The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a find command with a broad set of paths (including $HOME and $PWD) to locate its CLI tool (scripts/wbso). While intended for local path resolution across different IDEs (Claude Code, Cursor), searching broad user directories is a heavy operation.
[DATA_EXF0ILTRATION]: The scripts/wbso script collects local signals including Git commits (git log) and recent user prompts from Claude/Codex local logs (~/.claude/projects/, ~/.codex/sessions/). This data is appended to the context sent to the vendor's API (portal.wbso.ai). This is documented as intended functionality for compliance tracking but involves transmitting local activity data.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external data via the cmd_context function which fetches content from the WBSO.ai API. Evidence Chain:
Ingestion points: Fetches context.md from portal.wbso.ai via curl.
Boundary markers: The output is wrapped in <status> tags and specific project XML tags, but no explicit 'ignore instructions' markers are present in the bash script.
Capability inventory: The skill can write to ~/.config/wbso/config and execute various CLI subcommands.
Sanitization: Uses python3 for JSON parsing and sed/grep for basic text extraction, providing some structure-based sanitization.

wbso-auth