The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is designed to collect diagnostic information such as environment details, user identity via superset auth whoami, and system logs to assist Superset engineers with debugging.
The instructions explicitly require the agent to warn users that logs or screenshots may contain private information and mandate obtaining user consent before including such data in a report.
Collected data is only sent to vendor-controlled destinations, specifically the support@superset.sh email address or the official superset-sh/superset GitHub repository.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied feedback which is subsequently used to generate report drafts and search documentation.
Ingestion points: User-provided text describing bugs, features, or general feedback.
Boundary markers: The instructions do not specify natural language delimiters for the LLM, but they define strict technical boundaries for tool interactions.
Capability inventory: The skill uses Bash to interact with the superset and gh (GitHub) CLI tools.
Sanitization: To prevent the user-supplied text from triggering unintended shell commands, the skill specifically instructs the agent to use heredocs (<<'EOF') or temporary files when passing data to the superset or gh commands.
[EXTERNAL_DOWNLOADS]: The skill fetches content from external sources to provide relevant documentation and version checks.
It retrieves documentation from docs.superset.sh and checks for software updates by reading version metadata from the official superset-sh/superset GitHub repository.
These operations are limited to the vendor's own infrastructure and are used for intended functional purposes.

feedback