s-ai

Coding agents (opencode, oh-my-pi, pi) sharing one canonical layer: skills, prompts, MCP servers, providers, themes. Each agent module translates the canonical options into its harness's native format - define once, land in every agent.

Options

• Agents: s-ai.agents.{opencode,omp,pi}.enable + per-agent toggles + extraSettings
• Skills: s-ai.plugins.<name> - builtin set (core, ponytail, claude-code, workmux) + your own
• MCP: s-ai.mcp.servers canonical set, merged with host-level programs.mcp.servers
• Providers: s-ai.providers.<name> (openai-compatible grids) + s-ai.defaultProvider
• Prompts: s-ai.prompts.{builtin, dirs} - numbered markdown → combined AGENTS.md
• Theme: s-ai.theme.palette - raw config.lib.stylix.colors or plain base16
• Escapes: extraSettings / extraOmoConfig deep-merge last on every agent

Declarative / imperative posture

Read-only data (skills, mcp.json, models, themes, AGENTS.md) renders as store symlinks - agents read them, they cannot rot. Agent-editable settings (omp's config.yml, OMO's omo.jsonc) render as writable copies: runtime changes work and survive until the next switch, which re-asserts the declared values.