rh-guardlisted

# rh-guard Reward Hack Guard (`rh-guard`, RH Guard) is a **live hazard gate** on coding-agent tools (Claude Code, Cursor, Codex, Grok Build, Pi, Amp, Prime Agent, and DeepSeek Harness / `dsh`). Same `scoreEvent` path; multi-host adapters per install surface. Structural detectors deny writes to designated evaluation assets. TypeSafe Jev (System One) scores remaining events. Code owns thresholds and steer copy. This skill is the protocol for *using* those verdicts and for designing evals they protect. It is **not** how to boot the Next server. **Do not merge products.** [Augustus](https://github.com/24601/Augustus) is design-judgment for *where* typed System One judgment belongs. rh-guard is the runtime gate on agent tools. Use Augustus to place judgments; use this skill when the judgment is reward-hack risk on a coding-agent hook. [JevLint](https://github.com/huntedman/JevLint) is semantic convention lint (plain-English plugins, file-level Nouls) in a write → check → fix loop— quality, not gaming. Distinct from [mizchi/jev-lint](https://github.com/mizchi/jev-lint) (name/comment/test truth-of-contract; findings are candidates, not verdicts). [jevgate](https://github.com/thevibeworks/jevgate) is a sibling Bash gate: an allowlist proves what may run; writers, wrappers, and credential-shaped input are refused in code and never sent to the model; Jev judges only unlisted verbs; it cannot block (unsure → the agent's own permission prompt). Hard envelope owns safety; soft judgmen