skills/alchaincyf/fanbox/fanbox-agent/Gen Agent Trust Hub

fanbox-agent

Warn

Audited by Gen Agent Trust Hub on Oct 13, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPERSISTENCEDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONPRIVILEGE_ESCALATION
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides the agent with the ability to send arbitrary commands to other terminal windows using the /send endpoint and to initiate new terminals with an autorun field in SKILL.md. This allows the agent to execute code and scripts across the user's terminal environment.\n- [PERSISTENCE]: The skill interfaces with a task scheduling system via the /cron/save endpoint. This allows for the creation of tasks that execute shell commands or agent prompts automatically based on cron expressions, fixed times, or intervals, which could be used to establish persistent access or recurring automated activities.\n- [DATA_EXFILTRATION]: The /read endpoint allows the agent to retrieve the command history and output of any terminal window (up to 2000 lines). The instructions acknowledge that this output may contain sensitive information, posing a risk of data exposure if the agent is compromised.\n- [INDIRECT_PROMPT_INJECTION]:\n
  • Ingestion points: Terminal output is ingested through the /read and /wait (long-polling) endpoints in SKILL.md.\n
  • Boundary markers: The instructions do not specify any boundary markers or delimiters when processing output retrieved from other terminal windows.\n
  • Capability inventory: The agent has extensive capabilities including cross-window command execution, terminal termination (/kill), and task scheduling for persistence.\n
  • Sanitization: There is no mention of sanitizing or escaping the output retrieved from terminals before it is processed by the agent, creating a surface for embedded instructions to influence agent behavior.\n- [PRIVILEGE_ESCALATION]: The skill documentation includes instructions on how to bypass safety confirmation prompts using parameters like full:true for automated cron tasks and the --dangerously-skip-permissions flag for sub-agents, which reduces human oversight and increases the agent's autonomous power.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 13, 2026, 09:08 AM