Opencode Config Advisor

Goal

To locate the active Opencode configuration file, read its current state, and propose valid JSON modifications to satisfy user requirements (e.g., switching models, increasing context limits, changing output format).

Usage Rules

1. Config Discovery

Always check for the configuration file in the following order of precedence:

1. •Project Local: ./opencode.json (or .opencode/config.json)
2. •Global (User): ~/.config/opencode/opencode.json
3. •Legacy: ~/.opencode.json

Command to use: ls -F ~/.config/opencode/opencode.json ./opencode.json 2>/dev/null

2. Analysis & Safety

• •Read First: Always run cat <path_to_found_config> before suggesting changes.
• •Validate JSON: Ensure any proposed changes result in valid JSON. Do not use trailing commas.
• •Backup: If executing a write operation, suggest backing up the original file first (cp config.json config.json.bak).

3. Common Modifications

• •Model Switching: Look for "model" or "default_model". Common values are gpt-4o, claude-3-5-sonnet, gemini-pro.
• •Context Window: Look for "max_tokens" or "context_limit".
• •API Keys: NEVER output full API keys in the chat. Use placeholders like "sk-..." when displaying file content.

Examples

Example 1: Switch Model

User: "Opencode is being too expensive, switch it to a cheaper model." Agent Execution:

1. •Locates ~/.config/opencode/opencode.json.
2. •Reads content.
3. •Proposes changing "model": "gpt-4" to "model": "gpt-4o-mini".

Example 2: Debugging

User: "Why is Opencode ignoring my ignores?" Agent Execution:

1. •Reads config to check the "ignore_patterns" or "gitignore" setting.
2. •Checks if .gitignore exists in the current directory.
3. •Suggests adding specific patterns to the config if the global setting is missing.