The Agent Skills Directory

[Data Exposure & Exfiltration] (LOW): The skill performs network requests to https://slack.com/api. While necessary for functionality, the domain is outside the predefined trusted source whitelist.
Evidence: scripts/slack_api.py lines 86-95 use urllib.request to send data to the Slack API.
[Indirect Prompt Injection] (LOW): The skill retrieves untrusted data from Slack (messages, threads, search results) which could contain instructions targeting the LLM.
Ingestion points: scripts/slack_api.py (functions get_history, get_thread, search_messages, my_posts).
Boundary markers: The script provides basic separators in text mode (=== {user} ({ts}) ===) to delimit messages in format_messages.
Capability inventory: The skill can read channel history and user data, and post messages back to Slack.
Sanitization: Content is truncated by truncate_text, but no filtering of prompt injection keywords or instruction-like patterns is implemented.

slack