hybrid · open-rua, robodojo — Claude Code prompt: what it does, how to install
The hybrid prompt for Claude Code reports RoboDojo task progress with evidence-backed subgoals. A reviewer checks action-model proposals before execution.
MIT ★ 488 updated 2026-11-15
.claude/commands/hybrid.mdWhat hybrid does
The prompt defines what task_progress sent to actor_act should contain: verified subgoals, the current attempt, and work still remaining. This is progress reported by the model, not an environment verification flag. If a previously achieved relation is disturbed, update the progress report.
When enabled, the reviewer assesses each new action-model proposal before its prefix runs. It receives current images and measured state, recent feedback, prior actor context, and robot-only FK, but not the agent’s full conversation. The assessment comes before execution and may differ from the final returned state, so check later effects against the returned images and state. A reviewer stop or phase_complete is evidence to inspect, not proof that the whole task is done.
What it helps with
- report verified and remaining RoboDojo subgoals in
task_progress - correct progress after an achieved relation is disturbed
- assess an action proposal before its prefix executes
- check task results against returned images and state
Example prompts for the agent
- Send the verified, current, and remaining RoboDojo subgoals to
actor_act. - Assess the action-model proposal before its prefix executes.
- Check the returned images and state to see whether the RoboDojo task is complete.
How to install hybrid
Manually: put the files in place
.claude/commands/hybrid.mdHow to use
Paste the text at the start of a chat or into project instructions
Questions about hybrid
How do I install hybrid in Claude Code?
Save the prompt file at .claude/commands/hybrid.md.
What does hybrid do?
It sets the RoboDojo progress format for actor_act and describes how a reviewer assesses action-model proposals.
Does phase_complete mean the whole task is finished?
No. A reviewer stop or phase_complete is evidence to inspect, not proof of global completion. Check the returned images and state for later effects.
Source text
hybrid.md · MIT
For RoboDojo actor_act, pass evidence-backed task_progress: verified_completed, currently_attempting and remaining subgoals. This is model-reported progress, not an environment verification flag. Correct it when a previously achieved relation is disturbed. When enabled, the RoboDojo reviewer assesses each fresh action-model proposal before its prefix executes. It receives current images and measured state, recent feedback, previous actor context and robot-only FK, not the agent's full conversation. Previous actor images may predate intervening Direct actions; they are not guaranteed before-images of the latest action. FK describes robot kinematics, not future object motion, attachment or collision-free passage. An assessment describes evidence available before that chunk executes. It may precede actor_act's final returned state, so verify later effects from the returned images and state. A reviewer stop or phase_complete report is evidence to inspect, not proof of global completion. Inspect whole-call execution totals when an early stop, partial execution or error follows earlier chunks.
© Open-RUA · MIT · Open-RUA/uni-robot-agent
Similar prompts
All prompts for Agent setup: memory, context, orchestration →hybrid by other authors comes first.