The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/run_lint.py script executes the languagetool and vale command-line tools using subprocess.run(). This is the primary mechanism for the skill's "Layer 0" proofreading. The commands are constructed using argument lists rather than shell strings, and they target manuscript files provided by the user.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted manuscript content provided by the user. While the primary analysis is performed by local tools, the agent is subsequently tasked with reviewing the findings ("Layer 2").
Ingestion points: Markdown files (.md) passed to scripts/run_lint.py and scripts/normalize_typography.py.
Boundary markers: The instructions explicitly separate findings into "AUTO" (mechanical) and "REVIEW" (judgment), directing the agent to only auto-apply mechanical fixes.
Capability inventory: The skill has the ability to read and write files within the project directory and execute specific local CLI tools via subprocess.
Sanitization: scripts/run_lint.py includes a strip_md function that removes markdown formatting and structural elements before passing text to the LanguageTool engine.

manuscript-finisher