Embodied Task Decomposition

Decompose complex physical tasks into atomic subtasks for robot execution. Use when user provides: (1) An image showing a physical scene (indoor/outdoor), an...

Security Scan

The skill's requirements, instructions, and included code are consistent with its stated purpose (decomposing image+task inputs into atomic robot actions); it requests no external credentials or installs and contains only a local validator script.

Purpose & Capability

Name/description (robot task decomposition from image+text) match the included files: an action bank, examples, and a local validator script. No unrelated binaries, credentials, or config paths are requested.

Instruction Scope

SKILL.md stays within scope: analyze image + task text, produce atomic subtasks from a fixed action bank. It also permits the agent to add new actions to action-bank.md; this is reasonable for extensibility but means the agent may modify skill files if allowed to write them. The instructions do not direct reading unrelated system files or sending data to external endpoints.

Install Mechanism

No install spec (instruction-only with one local validator script). Nothing is downloaded or written to disk by an installer; lowest-risk install posture.

Credentials

No environment variables, credentials, or external service tokens are required. The validator reads only the included action-bank.md and the provided subtasks; no access to unrelated secrets or system config is requested.

Persistence & Privilege

always:false and no elevated privileges. The only persistence-related behavior is the agent MAY add actions to action-bank.md per instructions; modifying the skill's own files is plausible for this feature but you should ensure the agent's runtime is permitted to write only intended files.

Assessment

This skill is internally coherent and does not request credentials or install external code. Things to consider before installing: (1) the skill allows adding entries to action-bank.md — if you allow the agent write access it could modify the skill files, so review any changes to that file; (2) run the included scripts/validate.py on outputs or proposed new actions to catch duplicates or malformed subtasks; (3) avoid sending sensitive or private images to an agent you haven't audited, and restrict the agent's file write permissions to the skill directory if possible.

This skill decomposes high-level natural language instructions into atomic subtasks that a robot can execute.

When to Use

• User provides an image AND a task instruction
• User asks to "decompose", "break down", or "split" a task
• User wants step-by-step actions for robot execution

Input Format

1. Image: Photo of the physical scene (any environment: kitchen, office, outdoor, etc.)
2. Task Instruction: Natural language description of what to accomplish

Example:

Task Instruction: take toasted bread from bread machine on white table place on plate
Image: [image path or description]

Output Format

Numbered list of subtasks, each following format:

{action} {target} {location/optional prepositional phrase} with {left/right/either} gripper

Process

1. Analyze the image - Identify objects, surfaces, locations, tools visible
2. Understand the task - What is the goal? What needs to be moved/ manipulated?
3. Break into atomic actions - Each subtask = one action from the action bank
4. Specify gripper - Always indicate left, right or either gripper

Action Bank

Refer to action-bank.md for the complete list of allowed actions. All subtasks MUST use actions from this bank.

Examples

See examples.md for detailed decomposition examples across different domains.

Important Notes

• Use ONLY actions from the action bank
• Each subtask = one primary action
• Always specify gripper (left/right/either)
• Include target object and location
• Keep subtasks atomic and sequential
• Consider object state changes (e.g., "open bag" before "take fruit")

Updating the Action Bank

The agent MAY add new actions to the action bank when needed. To add a new action:

1. Check for duplicates - Search existing actions for similar functionality
2. Verify functional difference - New action must serve a distinct purpose
3. Add with documentation - Include description and example usage

Duplicate Check Criteria

A new action is considered a duplicate if it:

• Has the same name as an existing action
• Describes the same physical movement (e.g., "lift" vs "raise")
• Can be used interchangeably with an existing action in all contexts

Adding a New Action

When adding to action-bank.md, follow this format:

| action_name | Description | Example Usage |
|-------------|-------------|---------------|
| new_action | What it does | "new_action the object"

Example of adding "insert" (different from "place" - "place" = put on surface, "insert" = put into container):

| insert | Put object inside a container or slot | "insert the key into the lock"