The Agent Skills Directory

[COMMAND_EXECUTION]: The Python utility scripts/cloudflare_deploy.py uses subprocess.run with list-based arguments to interact with the wrangler CLI. This approach prevents shell injection by ensuring that arguments are not interpreted as shell commands.
[EXTERNAL_DOWNLOADS]: Setup instructions in SKILL.md and references/gcloud-platform.md reference downloading the Google Cloud SDK via curl https://sdk.cloud.google.com | bash. This is a standard installation method for an official tool from a well-known service provider.
[PROMPT_INJECTION]: The skill documents workflows for web scraping and AI processing in references/browser-rendering.md, representing an indirect prompt injection surface.
Ingestion points: Untrusted data enters the context via page.content() in browser automation examples.
Boundary markers: The provided documentation examples do not include explicit delimiters or instructions to the AI model to ignore commands embedded within the scraped content.
Capability inventory: The skill includes powerful deployment capabilities via scripts/cloudflare_deploy.py and cloud management CLI tools.
Sanitization: No input sanitization or validation of the scraped web content is demonstrated in the processing examples.

devops