Advanced Workflows
Start simple. Add advanced workflows only after the team has validated the basics.
RouteLLM
RouteLLM can expose one OpenAI-compatible endpoint that routes actual requests between weak/routine and strong/heavy-lift models. It is optional and advanced; most tools should start with recommendation or static profile guidance.
Generated files:
.coding-scaffold/ROUTELLM.md.coding-scaffold/routellm.config.yaml
Use RouteLLM when endpoint-level model routing matters. For simple explainable recommendations, use
coding-scaffold tools select-model.
Context Compression
The built-in compressor writes token-saving sidecars for large knowledge bases without extra installation:
Caveman Compression is optional when you want to try the upstream engine explicitly:
Use either path after the team has reviewed the source notes. Compressed sidecars are agent input, not the canonical record.
Open Multi-Agent
Open Multi-Agent is an optional workflow backend for repeatable TypeScript automation:
Generated files:
.coding-scaffold/OPEN_MULTI_AGENT.md.coding-scaffold/open-multi-agent.team.jsonexamples/open-multi-agent/team-coding-workflow.ts
Recommended adoption path:
- Validate the workflow interactively in OpenCode.
- Capture it as a skill.
- Run the generated workflow in plan-only mode.
- Review permissions, traces, and verification signals.
- Automate only after the team trusts the process.