Prompt Explorer

The Prompt Explorer lets you browse every prompt you sent to an AI agent and see exactly which commands it triggered — giving you a complete audit trail from intent to execution.

Usage

suv agent prompts

This opens an interactive TUI showing prompts grouped by session. You can also access the Prompt Explorer by pressing Ctrl+P while in the Agent Dashboard.

Flags

Flag	Description
`--after <TIME>`	Show prompts after a specific time (default: 7 days ago). Accepts natural language like `"2 days ago"` or dates like `"2025-01-15"`.
`--executor <NAME>`	Filter to prompts from a specific agent (e.g., `claude-code`, `cursor`)
`--here`	Only show prompts from sessions in the current directory

Live Search and Filters

Type directly into the always-on search box to filter by prompt text. Plain letters and digits are search text; no slash is needed to start searching.

Key	Action in the prompt list
Ctrl+P	Cycle time period: Today, 7 days, 30 days, All time
Ctrl+A	Cycle executor filter
Ctrl+S	Open the selected prompt's session timeline
Up / Down	Navigate prompts
Left / Right	Change page
Enter	Open the prompt's commands and details
Esc	Quit the list

Inside prompt details, Tab toggles command detail, Ctrl+Y copies the selected command, and Esc, q, or Backspace returns to the list. Session panels now show the full session ID.

What the Prompt Explorer Shows

Prompt List

The left side of the TUI displays a chronological list of prompts, grouped by session. Each entry shows:

Prompt text — a truncated preview of what you sent to the agent
Commands triggered — how many commands the agent ran in response to this prompt
Executor — which agent received the prompt
Timestamp — when the prompt was submitted

Detail Preview

The right side shows a detailed preview for the selected prompt, including:

Captured prompt text — subject to redaction and the configured capture-length limit
Session — the session ID this prompt belongs to
Executor — the agent that processed it
Working directory — the path where the session was active
Timestamps — when the prompt was submitted and when the agent finished responding
Success/fail stats — how many of the triggered commands succeeded versus failed
Command list — every command the agent ran in response, with exit codes and risk levels

Prompt Capture Availability

Prompt capture requires hooks in the agent's configuration. The following agents support prompt capture:

Agent	Prompt Capture	Setup
Claude Code	Yes — via UserPromptSubmit hook	`suv init claude-code`
Cursor	Yes — via beforeSubmitPrompt hook	`suv init cursor`
OpenCode	Yes — via suvadu.js plugin	`suv init opencode`
Codex	Yes — via UserPromptSubmit and PostToolUse hooks	`suv init codex`
pi.dev	Yes — via extension	`suv init pi`
Antigravity	No — no hooks system available	`suv init antigravity`

For agents without prompt capture (Antigravity and agents using only shell detection, such as Aider), you can still see commands in the Agent Dashboard and Agent Stats — but the Prompt Explorer will not have prompt entries for them.

Only prompts associated with recorded commands appear. For Codex, restart after setup and trust the hooks; capture requires a version with UserPromptSubmit and PostToolUse hook support. See Agent Setup for capture limits.

Examples

Browse recent Claude Code prompts

suv agent prompts --executor claude-code --after "2 days ago"

Shows all prompts sent to Claude Code in the last 2 days, along with the commands each prompt triggered.

Explore prompts in the current project

suv agent prompts --here

Filters to prompts from sessions that were active in the current working directory.

Default view (last 7 days)

suv agent prompts

Shows all prompts from all agents over the past 7 days.

Tip: The Prompt Explorer is especially powerful for debugging agent behavior. If an agent produced unexpected results, find the prompt that triggered the issue and review every command it ran in response — including which ones failed and their risk levels.