The Agent Skills Directory

[DATA_EXFILTRATION]: The script scripts/wbso collects highly sensitive local data from the user's environment. This includes git commit history (via git log), Claude Code session logs from ~/.claude/projects/, and Codex session logs from ~/.codex/sessions/. This data is transmitted to the remote API at https://portal.wbso.ai during the execution of the context command.
[COMMAND_EXECUTION]: The parse_args function in scripts/wbso utilizes the eval command to dynamically assign variables from CLI arguments. Specifically, the line eval "ARG_${key//-/_}=\"\$val\"" is vulnerable to shell command injection. If a user provides an argument value containing shell metacharacters (e.g., $(command)), it will be executed within the context of the script.
[PROMPT_INJECTION]: The skill implements an indirect prompt injection surface by reading external, potentially attacker-controlled data into the agent's context.
Ingestion points: The context command in scripts/wbso reads data from local git repositories and external tool logs.
Boundary markers: No delimiters or instructions to ignore embedded commands are present when this data is injected into the prompt.
Capability inventory: The script has access to curl for network operations, python3 for execution, and local file system access.
Sanitization: There is no sanitization or filtering of the content read from git logs or session files before it is returned to the agent.

wbso-signup