Skrapi

Skrapi turns an unfamiliar repository into a concise, reusable map. It studies what the code actually does, then writes a SKRAPI/ documentation set that another developer or agent can use without repeating the discovery work.

The analysis is stack-aware but not stack-assumptive. It asks for Spanish, English, or Chinese output, defaults to English if no choice is given, identifies the kind of project in front of it, and follows real entry points, imports, configuration, and data flow before describing the architecture.

How it studies a project

Skrapi reads the manifests, package-manager evidence, configuration, source structure, entry points, and relevant AI-tooling metadata. It classifies the repository as a web application, mobile application, browser extension, library or CLI, or monorepo, then applies the matching review guidance.

Dependency descriptions must be supported by real code usage. A package name alone is not proof that a feature or architectural pattern is active. Lockfiles are useful evidence too: they help identify the package manager and make generated installation commands accurate.

The documentation set

Every run creates three core files:

• ARCHITECTURE.md maps layers, ownership, entry points, and important data flows.
• PACKAGES.md explains each declared dependency’s observed role and flags usage that could not be confirmed.
• PROMPT.md turns the discovered structure into implementation guidance that can be reused safely.

STATE-MANAGEMENT.md and STYLING.md are added only when the repository provides enough evidence to justify them. A separate SKILLS.md is created when at least one discovered skill besides Skrapi exists, and it lists every discovered skill. PROMPT.md retains its Skills section even when that separate file is unnecessary.

Monorepos and larger codebases

For a monorepo, Skrapi creates per-project documentation beneath the fixed SKRAPI/ directory so distinct applications do not get flattened into one vague description. Very large repositories are sampled around meaningful boundaries rather than being described as exhaustively read when they were not.

Non-JavaScript codebases are supported as an explicit best-effort analysis. The output records uncertainty and unread areas instead of hiding them.

Evidence over inference

The lasting value of the result is traceability: important claims point back to files, imports, configuration, or observed conventions. Generated prompts preserve the detected package manager and, when skills-lock.json proves a custom source, include the correct skill installation command.

Skrapi documents the repository that exists—not the architecture its dependencies merely suggest.