Definition
When intent is expressed directly in natural language, its structural ambiguity prevents reliable compilation, capability checking, or reconstruction of meaning. The absence of a formal intent layer results in fragile execution behavior and non-attributable decisions.
Semantic DSL defines a structured form of intent that is:
- capability-aware,
- machine-interpretable,
- deterministically lowerable into Semantic ISA,
- reconstructable from execution traces.
Semantic DSL is not a prompt format and not a general-purpose programming language. It is the semantic program layer that precedes execution and constrains how intent becomes admissible for compilation.
Role in the Execution Stack
Semantic DSL occupies the highest layer of the semantic execution hierarchy:
Its role is to:
- normalize intent into structured semantic programs,
- enforce capability constraints defined by AgentIDL,
- ensure every admissible semantic term has a deterministic ISA mapping,
- preserve semantic meaning independently of the underlying model.
Without this structured layer, language-driven systems collapse intent and computation into an inseparable operation, preventing reliable auditing or comprehension.
Motivation
As AI systems transition from suggestion engines to operational participants, the primary failure modes arise not from incorrect content but from unstable or ill-formed intent structures. Typical issues include:
- intent expressions that vary semantically across models or linguistic phrasing,
- inability to ensure consistency across repeated runs,
- ambiguity in determining whether an action aligns with declared capabilities,
- loss of semantic provenance once execution begins.
Semantic DSL resolves these issues by providing a stable, capability-aligned semantic surface where intent can be validated, transformed, and compiled.
Architectural Position
Semantic DSL serves a role analogous to a high-level programming model, but grounded in semantic rather than syntactic constructs.
Its defining properties are:
Structure
Intent is expressed in a normalized semantic form, not free text.Capability Alignment
Every DSL term must correspond to an admissible capability in AgentIDL.Deterministic Lowerability
DSL structures compile into unique Semantic ISA graphs.Reconstructability
DSL terms act as anchors that permit reverse interpretation from execution traces.
Semantic DSL does not replace natural language.
It provides the first point of semantic commitment before compilation.
Scope
Semantic DSL concerns:
- admissible forms of structured intent,
- validation of semantic programs against declared capabilities,
- preservation of intent meaning through lowering into execution layers,
- providing the human-aligned interpretive basis for post-execution reconstruction.
It does not concern itself with:
- surface syntax preferences,
- optimization strategies,
- runtime execution mechanics,
- vendor- or model-specific prompt conventions.
Those concerns belong elsewhere in the semantic execution stack.
What This Work Is Not
Semantic DSL is not:
- a general-purpose programming language,
- a prompt template or instruction tuning format,
- a workflow engine or agent scripting scheme,
- a domain ontology or reasoning formalism.
Such systems may consume or produce Semantic DSL, but they do not define it.
Relationship to AgentIDL and Semantic ISA
AgentIDL defines the space of admissible capabilities.
Semantic DSL expresses intent within those boundaries.
Semantic ISA provides deterministic execution semantics for DSL terms.
Their relationships are directional:
AgentIDL constrains the DSL.
The DSL compiles into ISA.
ISA maps into execution substrates.
The layers are distinct and non-interchangeable.
Status
This concept is considered stable.
Ongoing work focuses on DSL schema refinement, capability alignment rules, and formal compilation mappings into Semantic ISA, rather than changes to the architectural boundary defined here.