add-semantic-call
Skill240 starsChanged 12 days agoApache-2.0View on GitHub
Add or expose a high-level EmbodiChain Task Program Semantic Call, including registered call descriptors/lowerers, configured runtime-service decoder support, or a deliberately promoted built-in call. Use when a Task Program call is unknown or must lower to an Atomic Skill.
In this repository35 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- add-atomic-actionSkill
- add-embodiment-componentSkill
- add-functorSkill
- add-robotSkill
- add-solverSkill
- add-task-envSkill
- add-task-programSkill
- add-testSkill
- benchmarkSkill
- implement-performanceSkill
- pre-commit-checkSkill
- project-dev-contextSkill
- prSkill
- releaseSkill
- review-prSkill
- update-api-docsSkill
- add-atomic-actionSkill
- add-embodiment-componentSkill
- add-functorSkill
- add-robotSkill
- add-semantic-callSkill
- add-solverSkill
- add-task-envSkill
- add-task-programSkill
- add-testSkill
- benchmarkSkill
- implement-performanceSkill
- pre-commit-checkSkill
- project-dev-contextSkill
- prSkill
- releaseSkill
- review-prSkill
- update-api-docsSkill
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 3 are distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 22
Prefer a registered extension for new provider/task families.
- Very specific: line 29
Read agent_context/MAP.yaml and resolve task-programs plus atomic-actions.
- Distinctive: line 84
does not step the simulator, emit controller commands, or contain a task-local motion generator.
- Very specific: line 97
Add a task integration entry only after the core decoder recognizes its kind.
- Distinctive: line 122
call discovery and duplicate-ID rejection.
- Distinctive: line 131
Use focused tests in:
- Very specific: line 142
Run $add-task-program's inspector against every changed official deployment.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Add Semantic Call
Expose a reusable high-level instruction to Task Program while preserving the boundary between declarative task intent and executable Atomic Skills.
Route the request
Choose the narrowest mode:
| Situation | Action |
| Existing catalog already discovers the requested call | Change only the program/integration config; use $add-task-program |
| Existing Atomic Skill should be exposed under a task/integration-owned call ID | Add a registered Semantic Call extension |
| Required Atomic Skill does not exist | Invoke $add-atomic-action first |
| The concept must become a stable universal language primitive like Pick/Place/HandOver | Add a built-in Semantic Call only when explicitly justified |
Prefer a registered extension for new provider/task families. Do not add a new
SemanticCallSpec subclass for an extension; serialized extensions use
RegisteredSemanticCallCfg and runtime values use exact
RegisteredSemanticCall.
Load current context
Read agent_context/MAP.yaml and resolve task-programs plus
atomic-actions. Verify the current source of truth:
| Concern | Path |
| Call values/catalog | embodichain/lab/task_program/semantics/calls.py |
| Language config/decoder | embodichain/lab/task_program/language/schema.py, decoder.py |
| Static linking/lowering | embodichain/lab/task_program/compiler/lowering.py |
| Registered extension declarations | embodichain/lab/task_program/integrations/extensions.py |
| Configured integration decoder | embodichain/lab/task_program/integrations/configured.py |
| Allowlisted simulation factories | embodichain/lab/task_program/integrations/_configured_services.py |
| Runtime assembly | embodichain/lab/task_program/integrations/simulation/ |
Inspect the target Atomic Skill's exact SkillDescriptor, goal, options, and
binding contract before designing the Semantic Call.
Registered Semantic Call mode
1. Define stable declarative identity
Choose a lowercase dotted call ID such as simulation.articulation_link_slide.
Define the smallest JSON-compatible argument mapping needed to select semantic
entities or configured values. Arguments contain no tensors, callables, import
paths, live objects, or motion generators.
Program form:
kind: registered
call_id: vendor.my_call
arguments:
target: logical_scene_entity2. Declare descriptor and lowerer coverage
The integration catalog must include one exact SemanticCallDescriptor whose:
call_idmatches the serialized call;spec_typeisRegisteredSemanticCall;- target descriptor is the intended Atomic Skill descriptor.
It must also own exactly one fingerprinted
RegisteredSemanticLowererFactory for the same call ID and revision. The
factory creates a fresh live lowerer for each adapter assembly.
3. Keep lowering typed and narrow
The lowerer:
- strictly validates canonical arguments;
- resolves only declared semantic references/live services;
- produces the target Atomic Skill's typed goal;
- consumes action options from the bound profile/preset;
- declares look-ahead effects/targets when required by static analysis;
- does not step the simulator, emit controller commands, or contain a task-local motion generator.
Lowerers cannot replace the catalog-owned descriptor, resource binding, effects, or options contract.
4. Expose configured YAML only through an allowlist
If integration.yaml.runtime_services.registered_semantic_lowerers must
instantiate the new provider family, add an explicit closed decoder branch and
typed factory in the configured integration modules. Never accept a generic
class_type, dotted import, or arbitrary kwargs escape hatch.
Add a task integration entry only after the core decoder recognizes its
kind.
Built-in Semantic Call mode
Promote a call to the built-in language only when its semantics, arguments, resource contract, effects, and Atomic Skill target are stable across providers and tasks. Update the complete vertical slice:
- language schema and strict decoder;
- compiler conversion;
- immutable semantic call and built-in catalog descriptor;
- static linking/lowering and effect analysis;
- public exports;
- configured integration options/monitors if applicable;
- focused docs and tests.
This is a language change, not a task integration shortcut. Preserve strict unknown-field rejection and exact types.
Validation
Registered-call coverage should prove:
- valid and invalid serialized argument shapes;
- call discovery and duplicate-ID rejection;
- exact descriptor-to-Atomic-Skill match;
- exactly one lowerer factory and stable fingerprint declaration;
- fresh lowerer construction;
- typed goal/options output;
- look-ahead/effect behavior;
- compiler preflight with missing/duplicate lowerers rejected;
- configured deployment decoding when a new service
kindwas added.
Use focused tests in:
tests/lab/task_program/semantics/test_calls.py
tests/lab/task_program/test_decoder.py
tests/lab/task_program/test_semantic_compiler.py
tests/gym/envs/task_program/test_catalog.py
tests/gym/envs/task_program/test_configured_integration.py
tests/gym/envs/task_program/test_task_vertical_slices.pyRun $add-task-program's inspector against every changed official deployment.
Live provider behavior still needs a real environment qualification.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

