Expand description
Platonic Core: typed harness primitives for disciplined agent execution.
The crate intentionally starts as a small kernel, not an agent app. It models runs, context packs, tool calls, policy decisions, and event-log entries. The default stance is: every side effect is typed, policy-gated, and recorded.
Re-exports§
pub use context::ContextFragment;pub use context::ContextLane;pub use context::ContextPack;pub use error::Error;pub use event::HarnessEvent;pub use event::ModelUsage;pub use event::RecordedEvent;pub use ids::ActorId;pub use ids::AgentId;pub use ids::ArtifactId;pub use ids::ModelName;pub use ids::RunId;pub use ids::ToolCallId;pub use ids::ToolName;pub use ids::TurnId;pub use message::Message;pub use message::MessageRole;pub use policy::EffectClass;pub use policy::PolicyDecision;pub use projection::ReadbackEntry;pub use projection::RunReadback;pub use run::RunCommand;pub use run::RunPhase;pub use run::RunState;pub use tool::ResultVisibility;pub use tool::ToolCall;pub use tool::ToolProposal;pub use tool::ToolResult;
Modules§
- context
- Context assembly primitives with lane labels and budget validation.
- error
- Error types shared across Platonic Core modules.
- event
- Durable harness event ledger types.
- ids
- Compact identifier newtypes used across the harness event ledger.
- message
- Model-facing message primitives.
- policy
- Policy primitives for evaluating proposed side effects.
- projection
- Pure readback projections derived from recorded run events.
- run
- Pure run state machine.
- tool
- Tool-call and tool-result boundary types.