One configuration file cannot be used by version 1 and version 2 #41081

Description

Version 1 and version 2 can be installed at the same time. The two versions read the same configuration file. But one file cannot satisfy both versions.

Version 1 stops with an error when the file contains the version 2 keys. The keys are providers and agents. Version 1 shows this message:

Configuration is invalid at <config-dir>/opencode.json
↳ Unrecognized keys: providers, agents

The migration document is at https://opencode.ai/v2/docs/migrate-v1

The document says that version 1 fields and version 2 fields can stay together in one file. The document says: "Supported V1 and native V2 fields may coexist at the top level". The document also says: "when both forms set the same canonical value, a valid native V2 value takes precedence". Thus the user can write the provider with the version 1 fields.

Version 2 does not obey this rule for a custom provider package.

I declare a provider with the version 1 fields. I use the provider key. I use the npm field. Version 1 reads the file correctly. But version 2 shows no models for this provider. Version 2 shows no error message.

I declare the same provider with the version 2 fields. I use the providers key. I use the package field. Version 2 shows all 32 models. But version 1 stops with the error above.

Thus neither format is correct for both versions.

The fault does not occur for all packages. A provider with the @ai-sdk/openai-compatible package loads correctly with the version 1 fields. The fault occurs with the package commandcode-go-opencode-provider.

Environment

• opencode version 2: 0.0.0-next-16974 (next channel)
• opencode version 1: 1.18.15
• Operating system: macOS 25.5.0
• Architecture: arm64

How to repeat the fault

1. Install the package commandcode-go-opencode-provider from npm.
2. Put configuration A in the global configuration file.
3. Run this command with version 2: opencode models
4. Count the lines that start with commandcode/.
5. Run the same command with version 1.
6. Put configuration B in the same file.
7. Do step 3 to step 5 again.

Configuration A uses the version 1 fields:

{
  "provider": {
    "commandcode": {
      "npm": "commandcode-go-opencode-provider",
      "name": "Command Code",
      "env": ["COMMANDCODE_API_KEY"],
      "options": { "baseURL": "https://api.commandcode.ai" },
      "models": {
        "deepseek-v4-flash": {
          "id": "deepseek/deepseek-v4-flash",
          "name": "DeepSeek V4 Flash",
          "tool_call": true
        }
      }
    }
  }
}

Configuration B uses the version 2 fields:

{
  "providers": {
    "commandcode": {
      "package": "aisdk:commandcode-go-opencode-provider",
      "name": "Command Code",
      "settings": { "baseURL": "https://api.commandcode.ai" },
      "models": {
        "deepseek-v4-flash": {
          "modelID": "deepseek/deepseek-v4-flash",
          "name": "DeepSeek V4 Flash",
          "capabilities": { "tools": true }
        }
      }
    }
  }
}

Result

Configuration A: version 2 shows 0 models for the provider. There is no error message. Version 1 shows all models.

Configuration B: version 2 shows all models. Version 1 stops with the error message.

I did more tests with configuration A. The result stays the same for each test:

• The package field and the npm field together in the provider key.
• Only the package field in the provider key.
• Model entries with only version 1 fields. I removed modelID, capabilities and cost.cache.

A second file does not correct the problem. Both versions read the file opencode.jsonc. Version 1 also stops with an error for that file.

Expected result

Version 2 must load the provider when the configuration uses the version 1 fields. Then one file can satisfy both versions.

If version 2 cannot load the provider, then version 2 must show an error message.