The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-generated story content, which could theoretically contain embedded instructions designed to hijack the agent's behavior.
Ingestion points: The agent reads story.md, chapter markdown files (specifically under ## Chapter Text), character entity files, and style-sheet.md to perform editing and voice analysis.
Boundary markers: The instructions mandate working paragraph-by-paragraph and following structured checklists provided in the references/ directory.
Capability inventory: The agent can execute shell commands via the story CLI for analysis and building narration/HTML copies, and can invoke system text-to-speech utilities.
Sanitization: The workflow requires the agent to propose every change with a before/after comparison and a rationale, ensuring the user reviews and accepts edits before they are applied to the markdown files.
[COMMAND_EXECUTION]: The skill heavily utilizes a custom story CLI tool and system-level audio utilities to perform its functions.
Evidence: Throughout SKILL.md, there are numerous calls to story passes, story prose, story voices, story build, and story wordcount for manuscript management.
Evidence: The skill uses command -v to detect and then utilize OS text-to-speech tools such as say (macOS), espeak-ng, or spd-say (Linux) to perform a read-aloud pass.
[EXTERNAL_DOWNLOADS]: The skill references external third-party software for rendering and accessibility features.
Evidence: Mentions external paged-media engines like Paged.js CLI, WeasyPrint, and Prince for generating PDF/print versions of the story.
Safety check: The instructions explicitly direct the agent to "ask before installing anything" when a required system tool or package is missing, maintaining user control over environment changes.

line-editing