Crate gestura_core_pipeline 

Public pipeline types, persona, and reflection models for Gestura.

gestura-core-pipeline defines the stable data model shared across the agent execution stack: request/response types, persona defaults, compaction strategy, paused-execution state, and reflection/evaluation structures.

Design role

This crate intentionally contains the types and prompt assets of the pipeline rather than the full runtime implementation. The concrete pipeline orchestration lives in gestura-core, where it can coordinate tools, context, streaming, sessions, and provider selection.

In other words:

• this crate owns the pipeline vocabulary
• gestura-core owns the pipeline execution engine

High-signal exports

• AgentRequest, AgentResponse: the primary request/response types used by CLI, GUI, and tests
• RequestSource, RequestMetadata: request origin and routing metadata
• PipelineConfig: runtime configuration for pipeline execution
• CompactionStrategy: context-compaction policy shared with sessions
• PausedExecutionState: resumable execution state for confirmation flows
• default_system_prompt: the default runtime persona prompt
• reflection and reflection_eval: quality signals and evaluation helpers

Architecture boundary

Downstream code should usually import these types through gestura_core::pipeline::* so the facade remains the stable public entry point. Depending on this crate directly is most appropriate when working on pipeline data structures, persona content, or reflection logic in isolation.

Documentation direction

The goal is for cargo doc to surface pipeline concepts here, while more operational or end-user workflow documentation stays outside the API docs.

Re-exports

pub use persona::default_system_prompt;

pub use reflection::AgentReflection;

pub use reflection::QualitySignals;

pub use reflection::ReflectionConfig;

pub use reflection_eval::ReflectionEvalCase;

pub use reflection_eval::ReflectionEvalReport;

pub use reflection_eval::ReflectionEvalSummary;

pub use reflection_eval::ReflectionEvalToolOutcome;

pub use reflection_eval::ReflectionEvalToolResult;

pub use reflection_eval::ReflectionEvalTurn;

pub use reflection_eval::builtin_reflection_eval_cases;

pub use reflection_eval::evaluate_reflection_case;

pub use reflection_eval::evaluate_reflection_cases;

pub use types::AgentRequest;

pub use types::AgentResponse;

pub use types::CompactionStrategy;

pub use types::Message;

pub use types::PausedExecutionState;

pub use types::PipelineConfig;

pub use types::RequestMetadata;

pub use types::RequestSource;

pub use types::SessionLlmInfo;

pub use types::TokenLimitStatus;

pub use types::ToolCallRecord;

pub use types::ToolResult;

Modules

persona

Runtime agent persona + instruction hierarchy.

reflection

ERL-inspired experiential reflection types and pure helpers.

reflection_eval

Fixture-based reflection evaluation harness.

types

Pipeline types for unified LLM interaction

Structs

OutcomeSignal

Durable outcome observation attached to a reflection, task, or memory record.

Enums

OutcomeSignalKind

Durable outcome labels that corrective learning can attach to turns and tasks.