MCP Skill

Positioning: mcp_skill is an internal component of the MakeMind knowledge stack exposed through the mcp_knowledge facade. Application code should import package:mcp_knowledge/mcp_knowledge.dart — the symbols declared here are re-exported from there. Direct package:mcp_skill/ imports remain valid for advanced or integration scenarios but are discouraged in product code.

A typed skill definition and execution runtime for AI capabilities — JSON-Schema-validated inputs/outputs, multi-step procedures with branching and circuit breakers, embedded expression language, and Contract Layer integration with the rest of the MCP ecosystem.

Components

• Definition — Skill, Parameter, Step, Procedure models.
• Bundle — SkillBundle packaging with loader, validator, capability composition, capability verifier, permission enforcer.
• Registry — SkillRegistry and an in-memory implementation.
• Runtime — SkillRuntime with context builder, branch merger, circuit breaker, error aggregator, claim extractor, procedure executor.
• Multi-skill — chained execution and dependency resolution across skills.
• Expression language — lexer / parser / AST / evaluator / function registry for skill conditionals and templated values.
• Evaluation — rubric-based scoring of skill executions.
• FactGraph integration — bridge into mcp_fact_graph for claim recording.
• Contract Layer adapters — SkillRuntimePortAdapter, SkillRegistryPortAdapter implementing the mcp_bundle standard ports.

Quick Start

import 'package:mcp_skill/mcp_skill.dart';

final registry = MemorySkillRegistry();
await registry.register(mySkill);

final runtime = SkillRuntime(registry: registry, llm: llmPort);
final result = await runtime.run(skillId: 'summarize', input: {'text': '...'});

Support

• Issue Tracker
• Discussions

License

MIT — see LICENSE.