Descriptions:
IndyDevDan presents a practical solution for engineers managing AI agents, skills, and prompts across ten or more codebases: a ‘library meta-skill’ that acts as a centralized index for distributing private skills, agents, and prompts without duplicating files across repos. The system uses a single YAML library file — analogous to a package.json — that stores references to private GitHub repositories or local file paths rather than copying content directly.
The meta-skill supports commands like `/library add` and `/library use` to pull named skills into any codebase on demand. Dan demonstrates the full workflow using his custom PI coding agent configured with Claude Sonnet, cloning and installing skills from a private repository into a fresh temp directory. The approach handles deduplication, version sync, and cross-team distribution cleanly, with a single source of truth that both developers and agents can reference.
The video also introduces a hierarchy of agentic primitives: raw skills (discrete capabilities), agents (parallelism and scale), and single-file prompts (lightweight orchestration) — arguing that over-engineering everything into skills is a common mistake when simpler prompt files would suffice. The library system is explicitly designed for private professional work rather than public plugin registries, and the presenter emphasizes reviewing any externally sourced skills before installation. Compatible with Claude Code and other agentic frameworks that support skill and command file conventions.
📺 Source: IndyDevDan · Published March 16, 2026
🏷️ Format: Tutorial Demo
