Skip to content

Cross-harness delegation, without the duct tape

Let agents delegateto the competition.

Codex wasn't built to delegate to Claude.Claude wasn't built to delegate to Codex. dlgt was.

I built an entire company with 47 AI agents. Hey Sol, ask Fable to review this.

Supported harnesses

One bridge. Six harnesses.

Run another coding agent in a live, addressable terminal Session. Follow up in the same conversation and read its provider-reported result.

  • Codex
  • Claude
  • Cursor
  • Grok
  • OpenCode
  • Pi
Explore CLI usage

Every major harness already has subagents.

The missing piece is the bridge dlgt.

You've already tried the DIY routes. dlgt was built for this.

  • tmux send-keys
    • Your agent polls capture-pane and burns tokens on screen dumps
    • Or you script UI heuristics that break on a spinner
  • claude -p / codex exec
    • A cold start on every call — context is thrown away
    • Headless runs sometimes aren't covered by your subscription
  • dlgt

    One job: your agent uses the competitor's agent.

    • Completion is a lifecycle event — done means done
    • Durable sessions — follow-ups keep their context
    • Managed PTY, JSON results — no scraping, no tmux
    • On the plan you already pay for

Install once.Ask naturally.

Give either harness the install instructions. After that, delegation is part of the prompt.

codex "Install and verify dlgt. Fetch https://combinatrix.ai/dlgt/installation.md with curl and follow its instructions."

Then ask either agent normally.

Codex
codex -m gpt-5.6-sol "Create a great game. Ask Fable to review it."
Claude
claude --model claude-fable-5 "Think of 10 funny jokes. Ask Sol at xhigh effort to review them."

Bonus: naming an effort is also enough — native subagents can't choose theirs.

Codex
codex -m gpt-5.6-sol "Make the CLI faster. Have Luna do it at xhigh effort."

One bridge. Six harnesses.