openclaw-code-agent

About

Coding-agent sessions are ephemeral by default, which makes them a poor fit for chat platforms where a task spans hours and multiple follow-ups. This OpenClaw plugin treats each launched task as a managed session: Claude Code, Codex, or OpenCode runs in a git worktree in the background, the plan appears in the Telegram or Discord thread for approval, and subsequent plain-language replies steer the session — 'add unit tests', 'stop this session' — with wake routing deciding which session handles each message. Lifecycle operations go beyond launch: sessions suspend, resume, fork, and recover after restarts, and completed work surfaces as Merge, Open PR, Later, or Discard buttons. Optional goal loops run verifier-driven or 'Ralph-style' iterations unattended. Cost visibility comes from agent_stats, which reports per-session USD spend. OpenClaw users who want durable, approvable coding sessions from messaging apps are the audience.