Skip to main content
skill CX Codex

contact-effect — Claude Code skill: contact verification

The contact-effect Claude Code skill checks contact by observing how the target responds and guides adjustments. It activates for new or uncertain contact, or after a change to grip, tool geometry, or support.

MIT ★ 488 updated 2026-11-15

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add Open-RUA/uni-robot-agent --skill contact-effect -a claude-code
Open on GitHub

What contact-effect does

The skill tests whether a motion produces the intended physical effect, such as moving an object with a push or activating a button with a press. It proposes predicting an observable change, making a controlled probe, then comparing the target before and after.

For a button, contact alone does not prove activation. Evidence can come from cap travel or a visible functional response; hand movement and a reached pose are insufficient on their own. When the response is missing, the skill separates possible causes such as misalignment, grip slip, obstruction, or movement of the support.

It counts supported effects or distinct observed transitions, not tool calls or repeated frames of one state. A calibration probe can count toward the requested total. Exact-count guidance is available in general/gripper/contact-effect/references/exact-count-cycles.md.

What it helps with

  • check whether a push moves the target in the intended direction
  • distinguish button activation from mere contact
  • identify why contact failed or a grip slipped
  • count confirmed press cycles without double-counting

Example prompts for the agent

  • Check whether the button activated after the press, and separate contact from activation.
  • The object did not move after the push; identify likely causes and suggest a contact adjustment.
  • Count confirmed button presses and check whether release occurred between cycles.

How to install contact-effect

1Skills CLI (Vercel)

npx skills add Open-RUA/uni-robot-agent --skill contact-effect -a claude-code

2Manually: put the files in place

~/.claude/skills/contact-effect/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/contact-effect/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about contact-effect

How do I install contact-effect in Claude Code?

Run npx skills add Open-RUA/uni-robot-agent --skill contact-effect -a claude-code. The skill file is installed at ~/.claude/skills/contact-effect/SKILL.md.

What does contact-effect do?

It checks physical contact by observing the target’s response and helps revise the motion when the expected effect is missing. For button presses, it distinguishes evidence of activation from simple touch.

How does contact-effect count button presses?

It counts supported effects or distinct observed transitions rather than tool calls. A calibration press can contribute to the event total.

Source text

SKILL.md · MIT
---
name: contact-effect
description: Verify and adjust contact from observable effects on the target. Use for a new or uncertain contact, or after grip, working geometry or support changes during pushing, pressing, insertion or tool use.
---

# Test the effect at the target

## Predict, probe and compare

Before an unverified contact, identify the working surface and one observable
change expected on the target. Choose a short, controlled probe large enough to
distinguish outcomes while limiting disturbance. Set its extent from the current
geometry, clearance and tool limits; there is no universal distance or duration.
Use an existing pre-action view as the reference when it is still valid.

Inspect the target after execution. Distinguish movement in the intended direction
from rotation, tipping, slip or motion of the support. Evidence sufficient for
one layer below does not automatically establish the next.

| Evidence | Establishes | Still needs evidence |
|---|---|---|
| Hand or wrist moved | The command produced robot motion. | The intended object or tool followed. |
| Held object or tool followed | Control over that object during the observed motion. | Its working surface acted effectively on the target. |
| Target changed | A physical effect occurred. | The effect matches the required direction, state or event. |
| Required effect observed | The local requirement has evidence. | Other requirements and final relationships are satisfied. |

## Inspect the relevant response

| Contact | Compare before and after | When the response is wrong or absent |
|---|---|---|
| Push or slide | Target translation, rotation, tipping and maintained support. | Change the contact point, height or direction rather than extending the same ineffective stroke. |
| Held-tool contact | Working end relative to the grip, its target contact and the target response. | Separate grip slip from ineffective contact; re-localize the changed working geometry. |
| Press or trigger | Local displacement, visible state change or an available activation signal; release when another cycle requires it. | Keep contact, mechanical movement and functional activation distinct. |
| Insert or engage | Entry at the opening, exposed length, relative axes and edge deflection. | If wedged or tipped, withdraw safely before correcting alignment. |
| Constrained motion | Moving part relative to its base, and any unintended base movement. | Restore support or revise the contact if the whole assembly moves instead. |

A clear visual state change can confirm an effect without a separate event
signal. Conversely, apparent touching or pose_reached alone cannot establish
activation. Use only observations and feedback exposed by the tools; do not
assume access to hidden success flags, counters or simulator state.

