The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of system commands including tar, zstd, rsync, find, systemctl, and sha256sum to perform archive creation, compression, remote transfer, and verification tasks during the backup and restore processes.
[DATA_EXFILTRATION]: The skill is designed to package and facilitate the movement of sensitive data, specifically identifying that the backups contain API tokens, messaging pairing data, and OAuth refresh tokens. It provides instructions for transferring these artifacts to a user-defined remote host via rsync.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from the environment to drive its logic. Ingestion points: The skill reads the output of openclaw --version, openclaw backup --help, and configuration values like agents.defaults.workspace. Boundary markers: The skill relies on a manual "gate" protocol for verification rather than using explicit prompt delimiters for external data. Capability inventory: The skill utilizes powerful tools including tar, rsync, node, find, and systemctl. Sanitization: The bundled Node.js script performs basic escaping of single quotes in file paths used within SQLite commands.
[DYNAMIC_EXECUTION]: A bundled Node.js script (sqlite-snapshot.md) is used to perform transactional SQLite snapshots. The script uses child_process.execFileSync to discover databases and the node:sqlite module to execute VACUUM INTO commands.

openclaw-backup