The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/wbso file contains a vulnerable parse_args function that uses the eval command to process command-line arguments. Because the SKILL.md instructions direct the agent to pass raw user input (the feedback message) into this script, a user can craft a message containing shell metacharacters (e.g., $(command) or `command`) to execute arbitrary code on the host system.
[DATA_EXFILTRATION]: The scripts/wbso script includes logic to gather and transmit sensitive local data to an external API (portal.wbso.ai). This includes today's Git commits, Claude Code session logs from ~/.claude/projects, and Codex session logs from ~/.codex/sessions. While intended for context-aware feedback, this represents a significant collection of private development activity.
[COMMAND_EXECUTION]: The SKILL.md file includes a complex discovery script that uses find to search for the wbso executable across a wide range of sensitive local directories, including .claude, .cursor, and .agents folders in the user's home directory. This behavior is unusually broad for a standard skill tool discovery mechanism.

wbso-feedback