What problem does it solve? Repositories without project-local agent guidance cause AI coding agents to make wrong assumptions about package managers, build commands, and test runners, wasting context and introducing errors. This Skill inspects a repo, identifies guidance gaps, and creates a minimal scaffold that references global skills by name instead of copying them. ## Core Features & Use Cases - Guidance Inventory and Gap Analysis: Detects existing AGENTS.md, CLAUDE.md, GEMINI.md, .codex/project.toml, lockfiles, and framework markers before proposing any changes. - Minimal Scaffold Generation: Creates a concise AGENTS.md with package manager, key commands, safety constraints, and a machine-readable .codex/project.toml manifest with validated TOML syntax. - Context Engineering Reference: Includes absorbed guidance on session context setup, rules file hierarchy, and recovery when agent output quality degrades. - Use Case: When starting work on a medium-sized Next.js repo with no agent instructions, use this Skill to detect the pnpm lockfile, confirm scripts from package.json, and generate an AGENTS.md plus project.toml that point every future agent at the right commands and a curated set of at most seven skills and three agents. ## Quick Start Ask the agent to inspect this repository for existing agent guidance and propose a minimal AGENTS.md and .codex/project.toml scaffold before creating any files.
