brand-voice-analyzer

Warn

Audited by Gen Agent Trust Hub on Apr 10, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill instructions explicitly direct the agent to analyze sensitive data types if available, including "Internal comms (all-hands emails, Slack announcements, internal memos)" (SKILL.md). This exposes sensitive information to the agent's processing environment, where it could be vulnerable to misuse or accidental exposure.
  • [PROMPT_INJECTION]: The skill is designed to ingest untrusted data from URLs and local files (SKILL.md), creating a surface for indirect prompt injection.
  • Ingestion points: SKILL.md (Phase 1) specifies the use of WebFetch for content from URLs and Read/Glob for local files, including potential internal communications logs.
  • Boundary markers: The instructions lack requirements for boundary markers or delimiters to isolate untrusted data from the agent's internal instructions.
  • Capability inventory: The skill uses tools including Bash, Write, Edit, and WebFetch, which could be abused if an injection attack is successful.
  • Sanitization: There are no instructions to sanitize or validate the ingested content before it is processed for analysis.
  • [COMMAND_EXECUTION]: The skill instructs the agent to use the Bash tool (SKILL.md). While the current workflows do not detail specific shell commands, the availability of a high-privilege execution environment in the same context as untrusted data fetched from the web increases the potential impact of an exploit.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 10, 2026, 05:26 PM