akillness/synapse-skill

Overview

This skill is a multi-AI agent orchestration system that coordinates Claude, Gemini, and Codex into role-based workflows for planning, analysis, code generation, review, and execution. It supports pipeline, parallel specialist, swarm, and full end-to-end workflows and exposes local HTTP endpoints for integration. Use it to automate complex software tasks where multiple specialized AI agents must collaborate.

How this skill works

Synapse assigns roles (planner, analyst, coder, reviewer, executor) to configured models and routes requests to service endpoints (claude, gemini, codex) via a local gateway. You invoke endpoints for plan, code, analyze, review, execute, or call the /api/v1/workflow endpoint to run pipeline, parallel, or swarm patterns. The system aggregates results, synthesizes outputs, and can run sandboxed commands through the Codex executor for builds and tests.

When to use it

• Breaking down complex projects into multi-step plans and implementations
• Generating production code from specifications and running automated tests
• Reviewing large codebases or performing deep security/performance audits
• Coordinating parallel reviews or multi-stage pipelines (plan → code → test → review)
• Automating end-to-end workflows that require execution and verification

Best practices

• Ensure Docker and Synapse services are running before invoking endpoints (health check first).
• Select role-specific models by task: use Claude for planning/codegen, Gemini for large-context analysis/review, Codex for execution.
• Use the workflow endpoint for full automations and explicit endpoints for fine-grained control in pipelines.
• Limit local exposure: run Synapse in a controlled environment and sandbox execution commands.
• Provide concise constraints and expected outputs to the planner to reduce iteration cycles.

Example use cases

• Pipeline: research requirements with Gemini, plan and generate code with Claude, execute tests with Codex, then review with Gemini.
• Parallel specialists: spawn multiple Gemini reviewers (security, performance, simplicity) and aggregate findings into a single report.
• Swarm: create a task pool for many small review or lint jobs and let workers (Claude/Gemini/Codex) claim tasks concurrently.
• End-to-end: POST to /api/v1/workflow to build and validate a user authentication module with specified constraints and models.
• Automated CI: run Codex to execute build and test commands, then route failures to Gemini for root-cause analysis.