aegis-incident-response-lifecycle
Warn
Audited by Gen Agent Trust Hub on Aug 31, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute various shell commands for triage and evidence collection, including ps, ss, and journalctl, and to run local scripts such as tools/run-all-sweeps.sh.
- [PRIVILEGE_ESCALATION]: The skill explicitly identifies the requirement for root privileges to execute specific network diagnostic commands and describes the use of break-glass local admin accounts for emergency system access during incident response.
- [INDIRECT_PROMPT_INJECTION]: The skill defines a surface for indirect prompt injection through its data ingestion process. 1. Ingestion points: Untrusted data from incident intake fields including 'What happened', 'Detection source', and 'Evidence pointers' (alert URLs, log queries, and customer reports) are ingested into the agent context in SKILL.md. 2. Boundary markers: No delimiters or explicit instructions to ignore embedded commands are specified for these external inputs. 3. Capability inventory: The agent is authorized to execute shell commands, run local triage scripts, and perform database administration tasks. 4. Sanitization: The instructions do not define any escaping or validation mechanisms for the ingested content.
Audit Metadata