The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell_exec to run deployment commands for external services including netlify-cli, vercel, wrangler, and rsync. These operations are consistent with the skill's primary purpose of publishing rendered content to web hosts.
[SAFE]: Credentials and deployment tokens are managed securely by reading from environment variables (process.env) at runtime rather than being stored in the skill configuration.
[PROMPT_INJECTION]: The skill's blog post drafting feature uses project data as input, which constitutes a surface for indirect prompt injection. However, this is significantly mitigated by a required human review process before any generated content is published.
Ingestion points: Chapter summaries, user-model data, and craft-critic flags ingested via the POST /api/blog-posts/draft endpoint (SKILL.md).
Boundary markers: No specific delimiters or "ignore instructions" warnings are defined in the instructions for separating untrusted project data from the generation prompt.
Capability inventory: The skill possesses shell_exec and file_write permissions, which could be exploited if an indirect injection bypassed agent safety layers.
Sanitization: No programmatic sanitization is described, but the workflow enforces a "pending review" state, requiring manual author approval for all generated drafts.

website-publisher