The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains explicit instructions to override the agent's safety filters when analyzing fictional content. In SKILL.md, it states that the agent must not refuse chapters or the entire book based on 'sensitive' or 'uncomfortable' content such as violence or dark ethics, essentially attempting to bypass standard content safety protocols.
[COMMAND_EXECUTION]: The skill instructs the agent to execute complex shell commands and scripts. For instance, SKILL.md provides a Bash pipeline involving ls, sed, sort, and cut to aggregate chapter summaries. Additionally, references/style-profile-generator.md includes a Bash script that searches for Python interpreters to execute a dynamic Python script for linguistic analysis.
[REMOTE_CODE_EXECUTION]: The skill employs subagent orchestration (chapter-extractor). It programmatically calls subagents to process chapter data and checks for agent files in multiple local paths (.claude/agents/, .opencode/agents/, .codex/agents/), which is a form of dynamic loading from computed paths.
[DATA_EXPOSURE]: The skill performs extensive file system operations, reading novel files from user-provided paths and writing analyzed data into a project-specific directory structure (拆文库/{书名}/). These operations lack explicit input sanitization, potentially exposing the local environment to risks if the novel text contains malicious patterns.
[SAFE]: The inclusion of a GitHub source URL (worldwonderer/oh-story-claudecode) in the metadata is documented as a project reference and does not represent a direct threat on its own.

story-long-analyze