A robust execution layer that decouples Claude Code from the orchestrator using tmux for persistent, crash-resistant coding tasks.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install resilient-claude-agent
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install resilient-claude-agent using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Resilient Claude Agent?
The Resilient Claude Agent is a specialized skill designed to handle long-running coding tasks that are vulnerable to orchestrator restarts, process crashes, or network interruptions. By leveraging tmux to isolate the execution environment, this skill provides a stable platform for autonomous coding. It is a critical addition to any suite of Openclaw Skills, allowing users to delegate complex tasks like refactoring or debugging without needing to maintain an active connection.
This skill operates under a fire-and-forget philosophy, utilizing the --dangerously-skip-permissions flag to ensure Claude Code can execute bash commands and file modifications without human intervention. It includes a comprehensive monitoring system that watches for process health and output staleness, automatically resuming sessions if a failure is detected.
Resilient Claude Agent Use Cases
- Long-running architectural refactoring and codebase migrations.
- Automated test suite generation and iterative debugging cycles.
- Deep file exploration and documentation generation for large projects.
- Background security analysis and code review tasks.
- Maintaining persistent coding sessions in environments with unstable network connectivity.
How Resilient Claude Agent Works
- The skill initializes a secure, isolated temporary directory to manage the task lifecycle.
- A background tmux session is created, ensuring the coding process is decoupled from the parent shell context.
- The skill captures real-time output through a pipe-pane mechanism that strips ANSI escape codes for clean logging.
- A wrapper script launches Claude Code, delivering prompts via stdin to prevent sensitive data from appearing in process lists.
- A dedicated health monitor script runs with nohup to track PIDs, detect hangs, and trigger automatic session resumes using the claude -c command.
Resilient Claude Agent Setup
To utilize this skill within your Openclaw Skills environment, ensure you have tmux and the Claude Code CLI installed. The setup follows a specific sequence:
# Create a secure workspace
TMPDIR=$(mktemp -d) && chmod 700 "$TMPDIR"
# The orchestrator writes task details to a manifest
# and launches the monitor script in the background
nohup bash scripts/monitor.sh claude-<task-name> "$TMPDIR" >"$TMPDIR/monitor.log" 2>&1 &
Resilient Claude Agent Data Schema & Taxonomy
The skill organizes its state within a temporary directory using a structured manifest system:
| File | Purpose |
|---|---|
| manifest | Structured task state (status, PID, model, start time). |
| output.log | Sanitized continuous output capture from the tmux pane. |
| prompt | The raw task instructions delivered to Claude. |
| done | A completion marker file indicating the task lifecycle is finished. |
| exit_code | The numeric return code of the final Claude process. |
| pid | The current active process ID for the Claude Code agent. |
Resilient Claude Agent Advanced Features
- Layered health detection prioritizes done-file checks over PID liveness to prevent race conditions.
- Exponential backoff strategy for monitoring intervals to optimize system resources during failures.
- Pane state classification that detects waiting input prompts or internal agent crashes.
- Automatic session cleanup and abandonment protocols to prevent orphan processes.
- Secure prompt handling that bypasses shell interpolation for improved security and reliability.
SKILL.md
Loading
Related Openclaw Skills
TDD E2E PR Workflow
Automate the end-to-end development cycle from issue selection to pull request creation with mandatory visual verification.
Extract Design System
Reverse-engineer public website design primitives into localized design tokens and CSS variables automatically.
Ark Seedance Video Generation
An end-to-end video generation skill for AI agents using Volcengine Ark Seedance models for professional video synthesis.
Doc Doctor
Doc Doctor is an automated maintenance tool for markdown knowledge bases that scans for broken links, missing metadata, and orphan pages to ensure a high-quality documentation score.
Video Analyzer
A local AI tool for downloading and analyzing video content from Bilibili, Douyin, and Toutiao using frame extraction and local vision models.
Polymarket Micro Session Edge Trader
An automated trading agent designed to exploit predictable mean-reversion patterns in Polymarket 5-minute crypto markets during global session opens.
Featured*
