Skip to content

DexForce/EmbodiChain / .agents/skills/add-embodiment-component

add-embodiment-component

Verified

Add or update a reusable EmbodiChain embodiment YAML component that owns one simulation robot, its sensor suite, and an optional Task Program skill_profile. Use for configs/components/embodiments, new Task Program-compatible robots, or multi-embodiment task deployment.

FileKindSize
SKILL.mdskill md7.1 KB
agents/openai.yamlconfig118 B
---
name: add-embodiment-component
description: Add or update a reusable EmbodiChain embodiment YAML component that owns one simulation robot, its sensor suite, and an optional Task Program skill_profile. Use for configs/components/embodiments, new Task Program-compatible robots, or multi-embodiment task deployment.
---

# Add Embodiment Component

Create the reusable physical and semantic declaration that lets one robot
configuration serve many tasks.

## Establish scope

An embodiment component belongs at:

```text
embodichain/extensions/tasks/configs/components/embodiments/<embodiment_id>.yaml
```

It owns exactly:

- `embodiment_id`;
- one `simulation` robot mapping;
- the independent `sensor` list (mounted defaults belong to `RobotCfg.sensors`);
- optional `skill_profile`;
- optional `grippers` selecting concrete parallel-jaw Cfg defaults.

It does not own a task, physical scene, Task Program, execution policy,
task-local object IDs, or runnable Gym ID. Component files do not use a
compatibility `version` field.

## Load current context

Read `agent_context/MAP.yaml`, then resolve:

- `robot-system`;
- `sensor-system`; and
- `task-programs` when a `skill_profile` is requested.

Inspect the closest current component under
`embodichain/extensions/tasks/configs/components/embodiments/` and the intended runnable
task deployment.

## Reuse or add the robot

If an existing registered robot config can express the request, select it in
`simulation` and add only necessary config overrides. If a new
`RobotCfg`/URDF/control-part implementation is required, invoke `$add-robot`
first. This skill does not duplicate robot registration logic in YAML.

Keep robot-mounted camera calibration in the robot configuration. The component
selects that robot and declares additional independent sensors; do not repeat
mounted defaults in `sensor`. Sensor UIDs must be unique across both sources.
Gym initializes both sources by default and uses `enable_sensor: false` to
disable both, overriding the robot's standalone initialization switch.

## Reuse parallel-jaw hand defaults

Declare `grippers.<instance>` with exact `class_type`, explicit `mode`
(`mounted` or `existing`), `motion_part` and `control_part`. Mounted bindings
also name `component_type` and may supply the ordinary installation `transform`.
Existing bindings append no asset; select a native `variant` when required.
Do not duplicate their hand components or control groups under `simulation`.
Hand Cfgs own native physical data; embodiment bindings own installation and
optional `root_from_tcp`/contact calibration and complete command replacements.

A declared grasp endpoint opts in with `gripper: <instance>`, matching control
part and explicit capability. A unique kind in the hand Cfg's `grasp_generators`
supplies the default; multiple kinds require an explicit policy selection.
Empty support keeps the open/grasp preset but grants no sampler service, so
fixed-pose calls can still use the hand. Requesting an undeclared kind fails.
Matching endpoints may share one hand, generated preset and sampler service.
Composition fills its omitted command preset and supported generator; keep only
search-policy adjustments in runtime services. Do not repeat
physical command presets or old generator `model` fields. Solver TCPs are derived
from the hand reference and fixed installation; any explicit value must agree
after assembly resolves the solver end's fixed chain to the hand root.
Without a semantic profile the same declaration supplies physical assembly only.

## Add `skill_profile` only when needed

Handwritten and RL deployments may use an embodiment without semantic
metadata. A configured Task Program requires `skill_profile` with:

- `contract_id`: the compatibility contract tasks/policies require;
- `profile_id`: stable unversioned profile identity;
- `resources`: logical task-facing resources;
- `command_presets`: named control commands;
- optional `runtime_services`.

Each resource declares logical endpoints. Endpoint declarations map to actual
`control_part` names and list only capabilities the selected robot/runtime
can provide. Grasp endpoints that require open/grasp commands select a matching
command preset.

Use logical resource IDs such as `primary_manipulator`, `left`, or `right`;
Task Program integrations bind Semantic Call slots to these IDs. Do not expose
task-local names as reusable resources.

## Runtime service ownership

Put embodiment-calibrated services in `skill_profile.runtime_services`, such
as gripper-specific antipodal grasp generation or control-part evidence. Put
task-specific targets, poses, and overrides in the task integration.

For reusable joint-position constraint evidence, omit `object_ids` unless the
embodiment truly supports only an explicit global set. Omission lets runtime
assembly scope evidence to graspable objects in the selected task scene.

Only use service kinds accepted by the current configured decoder. Never
serialize a dotted import or callable.

## Attach without copying

A runnable deployment selects:

```yaml
embodiment:
  component: ../../../components/embodiments/<embodiment_id>.yaml
```

Optional robot installation `overrides` are limited to `uid`, `init_pos`,
`init_rot`, `init_qpos`, and bounded calibration: replacing existing named
`joint_drive_props.stiffness`/`damping` values and a declared hand's complete
`open`/`grasp` commands. Prefer these over copying a component for task-local
gains or commands. The separate `camera_overrides` mapping may select
an ordinary Camera UID explicitly declared in `sensor` or `simulation.sensors`
and change only intrinsics/extrinsics. Keep sensor membership, parent, resolution
and output flags with the component. Do not repeat inline `robot` or `sensor`
fields or copy an embodiment merely for scene-specific camera calibration.
Follow the closed [calibration contract](../../../agent_context/topics/env-framework/configuration.md#reusable-gym-deployment-components),
including representation-switch and finite/degenerate-pose checks.

One task can provide multiple thin deployments selecting compatible
embodiments. Keep the program embodiment-independent and validate the
integration and policy contracts against every variant.

## Validation

At minimum:

1. decode the YAML as a closed embodiment component;
2. build the selected robot config and verify control parts;
3. verify the combined independent/mounted sensor UIDs, parent links and shapes;
4. verify every profile endpoint/control part/capability and command preset;
5. compose at least one intended task deployment;
6. for Task Program use, run
   `$add-task-program`'s deployment inspector;
7. run a minimal live robot/sensor smoke test when available.

## Checklist

- [ ] Component selects exactly one robot and declares only additional sensors
- [ ] Existing robot config was reused, or `$add-robot` was completed first
- [ ] Profile is omitted when no semantic consumer needs it
- [ ] Contract/profile/resource/preset IDs are stable and unversioned
- [ ] Endpoint control parts exist on the robot
- [ ] Declared capabilities and commands are actually supported
- [ ] Reusable services contain no task-local object IDs or poses
- [ ] At least one intended deployment composes successfully
writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deploymentStars: 322kUpdated Claude CodeCodexVerifiedInstall
xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy.Stars: 119kUpdated Claude CodeCodexVerifiedInstall
childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release.Stars: 118kUpdated Claude CodeCodexVerifiedInstall