threnody-swarm-reviewlisted

# Threnody swarm review ## Overview This is the **deep specialist** review mode. It is intentionally more expensive than `threnody-fast-review`; use it when the user asks for deep/security-critical review, threat modeling, or a named dimension such as `[dims=performance]`. Complexity gating limits token blowout: | Band | LOC | Dimensions run | |------|-----|----------------| | trivial | < 50 | logic, edge | | moderate | 50–200 | logic, edge, types | | complex | > 200 | logic, edge, types, security, performance | Security is always added when the file contains risk signals (SQL, exec, auth, tokens, etc.), but ordinary risk words do **not** automatically force high tier. **Tier → model is chosen per agent from raw file size × dimension reasoning-weight** (no LLM call — pure heuristic, keeps fast-start): | File/profile | edge / types (light) | logic / performance / security (reasoning-heavy) | |--------------|----------------------|--------------------------------------------------| | small `< 230` | low → cheap tier | low, or medium for dense/security-risk code | | mid `230–600` | medium | medium, high only when dense | | large `> 600` | medium | high unless the file is flat/config-like | | ordinary security risk | — | medium by default | | concrete exploit primitive or explicit deep/critical request | — | high | Synthesis is `medium` by default. It escalates to `high` only for explicit deep/security-critical intent, concrete exploit primitives, or large finding sets (f