dd-apm
Pass
Audited by Gen Agent Trust Hub on Nov 12, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPRIVILEGE_ESCALATIONPERSISTENCEDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill downloads and executes an installation script from
https://install.datadoghq.com/scripts/install_script_agent7.shto deploy the Datadog Agent. This is a trusted vendor resource but represents remote code execution.\n- [COMMAND_EXECUTION]: The skill makes extensive use ofsshandkubectlto discover services, verify configurations, and manage workloads across remote systems and clusters.\n- [PRIVILEGE_ESCALATION]: Administrative tasks are performed usingsudoto modify systemd units, access kernel-level process information, and manage the Datadog Agent service.\n- [PERSISTENCE]: Single Step Instrumentation is enabled by modifying the/etc/ld.so.preloadconfiguration, which ensures the Datadog launcher is loaded into new processes at startup.\n- [DATA_EXFILTRATION]: The skill reads/proc/<PID>/environto verify Unified Service Tags. Accessing process environment data is a sensitive operation that could expose credentials if not properly scoped.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes output from numerous system discovery and management tools, creating an attack surface for indirect prompt injection.\n - Ingestion points: Outputs from commands such as
pgrep,ls,cat /proc/maps,kubectl get pods, and thepupCLI tool.\n - Boundary markers: No explicit markers or instructions are used to delimit or ignore potentially malicious content within these data streams.\n
- Capability inventory: The agent possesses high-privilege capabilities including SSH access, Kubernetes cluster administration, local privilege escalation via sudo, and network access.\n
- Sanitization: The skill uses filtering commands like
grepandtrto isolate specific Datadog variables, but it does not implement comprehensive validation for all ingested system data.
Audit Metadata