bts-reviewlisted

# Code Review Review code for: $ARGUMENTS ## Settings Quality, security, and architecture review require deep reasoning — all three use the main session model by default. Security is recall-bound: semantic flaws (authz bypass, state-dependent privilege escalation) that the initial sweep misses can never be rescued by the rebuttal step, so the sweep itself needs the strong model. Override any agent model via `.bts/config/settings.yaml`: `agents.reviewer_quality`, `agents.reviewer_security`, `agents.reviewer_arch`. ## Step 1: Determine Review Mode Parse $ARGUMENTS: - If first word is `--mid-run` → **mid-run mode** (Phase 11). The second argument is a window like `t-006..t-010` — review ONLY the files those tasks touched, plus the files they import. Architecture perspective only; quality + security agents are skipped. Adversarial Round 2 (rebuttal) is skipped to keep mid-run cost bounded. Output goes to `.bts/specs/recipes/{id}/reviews/midrun-{timestamp}.md`, NOT to `review.md` — the terminal review (no --mid-run) still runs once at implementation completion. - If first word is `security`, `performance`, or `patterns` → **focused mode**. Remaining words = file scope (or all if empty). - Otherwise → **full mode** (all perspectives), arguments = file scope. ## Step 2: Identify Scope and Context **File scope:** If inside a recipe (tasks.json exists): - Read tasks.json for the list of implemented files - If test-results.json exists, also include files from `te