What problem does it solve? When you need to write a PRD suite for an existing (brownfield) repository, manually reading the entire codebase is slow and error-prone. This Skill performs a read-only reconstruction scan that harvests grounded evidence about capabilities, artifacts, and boundaries so a downstream build step can compile PRDs without rereading the repository. ## Core Features & Use Cases - Reconstruction-Grade Scanning: Produces a complete scan package under .specify/prd-runs/<run-id>/ including coverage ledgers, capability ledgers, artifact contracts, state machines, and error semantics. - Evidence Labeling: Preserves Evidence, Inference, and Unknown labels on every consequential claim, with depth levels from L1 (exists) to L4 (reconstruction-ready). - Subagent-Orchestrated Lanes: Dispatches validated PrdScanPacket contracts to read-only subagent lanes and enforces structured worker result handoffs. - Freshness-Aware Routing: Consults .specify/prd/status.json to route fresh, targeted-stale, or full-stale scans, and resumes interrupted runs from workflow-state.md. - Use Case: Point it at a legacy service repository before writing requirements; it triages capabilities into critical/high/standard/auxiliary tiers and refuses handoff to the build step when critical capabilities lack reconstruction-ready evidence. ## Quick Start Ask the agent to run /sp-prd-scan on this repository to reconstruct its capabilities and produce a scan package ready for PRD compilation.
