The Agent Skills Directory

[SAFE]: The skill implements a robust human-in-the-loop interaction model. Instructions explicitly mandate a 'Confirmation Format' (Yes/Tweak/Rewrite) at every stage of the process, ensuring that no content is persisted to the file system or used in subsequent drafting steps without direct user verification.
[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection as it ingests user-supplied novel ideas, themes, and character details which are then interpolated into prompts for sub-agents using the delegate_task tool.
Ingestion points: Initial project setup (Step 1), Interactive Braindump (Step 3), and Character Building (Step 6) in SKILL.md.
Boundary markers: The skill uses a mandatory confirmation cycle for every individual section, serving as a significant boundary against unintended instruction execution.
Capability inventory: Uses delegate_task for parallel drafting/review and performs file-write operations within the user-specified writing directory.
Sanitization: Content is reviewed by the user at each step; no programmatic sanitization is applied, which is standard for creative writing applications. The risk is considered negligible due to the intended primary purpose of the skill and the iterative approval process.
[COMMAND_EXECUTION]: The skill utilizes background task delegation (delegate_task with run_in_background=true) to manage the computational overhead of drafting and reviewing multiple book chapters simultaneously. This orchestration is confined to the skill's creative writing tasks and follows the platform's standard operational patterns.

novel-architect