and input parameters with type definitions. However, descriptions are minimal and lack LLM-optimization guidance (e.g., when to call skill_list vs skill_view, what each returns). Output schemas are not documented. No error handling guidance or recovery paths. Parameter descriptions are sparse, e.g., 'slug' lacks context on format or how to obtain it. No pagination support despite file_tree potentially returning large structures.

• Document output schemas for all tools. E.g., skill_list returns [{slug, name, tags, description}]; skill_view returns {slug, name, version, tags, description, files}. Include field types and whether fields are always present.
• Expand parameter descriptions to 50-100 chars with format hints and origin guidance. E.g., 'slug: Unique skill identifier (lowercase alphanumeric + hyphens). Obtain from skill_list or skill_view.'
• Add error recovery guidance in tool descriptions. E.g., skill_read_files: 'If a path is invalid, call skill_file_tree first to list available paths.'
• Add pagination support to skill_file_tree and skill_read_files. Include limit (default 50, max 500) and offset/cursor parameters. Return total_count and next_cursor.
• Clarify tool selection in descriptions. E.g., skill_list: 'Use to discover available skills and filter by tags.' skill_view: 'Use to get full metadata for a specific skill.' skill_read_entry: 'Use to read the SKILL.md documentation file.' skill_read_files: 'Use to read arbitrary files from a skill package.'

Inferred effective spec: <=2025-11-25.

• Relies on Stateful initialize / Mcp-Session-Id (removed; protocol is stateless) - make each request self-contained