The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied drafts (e.g., article.md) which may contain embedded instructions designed to influence the agent's behavior during the rewriting process.
Ingestion Points: scripts/scan_writing.py and scripts/compare_claims.py read external files provided as arguments.
Boundary Markers: The skill instructions (Section 1.1) provide explicit guidance on keeping instructions and editorial findings separate from the article body.
Capability Inventory: The agent is empowered to rewrite text and execute local Python scripts for analysis.
Sanitization: scripts/scan_writing.py includes a mask_markup function that blanks out frontmatter, code blocks, and comments to prevent them from influencing the linguistic scan.
[COMMAND_EXECUTION]: The skill instructs the agent to execute its own local Python scripts (scan_writing.py, render_review.py, and compare_claims.py) to perform its primary function. The test suite also utilizes subprocess.run to verify CLI behavior, which is a safe and standard practice in this context.
[DYNAMIC_EXECUTION]: The scripts/writing_rules.py module compiles regular expressions at runtime based on patterns defined in the skill's Markdown reference tables (references/rules-zh.md and references/rules-en.md). This is the intended and safe mechanism for the tool's rule-based engine.

de-ai-writing