List view
Validate ChainWeaver's positioning against the ecosystem and verify all v1.0 release criteria are met before cutting the stable release tag. Scope: Ecosystem research (LangGraph, LlamaIndex Workflows, OpenAI Agents SDK, MCP community), v1.0 release criteria checklist, release documentation. Dependency notes: All prior milestones must be substantially complete. This is the capstone. Done when: Ecosystem research is documented with differentiation statement, all v1.0 criteria pass, docs/release-criteria-v1.md is committed, v1.0.0 tag is cut.
No due date•0/1 issues closedBuild the CLI tooling for flow inspection, validation, and visualization, and prove ChainWeaver's performance claims with benchmarks and an optional offline LLM-assisted flow compiler. Scope: CLI entry point (chainweaver command), inspect, validate, check, viz subcommands, benchmark suite (naive vs compiled), offline LLM-assisted flow compiler (build-time only, never runtime). Dependency notes: #45 (validate CLI) depends on #14 (serialization, M3). #46 (viz CLI) depends on #44 (CLI entry) and #79 (viz API, M4). #28 (LLM compiler) and #29 (benchmarks) are independently buildable. Done when: chainweaver inspect/validate/check/viz commands work, benchmark suite demonstrates N× speedup with LLM-calls-avoided metrics, LLM compiler proposes flow YAML files for human review.
No due date•0/4 issues closedBroaden the ecosystem with trace replay, VirtualTool adapter (flow-as-tool for tool-space reduction), export adapters for OpenAI/Anthropic, logging redaction, web playground, and LangChain/LlamaIndex bridge adapters. Scope: Replay mode from recorded traces, Flow-to-VirtualTool adapter, OpenAI/Anthropic/callable export schemas, logging redaction policy, interactive web playground (Streamlit or Pyodide), bidirectional LangChain + LlamaIndex adapters. Dependency notes: #21 (replay) depends on #20 (structured trace, M5). #24 (VirtualTool) builds on #75 (composition, M3). #25 (export) builds on #24. #81 (playground) benefits from all features. #82 (framework bridges) is independent. Done when: Flows replay from traces with diff detection, flows expose as virtual tools reducing tool-space, flows export as OpenAI/Anthropic schemas, logging redacts sensitive keys, playground is deployable, LangChain/LlamaIndex tools convert bidirectionally.
No due date•0/6 issues closedAdd schema fingerprinting and drift detection, flow status lifecycle, determinism/safety contracts, structured execution traces with trace IDs, multi-OS CI, and versioning policy documentation. Scope: Schema fingerprinting (SHA-256 of JSON schemas), drift detection on re-registration, FlowStatus enum (ACTIVE/NEEDS_REVIEW/DISABLED), determinism + safety contracts, structured ExecutionTrace with UUIDs + timing, multi-OS CI matrix, CHANGELOG + versioning policy. Dependency notes: #48 (fingerprinting) + #49 (FlowStatus) → #50 (drift detection). #20 (structured trace) builds on executor from M2. #34 and #35 are independent infrastructure. Done when: Schema changes are detected and flows flagged for review, FlowStatus guards execution, traces have unique IDs + per-step timing + JSON serialization, CI runs on Windows/macOS/Linux, CHANGELOG exists with semver policy.
No due date•0/7 issues closedAdd async execution support for I/O-bound MCP tools, persistent flow storage, flow/tool versioning, flow visualization API (Mermaid + ASCII), and the runtime chain observer with auto-flow suggestion. Scope: Async execute_flow_async() with mixed sync/async chains, persistent registry (file-based store), flow/tool schema versioning, to_mermaid() / to_ascii() visualization, runtime chain observer. Dependency notes: #16 (file store) depends on #14 (serialization, M3). #78 (chain observer) builds on #77 (chain analyzer, M3). #80 (async) is prerequisite for real async MCP integration. Done when: Async tool chains execute, flows persist to disk and survive restart, flows/tools carry semver versions, flows render as Mermaid/ASCII diagrams, chain observer detects patterns and suggests flows.
No due date•0/5 issues closedAdd flow composition (flows-as-steps), error recovery with retry policies, flow serialization, cost-savings tracking, and the observation-to-governance pipeline (capture runtime chains → score determinism → propose → approve). Scope: Flow composition (sub-flows), retry policies with backoff, YAML/JSON serialization, "Cost Avoided" calculator, runtime trace capture, determinism scoring, governance workflow, offline schema-based chain analysis. Dependency notes: #75 (composition) and #77 (chain analyzer) benefit from #71 (compile-time validation, M2). #76 (retry) complements #43 (guardrails, M2). Internal chain: #11 → #12 → #13 (observation pipeline). Done when: Flows compose as sub-steps with cycle detection, retries work with exponential backoff, flows round-trip through YAML/JSON, CostReport quantifies LLM calls avoided, tool chains are observed/scored/governed with approval gates.
No due date•0/7 issues closedExtend the execution engine beyond linear flows with DAG execution, conditional branching, checkpoints, compile-time schema validation, and bidirectional MCP protocol support — the features that make ChainWeaver a real orchestration layer. Scope: DAG flow model + topological execution, conditional branching, execution checkpoints, compile-time schema chain validation, MCP Tool Adapter (consume MCP tools), MCP Flow Server (expose flows as MCP tools), dry-run/explain mode, tool execution guardrails (timeout + output size). Dependency notes: Requires M1 testing infra to be effective. Internal chain: #10 (DAG) → #9 (branching) → #8 (checkpoints). MCP adapter (#70) and server (#72) are independent of each other. Done when: DAG flows execute with topological ordering, branches route correctly, schemas validate at compile time before execution, MCP tools wrap into ChainWeaver and flows expose as MCP tools, guardrails enforce timeout/size limits, explain_flow() previews execution without running tools.
No due date•0/8 issues closedEstablish project infrastructure, documentation standards, CI quality gates, and ergonomic APIs that enable efficient contribution and fast first-experience adoption. This is the base layer — nothing else ships without a solid foundation. Scope: Dev docs (AGENTS.md, CLAUDE.md, copilot-instructions.md, scoped instructions), CI improvements (mypy, pytest-cov), DX APIs (@tool decorator, FlowBuilder), README rewrite, real-world examples, bug fixes (license mismatch), contribution templates. Dependency notes: No dependencies. All other milestones benefit from the infra established here. Done when: CI enforces type checking + coverage, all agent instruction files exist, @tool and FlowBuilder are functional, README hero section fits in one viewport with 3+ runnable examples, license references are consistent.
No due date•6/12 issues closed