Context Seeding

Context seeding loads relevant information into the conversation before making your actual request. The AI gathers context first, then uses it to inform its response.

Examples

• Web search: “Search for the latest React 19 features”
• File reads: “Read the authentication module and related tests”
• Git history: “Check the recent commits touching this file”
• Slash commands: Triggering /docs or custom commands that inject context
• Skills: Having the AI load project-specific guidance files

Reducing Noise

One technique: after the seeding prompt, instruct the AI to respond with just “ready” or “done.” This prevents lengthy summaries from cluttering the context window. The information is loaded—no need for a recap.

Read all files in src/auth/ and say "ready" when done.

The AI processes the files, outputs “ready,” and you follow up with your actual question or task.