거북이코드 — 계획↔구현↔검증 게이트. Claude Code PreToolUse hook으로 구현 전 계획 케이스 누락·시나리오 미지정을 차단하고, gbc verify로 구현 후 케이스를 증거(JUnit·코드 독해)와 대조한다(verified>reviewed>unverifiable).
NestJS authorization — Laravel-style Gates & Policies, @Can guard, resource resolver (zero DB).
Flight checklist CLI that gates a Claude Code skill's phases — a stage cannot open until every check in every prior stage is recorded as passing
Hash-chained evidence ledger recording what each ADLC gate verified — agentic provenance with optional HMAC signing (C11).
EMILIA Gate — the Trusted Action Firewall. Deny-by-default enforcement for consequential machine actions: an action runs only with a valid, in-scope, sufficiently-assured, non-replayed EMILIA authorization receipt (proof a named human authorized this exac
Scheduled re-prosecution of hot paths — every frontier model release as a free re-audit (C12).
Mines human PR review objections into structured prosecution lenses (C13).
Conflict forecast and dispatch schedule for parallel ticket execution (D2).
N-version programming for failing tests — fan N LLM completions, group by agreement, recommend the smallest consensus fix (C7/P4).
Rail-freeze enforcement and suppression-marker gate — blocks builder edits to frozen paths (C5).
Diff-scoped mutation gate — the honest coverage check, mutating only changed lines (P3/C4).
Converts prosecution findings into permanent defenses — the compounding closer (P7).
Session-log flail analysis — mechanical two-strike rule for build agents (C6/P4 supervisor).
Standing red-team and gate calibration for the ADLC toolkit — generator-adversary fan-out (C-tool).
Measures reviewer recall via planted bugs — who reviews the reviewer (P5 meta-gate).
Deterministic per-ticket model assignment from the ticket DAG and ledger — no LLM calls (D1).
Behavior-space diff for the P6 human gate — review what changed in behavior, not 5,000-line diffs.
Surface — a deterministic gate that blocks merges when docs drift from the code they describe. Thin npm shim over the prebuilt surf binary.