The Agent Skills Directory

[SAFE]: The skill instructions and scripts are focused on creative writing workflows and do not contain malicious patterns such as prompt injection, data exfiltration, or unauthorized command execution.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection because it reads and interpolates previously written story content and settings into the agent's context to maintain consistency. This is a standard operational requirement for long-form writing, and the data is sourced locally from the project's own files rather than untrusted external network sources.
Ingestion points: The agent reads from the bible/, outline/, state/, and chapters/ directories as part of its 'load on demand' protocol defined in SKILL.md.
Boundary markers: No specific delimiters or boundary instructions are used when reading file contents back into the prompt.
Capability inventory: The skill executes local Python scripts (novel_ops.py, wordcount.py) for directory scaffolding and progress tracking. It also leverages other internal agent skills for text polishing and condensation.
Sanitization: No explicit sanitization or filtering of narrative content is performed prior to context interpolation.

novel-writer