$ lexprog.com

// notes from an old coder -- php, databases, and the occasional rant

[October 24, 2026] OpenCode

Pin the model id: on an untrusted clone

Set an explicit model so a provider alias cannot rewrite the team overnight. Keep network and write on ask until you read skills and MCP.

────────────────────────────────────────────────────────

Pin the model id: on an untrusted clone

Tip: latest is not a contract

Set an explicit model so a provider alias cannot rewrite the team overnight. Keep network and write on ask until you read skills and MCP.

{ "model": "anthropic/claude-sonnet-4-6" }

Gotcha: Leaving model unset (on an untrusted clone)

Each developer then hits a different default and reviews become model-vs-model. A malicious skill is an install script; YOLO permissions are how it runs.

{ "provider": "anthropic" }

Source: https://opencode.ai/docs/providers

────────────────────────────────────────────────────────
<-- back to posts