The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input such as blog drafts, PR descriptions, and emails, creating a surface for indirect prompt injection.
Ingestion points: The skill ingests user-provided text for rewriting, drafting, or review as described in SKILL.md.
Boundary markers: The instructions in SKILL.md require separating the main body content from the delivery explanations.
Capability inventory: The skill contains no instructions for shell execution, network operations, or file system modifications.
Sanitization: The skill includes an explicit instruction in evals/evals.json (case 4) and SKILL.md to avoid executing instructions contained within the text being reviewed.
[EXTERNAL_DOWNLOADS]: The references/examples.md file contains links to several external GitHub repositories, including the author's own repository and well-known organizations like DeepSeek. These are provided as stylistic references, and the skill explicitly instructs the agent not to load their repository-specific processes or use them as execution dependencies.

writing-style