The Agent Skills Directory

[DATA_EXFILTRATION]: The skill performs discovery sweeps that access sensitive file paths to audit security posture. Evidence in SKILL.md shows it searches for TLS private keys in /etc/ssl/private, application secrets in .env files, and backup-specific credentials such as RESTIC_PASSWORD and BORG_PASSPHRASE. While this is aligned with the purpose of a backup audit, it involves direct exposure of host credentials and encryption material.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data by parsing documentation and runbooks to identify recovery evidence. This creates a potential surface for indirect prompt injection.
Ingestion points: The audit sweep in SKILL.md uses grep to search for patterns like 'restore', 'RTO', 'RPO', and 'disaster' within README*, docs/, runbooks/, and other markdown files.
Boundary markers: No explicit delimiters or instructions are used to distinguish untrusted file content from the agent's instructions.
Capability inventory: The skill uses systemctl, grep, find, awk, ss, docker, and file for discovery. It does not define network-outbound or file-write capabilities for the audit phase.
Sanitization: No sanitization or filtering logic is present for the processed documentation content.
[COMMAND_EXECUTION]: The skill relies on complex shell command sequences for discovery, including enumerating systemd timers, searching crontabs, and scanning the filesystem for backup artifacts using find and grep across /usr/local/sbin, /opt, and /root/bin as defined in SKILL.md.
[PRIVILEGE_ESCALATION]: The skill explicitly requires root access to perform a complete audit of the host. This includes reading system cron spools in /var/spool/cron/crontabs and listing crontabs for all service accounts, as described in the 'What To Check' and 'Where To Look' sections of SKILL.md.

aegis-backup-dr