## Revise the failed hypothesis

If the expected change is absent, distinguish missed contact, slipping grip,
blocked target or an unsuitable force direction using available evidence.
Keep an unobserved cause as a hypothesis. Change the relevant contact factor and
test again; a longer stroke or unchanged repetition may preserve the failure.
With a held tool, locate the working end again after slip or reorientation.

## Establish an effective press before repeating it

Identify the movable button cap, its pressing direction, and an observable
response that distinguishes activation from mere contact. Record a useful
unloaded baseline while the cap is visible.

Align the actual contacting part of the finger or held tool with the cap.
The finger-center target is not automatically the physical contact point.
Approach with clearance, then use bounded contact motion along the intended
pressing direction.

Inspect the cap's travel or a visible functional response. Hand descent,
pose_reached, a tracking stall, or firm-looking contact alone does not prove
activation. A spring-return button may already be raised in a post-release
image; inspect evidence from the pressed state as well as the released state.

If the response is absent or ambiguous, keep the press unconfirmed:

- Check whether contact is on the movable cap or the fixed housing.
- Check alignment, pressing direction, obstruction and unintended base motion.
- Compare actual motion and target error with the required contact geometry.
- Check whether arrival tolerance allowed stopping before an effective press;
  tighten it only when the target and clearance justify it.
- Then choose a small additional stroke, a corrected approach, or another
  contact geometry based on the observed cause.

Do not conclude that a button is rigid or touch-activated merely because a
press failed to move it. Do not keep lowering an unchanged target when the
hand is stalled or the assembly is moving instead.

## Reuse verified effects and count events

Establish one effective press-and-release cycle before repeating its motion.
Count a press only when the required effect has supporting evidence, and
establish release or rearming before the next cycle.

Reuse a verified stroke while grip, working geometry, target and support remain
compatible. Recheck after slip, collision, changed alignment or an unexpected
response, not every known segment. Keep an observation boundary before motion
that depends on an unresolved effect; known approaches can still be batched.
Do not batch further unverified cycles to satisfy a numerical target.
For tool strokes, disengage the working end before lateral travel.

Count confirmed effects or distinct observed transitions, not calls, hand
travel or frames showing the same state. For exact-count tasks, keep confirmed
and uncertain events separate. Inspect an ambiguous attempt before repeating
it: an unseen activation could otherwise be counted twice. Tool-call history
proves attempted cycles, not the number of successful presses.
Use `general/gripper/state-verification/SKILL.md` when the remaining question
concerns the object's resulting state or completion evidence across operations.

A calibration probe is a real interaction and can already contribute to the
required event count. Do not reset the physical count when switching from
calibration to execution. If exact count or ambiguous rearming matters, read
`general/gripper/contact-effect/references/exact-count-cycles.md`.

© Open-RUA · MIT · Open-RUA/uni-robot-agent

Similar skills

SKILL.md skills catalog →

From the same set41

The whole set →
scene-memory The scene-memory Claude Code skill tracks object identities, reference states, and event order. It activates when an agent’s next action depends on an earlier observation, a demonstration, or another actor’s move. CX Codex +1 ★ 488repo stars robodojo-motion-control The robodojo-motion-control Claude Code skill explains RoboDojo motion planning, IK recovery, and diagnostic counters. It helps interpret why a robot did not reach its requested pose. CX Codex +1 ★ 488repo stars supported-contact The supported-contact Claude Code skill plans object contact and the assisting hand’s motion. It activates when a held load needs a receiver adjusted or restrained at the same time. CX Codex +1 ★ 488repo stars pose-targeting The pose-targeting Claude Code skill builds robot motion targets from a position and orientation. It activates for tool changes, reorientation, and aligning a feature on a held object. CX Codex +1 ★ 488repo stars placement-and-engagement The placement-and-engagement Claude Code skill plans and checks object placement, support, containment, and mechanical engagement from visible features. It applies when receiver stability or relative geometry shapes the requested result. CX Codex +1 ★ 488repo stars state-verification The state-verification Claude Code skill checks object states and relationships using distinguishing evidence. It activates when a next step depends on whether an object is held, placed, or correctly oriented. CX Codex +1 ★ 488repo stars grasp-geometry This Claude Code skill selects grasp geometry based on exposed material, finger access, and the object's next use. It covers thin, hollow, crowded, or partly blocked objects. CX Codex +1 ★ 488repo stars motion-recovery The motion-recovery Claude Code skill helps diagnose robot motion failures and choose a changed route. The agent consults it before a difficult reorientation or after motion stalls. CX Codex +1 ★ 488repo stars

Project sponsors

Companies that keep this analytics open