Status Detection
CACD monitors your AI agent sessions and detects their current state. This lets you see at a glance which sessions need attention and which are still working.
Session States
Section titled “Session States”Idle (Green)
Section titled “Idle (Green)”The agent is ready and waiting for your input. It’s finished processing and is sitting at its prompt.
Busy (Yellow)
Section titled “Busy (Yellow)”The agent is actively working - generating code, thinking, or executing commands. Give it time to finish.
Waiting for Input (Red)
Section titled “Waiting for Input (Red)”The agent is asking you a question or needs confirmation before proceeding. This usually means it wants to run a command or make a change and is waiting for your approval.
Pending Auto-Approval
Section titled “Pending Auto-Approval”If auto-approval is enabled, CACD evaluates whether the pending action is safe to approve automatically. This state is temporary while the check runs.
How Detection Works
Section titled “How Detection Works”CACD analyzes the terminal output from each session to determine its state. Different agents display their prompts and status in different ways, so CACD has detection patterns for each supported agent:
- Claude Code - Detects Claude’s prompts and permission requests
- Gemini CLI - Matches Gemini’s specific output patterns
- Codex CLI - Recognizes Codex’s prompt style
- Cursor - Identifies Cursor Agent markers
- GitHub Copilot - Matches Copilot CLI patterns
- Cline - Detects Cline’s tool approval prompts
Visual Indicators
Section titled “Visual Indicators”The sidebar shows colored dots next to each session:
- 🟢 Green = Idle
- 🟡 Yellow = Busy
- 🔴 Red = Waiting
Project entries also show aggregate counts like [3/1/0] meaning 3 idle, 1 busy, 0 waiting.
Status Hooks
Section titled “Status Hooks”You can trigger scripts when session states change. For example, get a desktop notification when an agent finishes working or needs your input. See Status Hooks for setup instructions.
Limitations
Section titled “Limitations”State detection is based on pattern matching, so it’s not perfect. If an agent displays unexpected output or your terminal has unusual formatting, detection might be delayed or incorrect.
If you notice detection issues with a specific agent, the patterns may need adjustment for your setup.