claude-code-legacy
by yaxuan42
Coding Agents & IDEsBrief
Trigger Claude Code development tasks in observable tmux sessions with stable startup, progress visibility.
Description
What This Skill Does
Orchestrates Claude Code development tasks inside named tmux sessions so long-running coding work stays observable and recoverable. Each task starts from a prompt file, runs claude with skipped permissions, and sends a completion callback with a structured JSON/MD report. Progress is inspectable live over SSH without blocking or losing output.
Running Claude Code inside a managed tmux session prevents silent hangs and preserves full pane output, making stalls detectable and sessions restartable without losing context.
When to Use It
- Running a multi-hour refactor job while away from the machine
- Monitoring AI coding progress via SSH from a different device
- Getting auto-notified when a Claude Code task finishes
- Reviewing diffs and risk summaries after automated code changes
- Listing all active Claude Code sessions in a single status sweep
Example Workflow
User asks: Run a gallery website redesign using Claude Code and notify me when done
- 1
Write the prompt to a file to avoid shell quoting issues
- 2
Start a named tmux session via start-tmux-task.sh with label, workdir, and prompt-file
- 3
Launch claude --dangerously-skip-permissions in interactive mode inside the session
- 4
Embed wake.sh callback in the prompt so task completion triggers a notification
- 5
Return the session name and SSH attach command to the user
User receives a wake notification referencing the completion report path; agent reads the JSON/MD report and provides analysis of what was completed, risks, and next steps
Requirements
- Claude Code CLI (claude) installed and authenticated
Installation
clawhub install yaxuan42/claude-code-legacyRun the CLI command, or paste the prompt into Claude Code, Cursor, Codex or whichever coding agent you use.