Team Orchestration for Coding Agents

Quick start

Start with one visible worker, watch its derived health state, then dry-run finish before closing disposable worker panes.

Launch a scoped worker prompt

forktty team ask review-team codex-worker --agent codex --task-id inspect-ui --prompt "Inspect the current diff and report risks" --submit

Watch worker health and inbox

forktty team watch review-team --stale-after-ms 120000 --limit 10

Plan final cleanup

forktty team finish review-team --dry-run

Finish and close disposable workers

forktty team finish review-team --close-workers

Roles, tasks, and mailbox dispatch

Team state records the leader, workers, tasks, messages, dispatch events, and health snapshots. Prompts can be queued and delivered only after the worker pane is ready.

Provider-aware worker handling

ForkTTY can choose providers from Settings > Agents when a worker is launched with auto selection, or use explicit providers when a task requires Codex, Claude Code, Pi, OpenCode, Antigravity, or a shell lane.

Finish with evidence

Team finish checks open tasks, pending messages, and live-looking worker final states before marking work done. Current-runtime launch-owned disposable panes can be closed deliberately; stale launch records are not enough.

Questions

Does ForkTTY run hidden autonomous agents?

No. ForkTTY records workflow loop state as metadata and keeps actual agent work visible through terminal panes and explicit socket/MCP/CLI calls.

Can team workers use different providers?

Yes. Workers can use explicit providers or auto-selection based on Settings > Agents and detected executables.

How does ForkTTY avoid closing user panes by mistake?

Close-worker flows are limited to disposable surfaces launched by the current ForkTTY runtime's team tools.

Explore related ForkTTY pages

Claude Code Codex MCP Docs

Get the alpha build

AppImage and .deb packages are published on GitHub Releases for Linux x86_64.

View releases

Coordinate coding-agent teams without hiding the terminals.