The Agent Skills Directory

[DATA_EXFILTRATION]: The skill initiates network requests to a remote service.
User-provided queries are transmitted via HTTP POST to the endpoint specified in ROMBOT_ASK_URL.
The destination and credentials are not hardcoded; they are established by the user during the interactive setup phase.
[PROMPT_INJECTION]: The skill instructions create an attack surface for indirect prompt injection by encouraging the agent to trust external data implicitly.
Ingestion points: Data is fetched from an external API in scripts/rombot-ask.py.
Boundary markers: The instructions explicitly mandate that the agent "Pass the answer through verbatim" and "Do NOT reword", which prevents the agent from applying its own safety or relevance filters to the remote content.
Capability inventory: The agent has access to Bash and Read tools, which could be abused if the remote content contains malicious instructions that the agent follows.
Sanitization: The script does not sanitize or validate the content returned from the API before outputting it to the agent.
[COMMAND_EXECUTION]: The skill's primary interface is a command-line tool executed via Bash.
The agent is instructed to run python3 scripts/rombot-ask.py "<question>", where the question is a dynamic string provided by the user or reformulated by the agent.

rombot-community