The Agent Skills Directory

[DATA_EXFILTRATION]: The wbso script automatically collects sensitive local data and transmits it to the vendor's portal (https://portal.wbso.ai). It reads Git commit history (git log --all) from the current working directory. It scans for and parses session logs from other AI agents, specifically reading .jsonl files from ~/.claude/projects/ and ~/.codex/sessions/ to extract user prompts. This collected data is sent as 'context' in the cmd_context function, which is polled repeatedly (every 4 seconds) in SKILL.md while waiting for account activation.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data and local file content, presenting it to the agent without sufficient isolation. Data is ingested from the vendor API and local file system (Git logs and session logs from other agents) in scripts/wbso and subsequently read by the agent from /tmp/wbso-ctx.md in SKILL.md. The skill lacks explicit delimiters or instructions for the agent to ignore potentially malicious content embedded in the ingested logs or API responses. No robust sanitization or validation is performed on the content of the Git messages or third-party agent prompts before they are injected into the agent's context.
[COMMAND_EXECUTION]: SKILL.md includes a shell routine that uses find to recursively search through multiple hidden application directories (.claude, .cursor, .agents, .codex) in the user's home directory to locate the wbso binary.

wbso-signup