The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions contain no attempts to bypass safety filters, override system prompts, or solicit restricted information. All instructions are focused on the task of voice analysis and guide generation.
[DATA_EXFILTRATION]: No evidence of hardcoded credentials, sensitive file path access (e.g., .ssh, .env), or unauthorized data transmission was found. Network activity is limited to the WebSearch tool for legitimate brand research purposes.
[REMOTE_CODE_EXECUTION]: The skill does not download external scripts or use dangerous patterns like piping network content into a shell.
[COMMAND_EXECUTION]: The skill uses allowed tools (Read, Write, Edit) to manage local documentation files within the user's project directory (e.g., docs/brand-voice.md). There are no patterns suggesting privilege escalation or unauthorized system modifications.
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests user-provided writing samples or content from WebSearch. However, the capability inventory (Read, Write, WebSearch) is standard for the task, and the instructions focus on qualitative pattern extraction rather than execution of the ingested data. No sanitization is explicitly defined, but the risk is consistent with normal agent-assisted research tasks.

brand-voice