context-creation

To initialize project state for an existing repository, the context-creation skill performs six parallel scout passes to inspect stack, infrastructure, architecture, security, testing, and data model signals. It then synthesizes these signals into .codearbiter documentation files and locks initialization with a marker.

Brownfield initialization detects when source code already exists but project context is not initialized, triggering extraction instead of scaffolding from scratch. This ensures the system understands existing code and can enforce gates safely without guessing prior state.

Scouted analysis uses six parallel scout passes to inspect stack, infrastructure, architecture, security, testing, and data model signals. This approach extracts project context without loading raw source code into the main context, keeping the analysis lightweight and focused.

Yes, context extraction is designed for mature repositories that need documentation, gating, and orchestration readiness. It flags unknowns, asks targeted questions, and records deferred items so the resulting project state is explicit, auditable, and prepared for normal feature work.

Automated codebase context creation may encounter gaps where signals are unknown or unclear. In these cases, it flags unknowns and records deferred items rather than guessing, meaning some manual gap resolution may be required to achieve a fully auditable project state.