Source of Truth

Legacy repositories and undocumented codebases create friction for every new developer or agent assigned to the project. This skill audits your existing files to build and maintain a high-fidelity knowledge base, ensuring every module, interface, and architectural decision is mapped and searchable.

What it does

• Map architecture by scanning project modules, data structures, and interfaces to define how components interact.
• Document conventions by identifying coding standards, delivery patterns, and testing protocols used in the source.
• Create searchable views by generating an offline-ready HTML interface with navigation for quick reference.
• Maintain state by updating the knowledge base when project logic changes or new architectural decisions are made.
• Onboard agents by providing a structured context layer that allows other LLMs to understand the repo quickly.

How it works

1. Initial Audit where the skill scans the directory to determine project type and scope the documentation requirements.
2. Drafting Phase using specialized templates for writers and reviewers to ensure technical depth and accuracy.
3. User Verification where the agent pauses to ask scoped questions and gather approvals before finalizing the documents.
4. View Generation where a Python script builds the offline HTML browser view including CSS and JS assets.

Frameworks & tools

Works with any project structure (software or general). The HTML view generator requires Python 3.10 or newer. It reads source files as data without executing them, making it compatible with any language or framework.

Why this beats prompting it yourself

Manual prompts often miss edge cases in large repos or hallucinate architectural patterns. This skill uses a multi-step verification process with specific writer and reviewer templates to ensure the output matches the actual source code.

Use cases

• Legacy handovers where a developer needs to document an inherited codebase they didn't write.
• Agent onboarding to create the "source of truth" context needed for other AI tools to operate safely.
• Decision logging to record architectural pivots and open questions in a centralized, versioned format.
• Project audits to identify undocumented modules or inconsistencies in project conventions.

Known limitations

Does not change project code or install dependencies. Large repos may consume significant AI tokens. The HTML generator requires a local Python environment.