Self-Correct

自纠错工具调用框架 - 轻量修正策略库 + 状态快照。对常见错误自动应对，高风险操作前保存快照。

This skill is coherent with its description, but review the following before installing: - Snapshot scope: nova-snapshot/cp -r will copy any path you pass. Do not snapshot /, /etc, home directories, SSH keys, or other sensitive locations unless you understand and accept that copies will be stored under /tmp/nova-snapshots. - Temp-file expectations: verify-result expects caller-created /tmp/stderr.$$ and /tmp/stdout.$$ files. Ensure your exec wrappers create these safely and avoid predictable names if running concurrently. - tag_error behavior: the tag_error function can print multiple lines (e.g., "success" plus other tags), which may confuse retry logic. Test how your agent consumes its output and consider normalizing a single tag return value. - Race/symlink considerations: cp and rm operations on files under /tmp can be susceptible to symlink/race issues if run as a privileged user. Run with least privilege and validate paths. - Unknown source: the skill's source/homepage is unknown. If you intend to use it in production, consider auditing or testing the script in a sandbox, and/or asking the author for provenance. Overall: safe to try in a controlled environment after addressing the above items; nothing in the package appears to exfiltrate data or require unrelated credentials.

Type: OpenClaw Skill Name: self-correct Version: 1.0.0 The skill bundle provides a framework for error diagnosis, automatic retries, and state snapshotting (backups) before high-risk operations. It includes shell functions in SKILL.md and a standalone script (scripts/snapshot.sh) to manage temporary copies of files in /tmp/nova-snapshots. The logic is transparent, lacks obfuscation, and aligns with its stated purpose of improving agent resilience through self-correction and recovery mechanisms.