The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it ingests data from local files created during the project lifecycle to guide the writing process.
Ingestion points: The agent reads from user-preferences.json, 00-人物档案.md (character profiles), and 01-大纲.md (story outlines) to generate chapter content.
Boundary markers: The skill lacks explicit delimiters to distinguish between its internal instructions and the story data ingested from these files.
Capability inventory: The skill can execute local shell commands (Python script) and spawn autonomous sub-agents.
Sanitization: Data from outline and preference files is used directly in prompts without filtering or escaping.
[DYNAMIC_EXECUTION]: The skill uses a provided Python script (scripts/check_chapter_wordcount.py) to automate the verification of Chinese character counts. This script is executed locally via the command line as part of the validation workflow.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill utilizes the host platform's native sub-agent and team collaboration tools to parallelize the writing of different chapters. The logic for these agents is defined through prompts stored in the skill's reference files (references/flows/phase3-writing.md).

chinese-novelist