The Agent Skills Directory

[DATA_EXFILTRATION]: The scripts/wbso script collects sensitive local activity data and sends it to the vendor's API (portal.wbso.ai) during the context command execution.
Evidence: The append_local_signals function in scripts/wbso reads git log, Claude chat history from ~/.claude/projects/*.jsonl, and Codex session history from ~/.codex/sessions/*.jsonl.
This data is aggregated and transmitted via curl to the vendor's endpoint to generate compliance reports.
[INDIRECT_PROMPT_INJECTION]: The skill aggregates data from external sources and local logs into the agent's context, creating a surface for indirect prompt injection.
Ingestion points: Data is ingested from the WBSO API (/api/v1/compliance/context.md), local Git logs, and local AI agent session files in ~/.claude/projects and ~/.codex/sessions.
Boundary markers: The script uses XML-style tags such as <git_commits_today> and <claude_user_prompts_today> to delimit the ingested content.
Capability inventory: The skill has the ability to perform network requests (curl), write configuration files to the filesystem (~/.config/wbso/config), and execute Python scripts for data processing.
Sanitization: The script performs basic prefix filtering on chat logs (e.g., skipping system reminders) but does not implement robust sanitization or escaping to prevent embedded instructions from influencing the agent.
[COMMAND_EXECUTION]: The skill executes shell commands to discover the location of its CLI tool and uses Python for JSON and text processing.
Evidence: SKILL.md contains a complex find command pipeline to locate the wbso executable across various common installation paths.
Evidence: scripts/wbso uses python3 -c and python3 - with heredocs to parse JSON responses from the API and extract specific fields from Markdown content.

wbso-auth