The Agent Skills Directory

[DATA_EXFILTRATION]: The skill performs extensive collection of sensitive system data, including restricted files like /etc/shadow and user SSH authorized_keys. While the collection is local or directed to vendor-specified S3 buckets, the exposure of these credentials to the agent environment is a significant capability.
[PRIVILEGE_ESCALATION]: The evidence collection scripts (e.g., scripts/collect-evidence.sh and the snippets in SKILL.md) require and explicitly instruct the user to execute them with root or sudo privileges. The AWS script also creates and applies new security groups, representing a significant privilege acquisition on the infrastructure level.
[COMMAND_EXECUTION]: The skill contains multiple shell scripts and references that execute a wide range of powerful system utilities (e.g., ps, ss, iptables, tcpdump, gcore). These tools provide the agent with deep visibility and control over the operating system and network traffic.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and analyze untrusted data from potentially compromised systems, creating a vulnerability surface for indirect prompt injection.
Ingestion points: The forensic scripts read from system logs (/var/log/auth.log, syslog), process lists, and network connection metadata.
Boundary markers: No delimiters or "ignore embedded instructions" warnings are used to separate the collected forensic data from the agent's internal logic.
Capability inventory: The agent has access to numerous high-privilege system commands, cloud APIs (AWS CLI), and orchestration tools (kubectl).
Sanitization: The collected log data and command outputs are not sanitized or validated before being processed by the agent.

incident-response