The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill is architected to ingest and refine sensitive user-provided academic data, including research abstracts, proposal drafts, and personal CVs (as seen in the workflows of SKILL.md and framework-guide.md).
Ingestion points: Untrusted data enters the agent context through chat interactions and the AskUserQuestion tool throughout the multi-step writing process.
Boundary markers: The skill lacks explicit delimiters or specific instructions to the agent to treat user-provided data as potentially adversarial or to ignore instructions embedded within research materials.
Capability inventory: The skill utilizes WebSearch and MCP-integrated scholarly tools (PubMed, Semantic Scholar). It also instructs the agent to generate and save reports (e.g., 点评结果.md in expert-scoring.md).
Sanitization: There is no evidence of sanitization, escaping, or validation of user-provided content before it is processed by the writing and reviewer modules.
[SAFE]: No executable code or scripts (.py, .js, .sh) are included in the skill. All literary search operations target well-known, trusted academic services (PubMed, Semantic Scholar). There are no hardcoded credentials or unauthorized network operations detected.

nsfc-writer