The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains explicit instructions to override the agent's safety filters. It directs the agent not to refuse processing content based on "sensitivity" or "discomfort" (e.g., violence, domestic abuse, dark ethics), which is a common pattern for bypassing content safety guardrails.
[DYNAMIC_EXECUTION]: The skill utilizes dynamic code execution by generating a Python script at runtime via a shell heredoc. This script is then executed using the local Python interpreter (python3, python, or py) to calculate text statistics and sentence length distributions during the style analysis phase.
[COMMAND_EXECUTION]: The skill makes extensive use of shell commands to manage project files. It uses a pipeline of ls, sed, sort, cut, and cat to aggregate chapter summaries into a single file (_章节摘要汇总.md). It also uses grep for extracting metadata and searching for patterns across the generated analysis files.
[DATA_EXFILTRATION]: The skill probes sensitive internal configuration directories of the agent environment, such as .claude/agents/, .opencode/agents/, .codex/agents/, and .agents/agents/. While these checks are intended for sub-agent compatibility, accessing these paths exposes the internal structure and configuration of the agent's runtime environment.
[INDIRECT_PROMPT_INJECTION]: As an analysis tool for long-form fiction, the skill is designed to ingest large amounts of untrusted text from external files or user input. The skill lacks robust sanitization for this ingested data and has significant system capabilities (file writing and shell execution), creating a risk that malicious instructions hidden within novel text could influence subsequent analysis stages.

story-long-analyze