The Agent Skills Directory

[SAFE]: The skill is designed to manage local markdown files within a novel project directory. It establishes clear boundaries and coordination protocols with other assistant skills like novel-lite and novel-review.
[PROMPT_INJECTION]: A thorough scan for bypass markers, 'Ignore previous instructions', or role-play injections returned no findings. The instructions are focused on guiding the user through a multi-step 'cold start' and project maintenance process.
[DATA_EXFILTRATION]: No sensitive file paths (e.g., .ssh, .env, .aws/credentials) or network operations (curl, wget, fetch) were identified. The skill only interacts with the user-defined novel project directory.
[REMOTE_CODE_EXECUTION]: There are no patterns suggesting the download or execution of external scripts. The skill relies on natural language interaction and the generation of markdown templates.
[INDIRECT_PROMPT_INJECTION]: The skill includes an ingestion point where it processes 'reference novels' provided by the user to extract writing styles. While this presents a surface for indirect prompt injection, the skill uses structured markdown templates and delegatory logic (via novel-style-extract) to mitigate accidental instruction following. Evidence is found in references/reference-冷启动流程.md where the process for handling user-provided text is described.

novel-setup