The Agent Skills Directory

[COMMAND_EXECUTION]: The workflow involves local shell commands to execute Python-based utilities stored within the skill's repository. In Phase 6.5 of SKILL.md and the 'Gate Steps' in references/delivery.md, the agent is instructed to run python3 scripts/blog_preflight.py, python3 scripts/generate_hero.py, and python3 scripts/blog_render.py for document rendering, hero image generation, and pre-delivery validation.
[EXTERNAL_DOWNLOADS]: The skill fetches visual assets from third-party image providers (Unsplash, Pixabay, Pexels, Openverse) and metadata from YouTube. To mitigate risks associated with network operations, the skill defines a strict 'URL and Asset Safety' protocol in references/delivery.md. This includes Server-Side Request Forgery (SSRF) protections such as resolving DNS to verify and reject loopback, private, and reserved IP ranges, as well as blacklisting dangerous protocols like javascript:, data:, and file:.
[INDIRECT_PROMPT_INJECTION]: The skill manages a surface for indirect prompt injection as it ingests untrusted data from the web (Phase 2: Research) to generate article content.
Ingestion points: WebSearch results, YouTube video metadata, and external statistics are incorporated into the agent context in Phase 2.
Boundary markers: None explicitly defined in the prompt instructions to separate research data from core instructions.
Capability inventory: The skill has the capability to write to the local file system and execute local Python scripts via python3.
Sanitization: The skill incorporates a mandatory 'Delivery Contract Enforcement' (Phase 6.5) that uses a blog-reviewer agent and specialized preflight scripts to validate content and ensure asset safety before the final output is delivered.

blog-write