perspt-coding

The coding domain package built on perspt-sdk: it declares the coding residual schema, supplies the coding energy model (weights, rho_gate, correction budget), and maps dominant residuals into coding correction directions.

Modules

Module	Description
barrier	OperationalSafetyBarrier - versioned operational safety channels; correctness evidence stays in V, never duplicated into h
diag	The structured verifier plane: adapters preserve as much source structure as each tool provides
lang	Language adapters (Rust, Python, TypeScript) as verifier suites
prompts	The coding prompt section library; branch_correct renders a typed CorrectionPacket for the model
registry	CodingAdapterRegistry/LanguageId - open language-adapter registry keyed by id, never enum dispatch
runtime	Generic runtime smoke-probe scheme for exercising built artifacts
symbols	Symbol extraction for the goal-presence sensor

Key Types

CodingDomain - The AgentDomainPackage implementation:

pub struct CodingDomain;

impl AgentDomainPackage for CodingDomain {
    fn domain_id(&self) -> DomainId;                 // "coding"
    fn detect(&self, workspace: &WorkspaceSnapshot) -> DomainDetection;
    fn residual_schema(&self, scope: &DomainScope) -> ResidualSchema;
    fn energy_model(&self, scope: &DomainScope) -> EnergyModel;
    fn correction_directions(&self, residuals: &[ResidualEvent])
        -> Vec<CorrectionDirection>;
    // ... hard_gate_policy, verifier_suite, safety_barrier
}

LanguageId - Open, stable language identifier:

pub struct LanguageId(pub String);

pub struct CodingAdapterRegistry {
    adapters: BTreeMap<LanguageId, Box<dyn LanguageAdapter>>,
}

The coding domain operates on discrete verifier residuals (compiler, LSP, AST, tests), so its analytic constants remain NotClaimed; only the measured discrete gate and the spectral mu apply.