aegis-service-sandboxing
Pass
Audited by Gen Agent Trust Hub on Aug 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill focuses on defensive security auditing and includes clear instructions for identifying and remediating containment weaknesses in Linux environments. It defines a multi-layered containment model and provides hardened templates for common application stacks.
- [INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection because it audits potentially attacker-influenced system configuration data, such as service unit files and cron entries. However, this risk is mitigated by explicit constraints that limit the agent to non-mutating commands and require it to interpret data through specific security rubrics.
- Ingestion points: systemd unit files, crontab files, process metadata, and container configuration (SKILL.md, sections 1-7).
- Boundary markers: The skill provides a logic-level boundary by instructing the agent to 'judge them against this module's rubrics, never against raw output alone.'
- Capability inventory: The skill utilizes standard diagnostic tools including systemctl, ps, grep, awk, stat, and docker inspect.
- Sanitization: Safety is enforced through strict operational constraints that forbid modifications to the host environment or service state.
Audit Metadata