The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides commands for network sniffing and traffic manipulation.
Evidence: Use of tcpdump -A -vvv for full traffic capture and socat -v TCP-LISTEN:8080,fork TCP:<target-svc>:80 for man-in-the-middle attacks.
[DATA_EXFILTRATION]: The skill facilitates the capture and potential extraction of sensitive information from internal service traffic.
Evidence: Instructions to save traffic to /tmp/capture.pcap for later analysis and recommendations to monitor service-to-service communication for flag or token data.
[CREDENTIALS_UNSAFE]: The skill describes methods to specifically target and harvest authentication secrets.
Evidence: Grep patterns are provided to extract flag, secret, password, token, key, Authorization headers, and Cookie values from HTTP payloads.
[INDIRECT_PROMPT_INJECTION]: The skill exposes the agent to untrusted external data by reading network traffic which may contain malicious instructions.
Ingestion points: Live network traffic captured via tcpdump and redirected via socat (SKILL.md).
Boundary markers: None used to separate captured data from the agent's instructions.
Capability inventory: The skill uses ps, tcpdump, and socat to inspect and intercept environment data.
Sanitization: No sanitization or validation is applied to the intercepted network payloads before they are processed by the agent.

k8s-sidecar-attack