ChainWeaver v0.0.1
ChainWeaver v0.0.1 — Initial Release
Deterministic orchestration layer for MCP-based agents. No LLM calls between steps — all data passing is structured and schema-validated via Pydantic.
Core Features
Tool— schema-validated callable wrapping Pydantic input/output modelsFlow/FlowStep— ordered, deterministic step sequences with input mapping (context keys + literal constants)FlowRegistry— in-memory registry with register, get, list, and intent-based matchingFlowExecutor— sequential runner with full error recording inExecutionResult- Typed exceptions —
ToolNotFoundError,FlowNotFoundError,FlowAlreadyExistsError,SchemaValidationError,InputMappingError,FlowExecutionError - Logging — library-safe
NullHandlerpattern; step-levellog_step_start/log_step_end/log_step_errorhelpers; DEBUG-level context key collision warnings - PEP 561 —
py.typedmarker included
Test Suite
32 tests covering: successful execution, error wrapping, schema validation, input mapping (including empty passthrough), empty flows, Tool.run() isolation, and missing tool/flow handling.
Requirements
- Python >= 3.10
- pydantic >= 2.0
What's Changed
- feat: ChainWeaver v0.1.0 — deterministic MCP tool-chain orchestration by @Copilot in #1
New Contributors
- @Copilot made their first contribution in #1
Full Changelog: https://github.com/dgenio/ChainWeaver/commits/v0.0.1