hybrid · open-rua, runtime — Claude Code prompt: what it does, how to install
The hybrid Claude Code prompt helps choose between actor_act and direct robot motion. It guides a bounded control phase and the checks that follow.
MIT ★ 488 updated 2026-11-15
.claude/commands/hybrid.mdWhat hybrid does
The prompt helps an agent choose a control method for a robot phase: actor_act for repeated visual feedback, or direct motion when the target, route, and observation checkpoint are clear. The instruction for actor_act should briefly name the object and intended effect.
It follows the tool’s limits, including reviewer availability, internal chunk budgets, and allowed control groups. After the call, the agent checks the execution total, stop reason, state, and images. One call marks a main-agent decision boundary; a mid-phase check requires separate calls.
A chunk budget does not establish success or failure. Partial execution can include completed chunks, and continuation depends on the integration’s recovery and operator-handoff conditions.
What it helps with
- choose between
actor_actand direct motion - split a control phase to check it midway
- verify an action using state and image evidence
- revise instructions for an unfinished phase
Example prompts for the agent
- Choose a control method for this robot phase and verify the result.
- Run this local phase with repeated visual feedback through
actor_act. - Split the phase into separate calls so the state can be checked midway.
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?
Add the prompt file at .claude/commands/hybrid.md.
What does hybrid do?
It guides the choice between actor_act and direct motion, then describes what to inspect after execution.
Can I check actor_act between its internal chunks?
No. One call is a main-agent decision boundary. Split the phase into separate calls to check it midway.
Source text
hybrid.md · MIT
Use actor_act for a bounded local phase that needs repeated visual feedback and adjustment. Use direct motion for a segment whose target, route and observation checkpoint are clear. Neither precision nor recovery alone requires a particular tool, and geometric pre-positioning is not an actor prerequisite. Give one short instruction naming the object and intended effect. It remains fixed throughout the call. Follow the tool's declared reviewer availability and chunk limits. Internal chunks obtain fresh observations and proposals. Review before execution occurs only when the tool declares an enabled reviewer. A chunk budget is not evidence of completion or failure. Choose active control groups for the whole phase. Use only constraints declared by the tool schema. They apply throughout the call, starting with its first action and including reviewer corrections. Omit a persistent actuator override when the policy needs to vary that actuator during the phase. One actor_act call is one main-agent decision boundary. Inspect its total execution, stop_reason, state and images after it returns. If a phase needs a main-agent check or different constraints midway, split it into separate calls. Do not plan to insert actions between internal chunks. An interruption or partial execution can follow earlier completed chunks; totals describe confirmed execution and do not resolve an uncertain request. Verify final effects with the evidence required for the environment's Direct actions. For an unfinished phase, reassess motion and intended effect and revise the instruction, viewpoint or approach. A short unsuccessful prefix does not establish that the action model cannot perform the task. Continue only while the integration's recovery and operator-handoff conditions permit it.
© Open-RUA · MIT · Open-RUA/uni-robot-agent
Similar prompts
Prompts catalog →hybrid by other authors comes first.