hubo-reviewlisted

# Hubo Review Review the user's specified programming target without changing it. Create exactly two persistent top-level roles: - `review_agent`: owns the canonical review. - `critic_agent`: tries to disprove the review, expose false positives, and find material omissions. Start this role in standby. ## Host mapping Use the first available host: - Codex: create with `spawn_agent`, resume with `followup_task`, and wait with `wait_agent`. - Claude Code: create with `Agent`, resume with `SendMessage`, and consume completion notifications. - GitHub Copilot CLI: create background roles with `task`, resume with `write_agent`, and consume completion notifications or `read_agent`. - OpenClaw: create kept sessions with `sessions_spawn`, resume with `sessions_send`, and wait with `sessions_yield`. Create each role once and reuse its ID for every round. The critic's initial task returns only `READY`. Optional descendants inherit their parent's read-only boundary. If the host cannot create and resume two addressable roles, ask whether the user accepts a sequential single-agent fallback. Both lineages are read-only. Block mutation tools mechanically when possible. Otherwise compare content-sensitive before/after evidence for the user-specified mutable target around every role turn; use target-specific evidence when no worktree represents it and report when the boundary cannot be verified. ## Shared core principles Apply this complete list to the coordinator, both roles, and ever