AI Agent Features

Conductore has deep integration with Claude Code, opencode, cursor, and other CLI-based AI agents. These features run passively in the background — you don't need to configure anything for basic operation.

Rate limit auto-resume

When Claude Code hits an API rate limit, it prints a message like:

Claude AI rate limit exceeded. Please wait 5 minutes before retrying.

Conductore detects this pattern, parses the reset timestamp, and schedules a continue command to run automatically at the exact moment the limit resets.

What happens:

1. Rate limit message detected in terminal output
2. Conductore parses the reset time and timezone
3. A countdown timer appears in the sidebar for that terminal
4. At the reset time, Conductore sends \ncontinue\n to the terminal
5. The session resumes without any manual intervention

Configuration: Enable/disable in Settings → Terminal → Terminal Automation → Auto-resume on rate limit.

Idle & stuck agent detection

Conductore monitors terminal output and triggers alerts when AI agents go silent.

Idle detection

After a configurable period of no output (default: 60 seconds), the terminal is marked idle:

• Amber pulsing dot in the sidebar
• Amber outline flash on the pane
• Optional sound alert (two short beeps)
• Optional toast notification

Stuck detection

After a longer period of silence (default: 120 seconds), the terminal is marked stuck:

• Red pulsing dot in the sidebar
• Red outline flash on the pane
• Different sound (one lower tone)
• Toast: "Agent stuck: [terminal name]"

Clearing alerts: Any new output from the terminal automatically clears the alert. You can also click the alert indicator in the sidebar to clear it manually.

Configuration: Settings → Terminal → Agent Alert System

• Toggle idle/stuck detection independently
• Set timeout durations (5–600 seconds)
• Configure notification types (pane flash, sidebar flash, sound, toast)

Session ID capture

Claude Code and opencode print a session ID when a conversation starts:

Session started: abc123de-f456-7890-abcd-ef1234567890

Conductore detects this pattern, captures the ID, and if the terminal is linked to a task, adds a session entry to that task's YAML frontmatter automatically.

This creates a traceable link from every task → every agent session that worked on it.

Configuration: Settings → Terminal → Terminal Automation → Capture session IDs.

Plan detection

When AI agents generate plans (structured output with numbered steps or markdown headers), Conductore can detect and capture them.

Detection triggers:

• Lines matching the pattern ## Plan, # Implementation Plan, **Plan:**, etc.
• Content is debounced (500ms) to capture the full plan before saving

What happens:

1. Plan content is extracted from terminal output
2. If the terminal is linked to a task, the plan is saved to the task's plan field in YAML frontmatter
3. The task file is updated immediately

Configuration: Settings → Terminal → Terminal Automation → Detect and attach plans.

Linking terminals to tasks

To enable session capture and plan detection for a terminal, link it to a task:

1. Open the Tasks tab in the right panel
2. Find the task for the current work
3. Click Link Terminal (or set the link from within the task edit form)

The terminal-task link is stored in the task YAML and persists across restarts.

Combining features

All four features work together. A typical session looks like:

1. Start Claude Code in a terminal linked to task CL-042
2. Session ID is captured → added to CL-042.sessions
3. Claude generates a plan → saved to CL-042.plan
4. Claude hits a rate limit → Conductore schedules auto-resume
5. Rate limit resets → session resumes automatically
6. You spot the amber idle alert → check back in, send a message