Repository navigation
v0.5.0
What's New
Skill Evals
- Added eval framework for testing skill activation across delivery modes (MCP, native, CLI, mcp+native)
- Includes xlsx skill from Anthropic skills repo and synthetic test skills
- Documents observed behaviors: noodling, context duplication, activation differences
Activation Triggers
- Added explicit activation language to MCP skill tool description
- Matches native Skill tool behavior: "IMMEDIATELY", "NEVER just announce", "BLOCKING REQUIREMENT"
- Improves skill activation rates for MCP-delivered skills
Key Finding
Native mode activated skills more readily than MCP due to tool description differences. Adding activation triggers to MCP tool descriptions brings behavior in line with native mode.
Source for native Skill tool description: https://github.com/Piebald-AI/claude-code-system-prompts/blob/main/system-prompts/tool-description-skill.md