What are Skills?
In the ToolVerse ecosystem, a Skill is defined as a discrete functional capability that an agent can invoke dynamically. While MCP (Model Context Protocol) servers provide the infrastructure for connectivity, Skills represent the logic and business rules that govern those connections.
Skills are typically packaged as .skill files or SKILL.md manifest structures. This allows agents to parse the capabilities, requirements, and execution parameters without needing to ingest entire documentation libraries.
Pro Tip: Lazy Loading
For complex agent workflows, use the lazy_init: true flag in your manifest. This prevents the agent from initializing heavy dependencies until the skill is actually called for the first time in a session.
Key Advantages
Token Efficiency
Skills use compressed manifest formats that minimize context window consumption. This allows for more tools to be loaded simultaneously without hitting model limits.
Domain Expertise
Define specific constraints and prompt-level guardrails within the skill logic to ensure the agent operates within narrow, expert parameters.
SKILL.md Structure
The standard manifest format for defining a skill. This markdown file acts as both documentation and machine-readable config.
# Skill: Database Analyzer v1.0
## Configuration
id: db-analyzer-core
runtime: python-3.11
permissions: [read-only, schema-access]
## Functions
### analyze_query_performance
Takes a SQL query and returns execution plan optimization suggestions.
- **Param**: query (string, required) - The SQL string to analyze.
- **Param**: engine (enum, optional) - [postgres, mysql, snowflake].
## Logic
if not permission_allowed(user, 'schema-access'):
return error('Missing permission')
plan = explain_analyze(query)
return optimize(plan)