The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): The skill facilitates the execution of powerful system utilities such as tcpdump, openssl, nc, and curl. Many of these tools typically require root or sudo privileges (e.g., tcpdump for raw socket access). An agent could use these to perform unauthorized reconnaissance, disrupt connectivity, or modify system network configurations.
[CREDENTIALS_UNSAFE] (HIGH): The Nginx configuration template provided in SKILL.md explicitly points to a sensitive private key path: /etc/ssl/private/api.key. If an agent is permitted to implement or 'debug' this configuration, it might attempt to read, display, or exfiltrate the contents of this key.
[DATA_EXFILTRATION] (HIGH): The tcpdump commands (especially tcpdump -i any -A) capture network packet content in ASCII format. This can expose passwords, session tokens, and other PII transmitted over unencrypted protocols (HTTP, Telnet). Combined with curl, an attacker could exfiltrate this captured data to a remote endpoint.
[INDIRECT_PROMPT_INJECTION] (HIGH): The skill is designed to ingest and analyze data from untrusted external sources (network traffic via tcpdump and HTTP responses via curl).
Ingestion points: Output from curl requests and live network traffic captured by tcpdump.
Boundary markers: None present. The agent processes raw command output directly.
Capability inventory: Shell execution (bash), sensitive file path access (SSL keys), and network transmission (curl).
Sanitization: No sanitization or filtering logic is provided for the data processed by the agent.

network