| name | skill-agentic-initializer |
| description | Professional bootstrapper for Vercel-style Agentic Repositories using compressed AGENTS.md manifests. |
Agentic Initializer (Vercel-Standard)
This skill automates the creation of a high-performance AGENTS.md manifest. Inspired by Vercel's research, it prioritizes "retrieval-led reasoning" over model pre-training by embedding a compressed documentation index directly into the repository root.
Logic Overview
- Deterministic Scanning: Scans a designated documentation directory (default:
{PROJECT_DIR}/docs/) using {SKILL_DIR}/scripts/scanner.py.
- Compressed Indexing: Builds a pipe-delimited 8KB "World Map" that maps paths to documentation files.
- Passive Context Delivery: Generates an
AGENTS.md that acts as the primary cognitive guide for LLM agents, ensuring they always have access to version-matched documentation.
- Retrieval-Led Reasoning: Injects explicit instructions to prefer retrieval from local docs over potentially outdated training data.
Usage
python3 {SKILL_DIR}/scripts/logic.py <target_path> <project_name> [docs_dir]
References
- See
{SKILL_DIR}/references/vercel-agents-md.md for the original research findings by Jude Gao (Next.js/Vercel).
