The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill documentation includes a command to install an external Model Context Protocol (MCP) server: npx -y @mikoto_zero/minigame-open-mcp@latest. This downloads and executes code from a package under the @mikoto_zero npm scope, which does not match the skill's declared vendor (taptap) and is not a verified or trusted source.
[EXTERNAL_DOWNLOADS]: The skill depends on a remote package from the npm registry that is not maintained by a trusted organization or the skill author, introducing a supply chain risk.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-generated content from TapTap reviews and community forums using tools like get_current_app_reviews and get_current_app_forum_contents. This content could contain malicious instructions designed to bypass agent safety filters or manipulate the generated operations briefs.
Ingestion points: get_current_app_reviews and get_current_app_forum_contents in SKILL.md.
Boundary markers: The skill requires human confirmation for write actions (like_current_app_review, reply_current_app_review), which helps mitigate injection but does not prevent the agent from being influenced during the summary generation phase. No explicit delimiters or boundary warnings are provided for the prompt interpolation of this data.
Capability inventory: The skill can fetch store metrics, snapshot data, and forum content, and can perform write actions such as liking or replying to reviews.
Sanitization: No explicit sanitization or validation of the ingested external content is described in the skill instructions.

taptap-dc-ops-brief