specflow-onboard-codebaselisted

# Specflow: Onboard Existing Codebase Turn an existing codebase into a spec-managed project. Read code, extract what the specs *should have been*, challenge every judgment, verify every spec, then let the human correct. ## What this skill produces Two parallel spec trees, bidirectionally linked, plus folder overviews: - `.specflow/specs/` — developer specs with entity references (READS/WRITES), rules, Given/When/Then criteria. Specs reference entities by name but do not define schemas — the model/migration is the single source of truth for field definitions. - `.specflow/specs-business/` — business specs with outcomes, journeys, business rules, success metrics. Organized in domain subfolders. Filenames are journey-oriented, starting with the persona (`user-fills-dynamic-form.business.md`, not `form-submission.business.md`). - `_overview.md` in every folder of both trees - Bidirectional `implements:` (dev, single value) / `implemented_by:` (business, list) - `onboarding-scratch/atoms/` — retained atom extraction records (the audit trail) - Delta analysis: bug findings filed in the `.cortex/compass/bugs/` ledger, plus implicit-behaviors.md, dead-features.md, link-map.md - Investigation records and verification feedback in proposed-notes.md ## Agent Delegation Model Onboarding a large codebase exceeds what a single agent can hold in context. Each subagent in Claude Code gets a ~200K token context window, of which ~150K is usable after system prompt, tools, and sk