The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes git subprocesses via scripts/input.ts to identify the repository root and discover changed files. These executions use safety best practices, such as spawn instead of a shell and the --end-of-options flag to prevent argument injection.
[DYNAMIC_EXECUTION]: The skill instructions involve executing its own bundled TypeScript logic (scripts/scan.ts) via bun or node to perform automated document analysis and linting.
[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted Markdown data from local files or standard input.
Ingestion points: External document content is read into the agent's context through scripts/input.ts.
Boundary markers: The SKILL.md provides explicit instructions for the agent to "Preserve Meaning" and "Protect Content," specifically identifying code fences, frontmatter, commands, and URLs as protected regions that must not be altered during editing.
Capability inventory: The skill has capabilities for file system reads/writes, git subprocess execution, and running its own TypeScript scanner scripts.
Sanitization: The scanner implementation in scripts/regions.ts uses masking logic to isolate protected technical content from prose analysis, reducing the risk that embedded instructions in technical blocks are misinterpreted as agent instructions.

prep-that-doc