The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection because it is designed to ingest and process untrusted external text from websites. However, the instructions include robust safety practices to mitigate this risk.
Ingestion points: Processes visible site text, components, and factual sources as described in SKILL.md and references/technical.md.
Boundary markers: The skill explicitly instructs the agent to separate prose from code, frontmatter, and JSX expressions, and to use delimited editing or compatible parsers to maintain structural integrity.
Capability inventory: The skill is capable of modifying files (editorial changes) and executing the astrofy check command.
Sanitization: Instructions emphasize preserving literal citations, URLs, and property names, and strictly forbid inventing facts or numbers that are not present in provided factual sources.
[COMMAND_EXECUTION]: The skill uses a project-specific CLI tool astrofy check --category editorial for validation. This command is consistent with the skill's stated purpose and author context.
[EXTERNAL_DOWNLOADS]: The skill references the official Astro documentation (docs.astro.build) and a public GitHub repository (github.com/miqdadbadjuber/anti-slop) for conceptual reference. These are well-known or contextually appropriate resources and do not involve remote code execution.

astrofy-editorial-review