The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user data through several commands, including /文豪 <作家> <选题>, /literary <author> <topic>, and /文豪润色 <文本>. Ingestion points: User-provided topics, story text for polishing, and project names are directly incorporated into the agent's context. Boundary markers: The instructions do not define delimiters or specific 'ignore' rules for user-supplied content during the imitation and editing workflows. Capability inventory: The skill instructs the agent to read/write files within project directories and execute a shared memory CLI tool. Sanitization: There are no specific instructions for the agent to sanitize or validate the content of the user-provided text before processing.
[DYNAMIC_EXECUTION]: The skill relies on a local script ../_shared/memory/cli.py for session persistence. The agent is instructed to invoke this script using python at the start of a session and after user operations to load and save state.
[COMMAND_EXECUTION]: The SKILL.md file provides command-line templates that pipe user-controlled strings (like <项目名>) into a Python script via echo. This pattern, if implemented naively by an agent, could lead to command injection if the user-supplied project name contains shell metacharacters.

literary-ghostwriter