from claude-code-dispatch44
Dispatch long-running development tasks to Claude Code with automated callbacks, Agent Teams support, cost controls, and git worktree isolation.
Claude Code Dispatch provides a CLI-driven dispatch system that launches Claude Code runs in the background, manages task metadata, and triggers Stop/TaskCompleted hooks to notify users (e.g., via Telegram). It supports Agent Teams for parallel dev+test workflows, model fallback, budget caps, and worktree isolation for safe repo changes.
Use this skill when you need to run a multi-step development task that should execute in the background, report results automatically, or coordinate multi-agent workflows (build+test). It's the default for requests like 'use Claude Code to build X' or any job that takes >2 minutes and needs zero-polling callbacks.
Designed for environments that can run PTY-wrapped model runners (Claude Code) and handle webhook/callback processing; useful with OpenClaw agents orchestrating background runs.
Skill dispatches Claude Code tasks with auto-callback, Agent Teams support, cost controls, and git worktree isolation. Python runner is well-structured; dispatch.sh is comprehensive but has a hardcoded OpenClaw gateway token as fallback. notify-agi.sh works but has hardcoded /home/ubuntu/ paths causing failures outside that environment. Scripts ran in dry-run mode: only notify-agi.sh exited cleanly despite directory errors.
claude (Claude Code CLI binary)Hardcoded credential token in dispatch.sh is a security concern — should require env var without fallback default. Skill is genuinely useful for OpenClaw users wanting background Claude Code execution with Telegram notifications.
