The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes the Agent() tool to spawn a sub-agent defined in .claude/agents/moshu-researcher.md. This capability allows the skill to delegate complex research and search tasks to a specialized component.
[EXTERNAL_DOWNLOADS]: The workflow involves fetching data from external web sources (wikis, platform directories, and analysis articles) via the moshu-researcher sub-agent. This is an intended functionality for its research purpose.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core function of ingesting and processing untrusted data from the internet.
Ingestion points: External web content, character wikis, and plot analyses retrieved by the research agent at runtime.
Boundary markers: The skill explicitly implements a "Transfer Discipline" (转译纪律) in caifeng-methods.md that instructs the agent to transform content and avoid direct copying, acting as a functional boundary against accidental instruction following.
Capability inventory: The skill possesses the ability to spawn sub-agents, read/write files within the project directory (设定/ and references/), and prompt the user for input.
Sanitization: The skill mandates a "Translation" process (转译深度三问) and manual author verification for character-related data to ensure external content is sanitized and adapted to the local project context before use.

moshu-research