h0ngcha0/hubo

# Hubo

Run Hubo only after explicit user activation. Coordinate exactly two persistent top-level roles in the current conversation:

- a pragmatic work agent whose lineage alone may edit files; and
- an independent, read-only review agent that challenges the implementation and its evidence.

Read [the complete Hubo workflow](references/workflow.md) before creating either role. Select exactly one current host from [the supported host adapters](references/hosts.md); never mix host APIs.

The work agent implements the smallest complete change and reports its diff, decisions, and verification. The review agent inspects the actual code and returns `CLEAR` or numbered findings. Relay every finding back to the same work agent, which must respond `FIXED`, `PUSHBACK`, or `NEEDS_USER` with evidence. Resume the same reviewer to verify the response, then continue without an arbitrary round limit.

Stop only when the reviewer returns `CLEAR` with every finding closed or withdrawn, or when a genuine requirement or technical-direction decision needs the user. Mirror every completed work and review report into the conversation, then include the complete chronological exchange in the final response. Never store the exchange in a file.

# 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`.