The original title is about upgrading Claude Code configuration from skills to a three-layer system with skills, agents, and commands.

Short summary

The author upgraded their Claude Code configuration from a single 'skills' layer to a three-layer system: skills (how), agents (who with what tools), and commands (run now). Agents are thin personas with tool allowlists that prevent reviewers from editing code. The repo ships 29 skills, 20 agents, and 10 commands, installable via curl or plugin marketplace, with automatic migration from the old structure.

• •Three-layer architecture: skills, agents, commands — each answering a distinct question
• •Agents enforce tool boundaries (e.g., read-only for reviewers) instead of relying on prompts
• •One-line installer migrates from old agent-skills repo; also bundled via Kickoff 1.36.0

Generated with AI, which can make mistakes.