Prompt Patterns

eye works best when the prompt tells the agent how to navigate instead of asking for a full-repository summary.

Good patterns

• Use eye to inspect this repository before answering. Start with get_project_structure at depth 2, then read the relevant files before you summarize the architecture.
• Use eye to inspect this repository before answering. Resolve createProgram with query_symbol from an anchor or symbol name, then follow references using symbolId and summarize the main call sites.
• Use eye only inside packages/next/src/server. Refresh the index for that scope, then use query_symbol for definition and references of loadConfig.
• Read the source around django/core/handlers/wsgi.py before explaining the request path. Do not answer from memory.

Prompts to avoid

• Explain this repo.
• Find where this is used. without a symbol, file, or scope.
• Read everything and summarize. on a large repository.

Those prompts force broad scans and make it harder to pick the right tool sequence.

Recommended flow

1. get_project_structure
2. read_source_range
3. query_symbol with action: "definition"
4. query_symbol with action: "references" or action: "context"
5. get_index_status or refresh_index only when needed

Next step: core tool reference.