The Agent Skills Directory

[NO_CODE]: The skill does not contain executable code or external scripts. It provides documentation for using a Model Context Protocol (MCP) plugin and a CLI tool (sam_cli) that are expected to be part of the host environment.
[DATA_EXFILTRATION]: While the skill involves uploading content to a backlog server, it uses specific, internal tools (mcp__plugin_dh_backlog__artifact_register) and follows established registry protocols. There are no signs of exfiltration to unauthorized external domains.
[INDIRECT_PROMPT_INJECTION]: The skill processes artifact content provided by other agents or users. While this presents an attack surface for indirect prompt injection, the skill defines clear boundary markers (parameters like content=, artifact_type, etc.) and instructs agents to return only logical IDs rather than echoing content, which mitigates risk.

create-artifact