# Action And Motion Understanding

> Use this skill when the user wants questions about what someone did, what happens next, which way something moved, whether one movement is faster or slower, or when two clips look almost the same unless you watch the motion carefully. Trigger it for requests like 'make questions about tiny movement differences', 'ask what happened right after', 'which way did it go', 'who moved faster', or 'make the answer depend on the action sequence rather than the still image.'

- Skill: `dingxingdi/action-and-motion-understanding-2` (Agent Skill, multi-file: 5 files)
- Install (CLI): `npx skillmds@latest add dingxingdi/action-and-motion-understanding-2`
- Raw SKILL.md: https://api.skillmd.com/api/skills/dingxingdi/action-and-motion-understanding-2/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: dingxingdi (https://skillmd.com/u/dingxingdi)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/dingxingdi/action-and-motion-understanding-2

---


# Skill: action-and-motion-understanding
## 1. Capability Definition & Real Case
* **Professional Definition**: The capability to infer what agents or objects are doing from temporally evolving evidence, including fine-grained action type, relative motion, motion direction, and immediate next-step dynamics that cannot be resolved from a single frame.
* **Dimension Hierarchy**: Temporal-Spatial Understanding->Dynamic Event Perception->action-and-motion-understanding
### Real Case

**[Case 1]**
* **Initial Environment**: A 16-second third-person kitchen clip. A person places a food box on the counter, reaches toward a cabinet, and slightly shifts their body orientation before touching the handle. There are no subtitles, and the background objects stay almost unchanged throughout the clip.
* **Real Question**: What will the person do next?
* **Real Trajectory**: Observe the hand trajectory after the box is placed, note the person reorients toward the cabinet rather than the table, and verify that the motion continues into a handle-reaching gesture instead of a pickup or turning-away action.
* **Real Answer**: Open the closet/cabinet.
* **Why this demonstrates the capability**: The question is only answerable by reading motion continuation, not by recognizing objects in a still frame. The decisive evidence is the evolving reach and body turn, which makes this a genuine temporal action-understanding problem rather than static scene recognition.

---
## Pipeline Execution Instructions
To synthesize data for this capability, you must strictly follow a 3-phase pipeline. **Do not hallucinate steps.** Read the corresponding reference file for each phase sequentially:

1. **Phase 1: Environment Exploration**
   Read the exploration guidelines to discover raw knowledge seeds:
   `references/EXPLORATION.md`

2. **Phase 2: Trajectory Selection**
   Once Phase 1 is complete, read the selection criteria to evaluate the trajectory:
   `references/SELECTION.md`

3. **Phase 3: Data Synthesis**
   Once a trajectory passes Phase 2, read the synthesis instructions to generate the final data:
   `references/SYNTHESIS.md`