The Agent Skills Directory

[DATA_EXFILTRATION]: The skill's CLI wrapper collects local user data and transmits it to the vendor's external API (portal.wbso.ai).
Evidence: The append_local_signals function in scripts/wbso scrapes Git commit logs and previous user prompts stored in ~/.claude/projects/ and ~/.codex/sessions/.
This data is included in the compliance context sent via curl to the vendor's portal during the context command execution.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from PDF files and local logs, creating a surface for potential instruction injection.
Ingestion points: PDF file uploads (cmd_upload) and Git/session log extraction (cmd_context) in scripts/wbso.
Boundary markers: None identified; the data is interpolated into the context provided to the agent.
Capability inventory: The skill performs file system searches, writes configuration to ~/.config/wbso, and executes network requests.
Sanitization: Basic filtering is performed via embedded Python scripts, but it does not account for adversarial prompt injection in the ingested logs or documents.

wbso-signup