dmae97/open-multi-agent-kit — Evidence-gated runner for Codex, Claude Code, OpenCode, and local coding agents.

Description and use

**Brief Summary**
dmae97/open‑multi‑agent‑kit is an evidence‑gated runner that orchestrates Codex, Claude Code, OpenCode and other local coding agents into scoped DAG lanes, producing replayable artifacts for each step. By turning isolated prompts and tools into repeatable, memory‑aware workflows, it lets teams coordinate multi‑agent pipelines, add tool‑use stages, and standardize agent state across runs.

**Value**

• **Deterministic orchestration** – tasks are routed through a directed‑acyclic‑graph, so the order of agent calls and tool invocations is explicit and reproducible.
• **Replayable artifacts** – every lane’s inputs, outputs, and evidence are persisted, enabling debugging, audit trails, and fine‑grained performance analysis.
• **Unified memory model** – the kit provides a simple abstraction for persisting and retrieving agent “memory,” making it easy to build stateful, long‑running workflows without custom glue code.
• **Multi‑model flexibility** – supports both hosted LLMs (Codex, Claude) and self‑hosted coding agents, allowing teams to pick the best model for cost, latency, or security.

**Practical Adoption Path**

1. **Prototype** – Install the TypeScript SDK/CLI, point it at your existing coding agents, and define a minimal DAG (e.g., a lint → generate → test lane). The kit’s API surface is straightforward, with clear run, addLane, and replay methods.
2. **Integrate tooling** – Replace ad‑hoc script calls with kit‑managed lanes; plug in your CI/CD, issue‑tracker, or IDE extensions via the provided hooks.
3. **Standardize memory** – Migrate any custom context‑passing logic to the kit’s built‑in memory store, gaining consistent state handling across agents.
4. **Scale** – Expand the DAG to include parallel lanes (e.g., multiple model candidates) and use the evidence‑gating feature to automatically select the best result based on test outcomes.
5. **Productionize** – Deploy the kit as a containerized service behind your internal API gateway; monitor lane health via the built‑in telemetry and use the replay feature for post‑mortem analysis.

**Production Readiness**

• **Activity & community**: 84 ★, 12 forks, recent commits (as of 2026‑06‑24), and a growing set of 20 topics indicate an active, engaged community.
• **Maturity**: The core orchestration, evidence‑gating, and replay mechanisms are stable and have been used in pilot projects, suggesting a low risk of breaking changes.
• **Integration ease**: Exposes a clean API/SDK/CLI, language metadata, and clear documentation, making it straightforward to embed in existing TypeScript/Node.js stacks.
• **Risks**: Licensing and security posture still need a final audit, and long‑term maintainer commitment should be confirmed before a full production rollout.

Overall, the kit is a strong OSS candidate for teams looking to move from one‑off prompt scripts to robust, auditable multi‑agent pipelines.

Problem

dmae97/open-multi-agent-kit helps turn isolated prompts and tools into repeatable agent workflows.

Outcome

High for an OSS candidate: recent activity, adoption, and ecosystem signals are strong enough for a serious pilot.

Workflow

• coordinate multi-agent workflows
• add tool-use pipelines
• standardize agent memory

Evidence

Evidence-gated runner for Codex, Claude Code, OpenCode, and local coding agents. Routes tasks into scoped DAG lanes with replayable artifacts.

Source