smithery/victorstein

agent-monitor

Real-time monitoring of Claude Code sub-agent activity in a tabbed terminal interface. Auto-spawns as soon as any sub-agent starts running.

Installation

$ npx skills add smithery/victorstein --skill agent-monitor

Similar popular skills

Related neighbors and high-traction skills in the same topics — useful to compare before installing.

More details

Agent compatibility

Declared targets from SKILL.md / docs. Unmarked agents are not listed — the skill may still install via the CLI.

Claude Code Declared
Cursor Not declared
Codex Not declared
GitHub Copilot Not declared
Windsurf Not declared
Gemini CLI Not declared
Cline Not declared
OpenCode Not declared

Skill metadata

Parsed from SKILL.md frontmatter.

Declared agents claude-code

Package contents

Files included with this skill beyond the listing page.

  • skill md SKILL.md 3,423 B
  • docs SUMMARY.md 161 B

History

  1. First recorded snapshot · 0 installs

SKILL.md

Agent Monitor Canvas

Displays real-time sub-agent activity with:

  • Tabbed interface (one tab per agent)
  • Scrollable output streams
  • Tool call display with results
  • Git diff visualization

Auto-Spawn Behavior

The monitor automatically spawns as soon as Claude Code dispatches any sub-agent using the Task tool. It uses Claude Code hooks to detect agent lifecycle events:

  • PreToolUse hook: Detects when Task tool is invoked
  • SubagentStop hook: Tracks when agents complete

Manual Usage

# Show agent monitor in current terminal
bun run src/cli.ts show agent-monitor --scenario monitor

# Spawn agent monitor in a new WezTerm/tmux pane
bun run src/cli.ts spawn agent-monitor --scenario monitor

# With initial configuration
bun run src/cli.ts spawn agent-monitor --scenario monitor --config '{
  "parentSessionId": "session-123",
  "agents": {}
}'

Keyboard Controls

Key Action
Tab Switch to next agent tab
1-9 Jump to agent by number
↑/↓ Scroll content up/down
PgUp/PgDn Fast scroll (10 lines)
g Jump to top
G Jump to bottom
q/Esc Close monitor

Status Indicators

Each tab shows a status indicator:

  • 🟡 Yellow: Agent is running
  • 🟢 Green: Agent completed successfully
  • 🔴 Red: Agent encountered an error

Content Display

For each agent, the monitor displays:

  1. Output Stream: Text output from the agent's thinking and responses
  2. Tool Calls: Each tool invocation with:

- Tool name (highlighted in cyan) - Input parameters (truncated preview) - Status icon (✓ success, ✗ error, ⋯ pending) - Result (if available)

  1. Git Diffs: File changes with syntax highlighting:

- Green for additions (+) - Red for deletions (-) - Cyan for hunk headers (@@)

Configuration

interface AgentMonitorConfig {
  parentSessionId: string;
  agents: Record<string, AgentState>;
}

interface AgentState {
  id: string;                    // Unique agent identifier
  description: string;           // Task description
  status: "running" | "completed" | "error";
  startTime: number;             // Unix timestamp
  endTime: number | null;
  toolCalls: ToolCallRecord[];
  gitDiffs: DiffRecord[];
  output: string[];
}

Hook Configuration

The plugin uses the following hooks (defined in hooks/hooks.json):

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Task",
        "hooks": [
          {
            "type": "command",
            "command": "bun \"${CLAUDE_PLUGIN_ROOT}/src/hooks/pre-tool-use.ts\""
          }
        ]
      }
    ],
    "SubagentStop": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "bun \"${CLAUDE_PLUGIN_ROOT}/src/hooks/subagent-stop.ts\""
          }
        ]
      }
    ]
  }
}

Tracking File

Agent state is stored in /tmp/claude-agents-{session}.json and includes:

  • Parent session ID
  • Canvas pane ID (when spawned)
  • Socket path for IPC
  • All tracked agents with their current state

Example Prompts

  • "Run these three tasks in parallel and show me the monitor"
  • "I want to see what the sub-agents are doing"
  • "Open the agent activity viewer"