The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates indirect prompt injection by design. It instructs the agent to fetch content from arbitrary external URLs and then perform editorial tasks on that content.
Ingestion points: The scripts/import_article.py script described in SKILL.md fetches content from user-provided URLs.
Boundary markers: The skill mentions scoping CSS under .sr-article to prevent style leakage, but does not specify delimiters or instructions for the agent to ignore embedded commands within the markdown body during the editorial pass.
Capability inventory: The skill uses python3 subprocesses for building and importing, file system writes for drafts and builds, and network access to fetch article content.
Sanitization: The importer performs HTML neutralization and typography normalization as described in the "Formatting imported / rich articles" section.
[EXTERNAL_DOWNLOADS]: The skill utilizes a local script to download content from external web pages to create drafts.
Evidence: The instruction python3 scripts/import_article.py "<URL>" in SKILL.md allows the agent to download resources from arbitrary remote sources.
[COMMAND_EXECUTION]: The skill relies on the agent executing shell commands to run its build and import scripts.
Evidence: SKILL.md provides multiple examples of shell execution, such as python3 build.py and python3 scripts/import_article.py. If the agent fails to properly sanitize the URL argument, it could potentially lead to shell command injection.

starchild-blog