Developer speed with program-controlled execution.
KyzoCode is an opinionated OpenCode distribution for teams that want capable coding agents without making model behavior their control plane. It combines practical permissions, typed development standards, deterministic gates, and inspectable shared project memory.
From the repository root:
npm install --prefix .opencode
opencodeOpenCode loads the project permission profile, local skills, agent instructions, and shared wiki protocol automatically.
| I want to... | Start here |
|---|---|
| Build an OpenCode plugin | OpenCode Plugin Architecture |
| Write or review Python | Python Development |
| Understand the project | KyzoCode Knowledge |
| Review the trust boundary | Security Baseline |
- Safe development defaults: routine workspace work stays fast; shell, network, and external access cross an approval boundary.
- Shared project memory: humans and agents use the same visible, version-controlled knowledge instead of opaque model memory.
- Typed construction standards: domain meaning lives in structures that prevent invalid states from moving forward.
- Native OpenCode extension: configuration and plugins stay inside OpenCode's supported surfaces rather than patching upstream.
- Deterministic knowledge gates: ontology, citations, links, and reachability are checked mechanically.
KyzoCode treats models as untrusted components. Its permission profile is a practical development baseline, not a sandbox; execute untrusted repository code inside real process, filesystem, and network isolation. The canonical policy is the Security Baseline.
Validate the shared project knowledge:
npm run verify --prefix .opencode| Path | Purpose |
|---|---|
.opencode/ |
OpenCode configuration and runtime dependencies |
.agents/skills/ |
Reusable development and agent capabilities |
wiki/ |
Project architecture, policy, and shared knowledge |
AGENTS.md |
Always-on behavioral contract for agents |
The repository currently ships the OpenCode foundation: project configuration, the security baseline, Project Knowledge lifecycle, Python development standard, local-plugin architecture, and a Project Knowledge runtime with trust-aware retrieval and permission-gated reconciliation. Additional product plugins are the next layer to be implemented.