The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires users to install the package alchaincyf/nuwa-skill via npx. This source is not recognized as a trusted vendor and is external to the stated author context of jwangkun.
[EXTERNAL_DOWNLOADS]: The research workflow in SKILL.md explicitly suggests downloading books and papers from shadow libraries such as Z-Library and LibGen, which are considered high-risk external sources.
[COMMAND_EXECUTION]: The skill utilizes the yt-dlp tool to download video content and transcripts, creating a command execution path using variables from external search results.
[PROMPT_INJECTION]: Personality template files (e.g., Steve Jobs, Elon Musk) include instructions intended to suppress safety disclaimers and limit the agent's meta-analysis, which can be used to bypass behavioral guardrails.
[PROMPT_INJECTION]: The skill maintains an indirect prompt injection vulnerability surface through its data collection process.
Ingestion points: Phase 1 in SKILL.md collects data from social media, articles, and transcripts.
Boundary markers: No delimiters or 'ignore instructions' warnings are present to isolate fetched data.
Capability inventory: The skill possesses the ability to spawn subagents, perform network operations, and write to the filesystem.
Sanitization: Fetched content is synthesized into instruction files without verification for malicious prompt injection patterns in the source material.

huashu-nuwa