characterize-brownfieldlisted

# Characterize Brownfield Reverse-engineers an existing codebase into structured, confidence-scored artifacts. Every claim is either a FACT (from a tool call — S7) or an INFERENCE (LLM interpretation), and every inference carries a confidence: **High** (directly observed, e.g. a parsed OpenAPI file) / **Medium** (inferred from strong signals, e.g. route patterns + naming conventions) / **Low** (guessed from weak or absent evidence). Honesty about confidence is the point — a brownfield PRD built on fabricated certainty is worse than one that says "unknown." **Boundary.** Read-only. Never edits code. Does not write a PRD (see `compose-brownfield-prd`). Does not create issues or stories. ## Inputs - Repository path (required). - Depth: `quick` (2-5 min, pattern-based, no source reads) / `deep` (10-30 min, critical dirs — **default**) / `exhaustive` (30-120 min, all source, opt-in only — warn the user of cost first). - Focus directories (optional) — narrows deep/exhaustive scans. ## Procedure ### 1. Structure & stack (FACT, via tool calls — S7 only, never recall) - `find . -maxdepth 3 -type d` (or workspace tools) for top-level layout. - Read manifest files present: `package.json`, `*.csproj`/`*.sln`, `requirements.txt`/`pyproject.toml`, `pom.xml`/`build.gradle`, `go.mod`, `Cargo.toml`. Extract exact framework + versions — never guess. - Read config files present: `tsconfig.json`, `.editorconfig`, lint configs, CI workflow files. - `git log --since="90 days ago" --name