The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple local Python scripts as part of its automated workflow. Evidence in SKILL.md and references/delivery.md shows the use of python3 to run scripts/blog_preflight.py, scripts/blog_render.py, and scripts/generate_hero.py for task validation, file format conversion, and image generation.
[INDIRECT_PROMPT_INJECTION]: The skill has a large ingestion surface for processing untrusted external data which could contain malicious instructions.
Ingestion points: The skill performs broad web searches for statistics (Phase 2), queries YouTube for video content, and processes metadata from third-party image APIs (Pixabay, Unsplash, Pexels). It also optionally ingests content from user-uploaded documents via blog-notebooklm.
Boundary markers: While the skill uses summary boxes and specific citation formats for output, it does not specify the use of delimiters or 'ignore embedded instructions' warnings for the raw data ingested during the research phase.
Capability inventory: The agent has the ability to write files to the local system, execute shell commands (via the mentioned Python scripts), and delegate tasks to other sub-agents (blog-researcher, blog-reviewer).
Sanitization: The skill contains robust URL and asset safety guidelines in references/delivery.md, specifically implementing SSRF protections by blacklisting loopback, private, link-local, and multicast IP ranges, and enforcing protocol restrictions (HTTP/HTTPS only).

blog-write