Crate mdx_rust_core

Expand description

Core primitives for the mdx-rust CLI.

mdx-rust-core contains the optimizer, hardening engine, safety pipeline, registry, evaluation, ledger, and audit primitives used by the mdx-rust binary.

§Stability contract

The supported product surface for 0.3.x is the mdx-rust CLI. This crate is published so the CLI can be installed from crates.io and so advanced users can inspect the internal data structures, but the library API is not yet stable. Public items may change before 1.0.

The intentionally documented facade is the set of pub use exports below. Module paths are left public for the CLI and tests, but most modules are hidden from rustdoc because they are implementation detail for now.

Structs§

AcceptedEditSummary: Optimizer entrypoint and run records. These are unstable before 1.0. Summary of the accepted source edit.
AgentRunResult: Agent runner result and trace events. These are unstable before 1.0. The result of running an agent on a single input, including traces.
AuditFinding: Deterministic static audit reports.
AuditPacket: Optimizer entrypoint and run records. These are unstable before 1.0. Versioned machine-readable evidence for one accepted optimizer change.
AuditProvenance: Optimizer entrypoint and run records. These are unstable before 1.0. Provenance fields that identify what inputs and tools produced the change.
Candidate: Optimizer entrypoint and run records. These are unstable before 1.0. A proposed improvement generated during an optimization iteration.
CandidateExecutionConfig: Candidate safety pipeline. Direct use is unstable before 1.0.
CandidateExecutionContext: Candidate safety pipeline. Direct use is unstable before 1.0.
CandidateExecutionOutcome: Candidate safety pipeline. Direct use is unstable before 1.0.
Config: Configuration loading and defaults used by the CLI. Root configuration for an mdx-rust project
DatasetSplit: Experiment budget and ledger records. These are unstable before 1.0.
EvaluationDataset: Dataset and scorer metadata used by optimizer reports.
EvaluationSample: Dataset and scorer metadata used by optimizer reports.
ExperimentLedger: Experiment budget and ledger records. These are unstable before 1.0.
FailureSignal: Trace diagnosis records.
HardeningChangeSummary: Scoped Rust hardening engine for ordinary Rust modules. Unstable before 1.0.
HardeningConfig: Scoped Rust hardening engine for ordinary Rust modules. Unstable before 1.0.
HardeningOutcome: Scoped Rust hardening engine for ordinary Rust modules. Unstable before 1.0.
HardeningPolicyRecord: Scoped Rust hardening engine for ordinary Rust modules. Unstable before 1.0.
HardeningRun: Scoped Rust hardening engine for ordinary Rust modules. Unstable before 1.0.
HookContext: Built-in lifecycle hook primitives. These are unstable before 1.0.
HookDecision: Built-in lifecycle hook primitives. These are unstable before 1.0.
HookPolicy: Built-in lifecycle hook primitives. These are unstable before 1.0.
ModelProvenance: Optimizer entrypoint and run records. These are unstable before 1.0.
OptimizationRun: Optimizer entrypoint and run records. These are unstable before 1.0. A single optimization experiment / iteration result.
OptimizeConfig: Optimizer entrypoint and run records. These are unstable before 1.0. Configuration for a single optimization run.
PromptVariantRecord: Experiment budget and ledger records. These are unstable before 1.0.
RegisteredAgent: Agent registry types used by CLI commands. Represents a registered agent
Registry: Agent registry types used by CLI commands.
SafetyRejection: Candidate safety pipeline. Direct use is unstable before 1.0.
ScoreProvenance: Optimizer entrypoint and run records. These are unstable before 1.0. Score evidence for the accepted change.
ScorerMetadata: Dataset and scorer metadata used by optimizer reports.
SecurityAuditReport: Deterministic static audit reports.
TraceDiagnosis: Trace diagnosis records.
TraceEvent: Agent runner result and trace events. These are unstable before 1.0. A single trace event captured during an agent run. Made first-class for trace-to-patch optimization.
WorkspaceSummary: Scoped Rust hardening engine for ordinary Rust modules. Unstable before 1.0.

Enums§

AgentContract: Agent registry types used by CLI commands. The kind of agent contract detected
AuditSeverity: Deterministic static audit reports.
EditStrategy: Optimizer entrypoint and run records. These are unstable before 1.0.
FailureKind: Trace diagnosis records.
HardeningMode: Scoped Rust hardening engine for ordinary Rust modules. Unstable before 1.0.
HardeningStatus: Scoped Rust hardening engine for ordinary Rust modules. Unstable before 1.0.
HookAction: Built-in lifecycle hook primitives. These are unstable before 1.0.
HookStage: Built-in lifecycle hook primitives. These are unstable before 1.0.
OptimizationBudget: Experiment budget and ledger records. These are unstable before 1.0.
SafetyRejectionKind: Candidate safety pipeline. Direct use is unstable before 1.0.

Functions§

audit_agent: Deterministic static audit reports.
diagnose_run: Trace diagnosis records.
evaluate_builtin_hook: Built-in lifecycle hook primitives. These are unstable before 1.0.
execute_candidate_edit: Candidate safety pipeline. Direct use is unstable before 1.0.
mechanical_score: Optimizer entrypoint and run records. These are unstable before 1.0. Very rough mechanical scorer for the example agent. Gives higher score if the output is not the echo fallback.
run_agent: Agent runner result and trace events. These are unstable before 1.0. Run a registered agent with the given input. Currently supports Process contracts (spawns the agent binary and pipes JSON). NativeRust support will be added when we generate harnesses.
run_hardening: Scoped Rust hardening engine for ordinary Rust modules. Unstable before 1.0.
run_optimization: Optimizer entrypoint and run records. These are unstable before 1.0. Placeholder for the full optimization engine. In a real implementation this would orchestrate:
split_dataset: Experiment budget and ledger records. These are unstable before 1.0.