shinpr/sub-agents-skills

Sub-agents-skills is a Python library that enables cross-LLM sub-agent orchestration by routing tasks to multiple code-generation models from a single compatible tool.

The problem it solves is the fragmentation of specialized code-generation capabilities across different LLM providers. Rather than being locked into one model's strengths, developers need a way to dispatch specific tasks to the best-suited agent for each job. The tool works by exposing sub-agent routing as an Agent Skills interface, allowing any compatible orchestration framework to delegate work to Codex, Claude Code, Grok, GLM, Kimi, Cursor, Gemini, OpenCode, or Command Code based on task requirements.

This approach suits projects that already use an agent framework supporting the Agent Skills pattern and need flexibility across multiple code-generation models. It is particularly valuable when different models excel at different coding tasks—some may be stronger at certain languages, architectural patterns, or problem domains. Teams evaluating adoption should note that the tool requires an existing compatible agent orchestration system; it is not a standalone agent framework but rather an extension that adds multi-model routing capability to one.

The project shows active development with regular commits addressing both new model integrations and refinements to the routing logic. Work spans across multiple sub-agent implementations, indicating ongoing effort to maintain compatibility as new code-generation models emerge. The codebase demonstrates attention to supporting a growing roster of LLM providers, with each integration receiving dedicated implementation work.