Restart Recovery

Make OpenClaw agent workflows restart-safe using checkpoint files, idempotent step tracking, wake/resume handoff, and stale-checkpoint monitoring. Use when u...

This skill appears to do what it says: local checkpointing and resume. Before installing or enabling it, review and test these points: 1) Do not store secrets/tokens in checkpoint data — the tools may print checkpoint JSON to stdout and write it to disk. 2) Confirm the checkpoint directory (memory/checkpoints under the agent's working directory) is the right location for your deployment and set restrictive file permissions. 3) Review the startup change you will add to AGENTS.md; enable it in a staging environment first so you can validate resume behavior. 4) Understand the simple lock mechanism (file-based) and the '--steal-lock' behavior; if multiple agents may touch the same directory, consider stronger locking or agent coordination. 5) Establish retention/cleanup and expiry policies for old checkpoints. If you want additional assurance, run the scripts in a controlled test workspace and inspect all generated files/logs for any sensitive data exposure.

Type: OpenClaw Skill Name: restart-recovery Version: 1.0.0 The skill bundle implements workflow state management but contains a path traversal vulnerability in `scripts/checkpoint_tool.py`, where the `--workflow` argument is used to construct file paths without sanitization, allowing JSON files to be written outside the intended directory. Additionally, `SKILL.md` includes a 'Required operating rule' that instructs the AI agent to automatically resume unfinished workflows on startup; while aligned with the skill's purpose, this creates a persistence mechanism that could be abused if a malicious checkpoint file is planted on the filesystem.