@ai-devkit/agent-manager

Detect, inspect, and send prompts to running AI coding agent sessions.

This package powers the ai-devkit agent commands. Use it when you need the lower-level agent session management utilities that AI DevKit uses to find active Claude Code, Codex, Gemini CLI, and other supported coding-agent sessions.

What It Provides

• Session detection — Find running agent sessions across supported providers
• Session details — Inspect agent metadata, working directory, and status
• Prompt sending — Send follow-up instructions to an existing session
• Provider adapters — Shared adapter layer for agent-specific behavior
• Terminal control — Discover, focus, and type into the terminal pane that hosts a running agent. Supported emulators: tmux, WezTerm, iTerm2, and macOS Terminal.app. Resolution is automatic from the agent PID's TTY.

Typical Use

Most users should use the CLI:

ai-devkit agent list
ai-devkit agent send "run the tests and report back" --id <agent-name> --wait
npm test 2>&1 | ai-devkit agent send --id <agent-name> --stdin

Use this package directly only when building custom tooling around AI DevKit's agent detection and control surface.

Documentation

Full guides and workflow examples: ai-devkit.com/docs

License

MIT