moving-target-control — Claude Code skill: what it does
The moving-target-control Claude Code skill plans contact with moving physical objects and checks whether a grasp holds. It activates when a target or receiver may move during approach or transfer.
MIT ★ 488 updated 2026-11-15
npx skills add Open-RUA/uni-robot-agent --skill moving-target-control -a claude-codeWhat moving-target-control does
The skill plans a grasp around observed motion, then checks the result. It compares observations with their capture times or environment steps to estimate direction and local speed. From there, it selects a reachable contact area and accounts for the target's movement during approach and gripper closure. It also checks finger clearance and the target's drift through the closing area.
Geometry needs a fresh update near contact; a long chase based on an old position can miss. After closure, the skill checks how the object moves relative to the hand and can use a short lift when appropriate. If the object slips or the receiver moves, it revises the grasp or transfer plan. It does not assume speed stays constant after a disturbance.
What it helps with
- estimate a target's direction and local speed from observations
- choose a reachable contact area while accounting for delay
- check whether a closed gripper holds the object
- revise a transfer plan after the receiver moves
Example prompts for the agent
- Grasp the moving object using fresh observations near contact.
- Check whether the object stays attached after closure and a short lift.
- Revise the transfer plan because the receiver moved while the hand pursued the target.
How to install moving-target-control
1Skills CLI (Vercel)
npx skills add Open-RUA/uni-robot-agent --skill moving-target-control -a claude-code2Manually: put the files in place
~/.claude/skills/moving-target-control/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/moving-target-control/
1Skills CLI (Vercel)
npx skills add Open-RUA/uni-robot-agent --skill moving-target-control -a codex2Manually: put the files in place
.agents/skills/moving-target-control/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add Open-RUA/uni-robot-agent --skill moving-target-control -a cursor2Manually: put the files in place
.cursor/skills/moving-target-control/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install Open-RUA/uni-robot-agent uni_robot_agent/resources/skills/general/gripper/moving-target-control2Manually: put the files in place
.github/skills/moving-target-control/SKILL.md1Skills CLI (Vercel)
npx skills add Open-RUA/uni-robot-agent --skill moving-target-control -a gemini-cli2Platform CLI
gemini skills install https://github.com/Open-RUA/uni-robot-agent --path uni_robot_agent/resources/skills/general/gripper/moving-target-controlManually: put the files in place
.devin/skills/moving-target-control/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about moving-target-control
How do I install moving-target-control in Claude Code?
Run npx skills add Open-RUA/uni-robot-agent --skill moving-target-control -a claude-code. The skill file is located at ~/.claude/skills/moving-target-control/SKILL.md.
What does moving-target-control do?
It plans contact with a moving object using observed motion and fresh geometry, then helps distinguish a miss, a successful grasp, and a later slip.
How does the skill verify a grasp?
It checks the completed closure and, when appropriate, a short lift. A centered wrist image or blocked fingers alone do not confirm that the object was captured.
Source text
SKILL.md · MIT
--- name: moving-target-control description: Capture or interact with a moving physical target using fresh geometry and observed motion. Use when the target or its receiver can move during approach, closure or transfer. --- # Plan contact against motion, then verify capture Use `general/gripper/scene-memory/SKILL.md` to preserve identity and reference images. Recognizing an object does not establish its current contact position. ## Prepare a reachable contact window Compare observations with their capture time or native step to establish motion direction and a local speed estimate. Use actual environment advancement: in a step-driven scene, reasoning time alone need not move the target; in a continuously advancing scene, account for elapsed time and observation latency. Do not mix these clocks or assume constant speed after a disturbance. Choose a reachable contact region and prepare a compatible orientation before the target arrives when useful. Allow for movement during the executed approach and closure, not only the displacement between images. Check finger clearance, solid contact material and relative drift through the closing region. Refresh geometry near final contact; shorten the segment whose accuracy depends on an old moving-target position instead of queuing a long stale chase. Before capture, preserve the receiver and the subsequent transfer route. A container on moving support may leave reach while the other hand pursues the target. Use `general/gripper/placement-and-engagement/SKILL.md` for support roles. ## Distinguish a miss, capture and later loss Inspect the completed closure and, when lifting is appropriate, a short lift. Compare the target's original support with its motion relative to the hand; centered wrist imagery or blocked fingers alone do not establish capture. Use `general/gripper/state-verification/SKILL.md` for unresolved attachment. After a miss, update identity, current position and reachable contact window. After slip, reconsider grip geometry and the loaded motion as well as timing; one successful lift does not establish retention for the whole transfer. If the receiver moved or was lost, revise the dependent placement plan before continuing the carry. Wait for a return only when observations or the instruction support recurrence; do not assume a disappeared target will circulate back.
© Open-RUA · MIT · Open-RUA/uni-robot-agent