Prompt Quality

The agent continuously evaluates your prompt as you type — no button needed. Two things happen automatically:

1. Quality score — rendered as a thin gradient bar at the top of the composer (plus a passive score badge in the toolbar)
2. Agent understanding — a short preview of how the agent interpreted your prompt, shown directly above the composer input

Prompt Quality Bar

The quality bar is a 2px gradient strip at the top of the composer. It fills proportionally to a 1–100 score:

• Red (0–40): Vague, missing file paths or constraints
• Amber (40–70): Decent, could be more specific
• Lime (70–100): Well-scoped with clear inputs, behavior, and constraints

Hover the bar to see the score and detailed feedback.

The quality evaluation uses the current prompt text, the last assistant response, and the project's AGENTS.md (if present) as context.

Agent Understanding Preview

Once you stop typing, the composer shows a small, non-clickable "Agent understands:" card directly above the input. It restates — in the same language as your prompt — what the agent thinks you are asking for: the goal, key entities, and any constraints.

• It appears automatically a moment after you finish typing
• It updates when you keep typing, and disappears again while you type
• It is purely informational: there is no button to click, and clicking it does nothing
• The interpretation is generated server-side using the current prompt text, the last assistant response, the project's AGENTS.md, and memory context

Use the preview to catch misinterpretations before sending — if it doesn't match your intent, refine the prompt and it will update.

Prompt Principles

Based on what the agent uses for evaluation, good prompts are:

Next Steps

• Projects & Sessions
• Interval Scheduling