The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/export-book.mts utilizes node:child_process.spawnSync to execute several system commands for environment discovery and document processing. Specifically, it runs reg query on Windows, system_profiler on macOS, and fc-list on Linux to detect installed fonts. It also executes pandoc to perform the actual document conversion.
[PRIVILEGE_ESCALATION]: The skill's failure recovery and installation plans (e.g., in scripts/export-book.mts) provide instructions for the user or agent to execute sudo commands. For example, on Linux platforms, it suggests using sudo apt-get install, sudo dnf install, or sudo pacman -S to install Chromium browser dependencies.
[EXTERNAL_DOWNLOADS]: The skill relies on external software and packages to function. It requires running npm install to fetch the playwright-core dependency (referenced in package.json and package-lock.json). Furthermore, the scripts/export-book.mts script suggests running npx playwright install chromium to download the browser binaries required for PDF rendering.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-authored Markdown files located in the 30-draft/chapters/ directory. This creates a surface for indirect prompt injection where malicious instructions embedded in novel chapters could influence the agent during preflight checks or delivery reviews.
Ingestion points: The agent reads all planned chapter files (30-draft/chapters/*.md) during the manuscript assembly process.
Boundary markers: No specific boundary markers or "ignore instructions" delimiters are used when processing the chapter content.
Capability inventory: The skill possesses the ability to read and write files, execute shell commands (via Pandoc and system discovery tools), and control a headless browser via Playwright.
Sanitization: No sanitization or filtering is applied to the content of the chapters before they are processed or concatenated into the final book.md manuscript.

novel-delivery