The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (LOW): The skill instructions in SKILL.md and references/gcloud-platform.md recommend installing the Google Cloud SDK via curl https://sdk.cloud.google.com | bash. While piping remote scripts to a shell is a high-risk pattern, the source is a Trusted Organization (Google), resulting in a downgrade to LOW per [TRUST-SCOPE-RULE].
[EXTERNAL_DOWNLOADS] (LOW): The skill references and downloads tools/binaries (Wrangler, Docker, GCloud CLI) from reputable and trusted repositories. These are standard for DevOps workflows.
[PROMPT_INJECTION] (LOW): The skill contains an indirect prompt injection surface where the agent processes untrusted data from external websites.
Ingestion points: page.content() in references/browser-rendering.md and references/cloudflare-workers-advanced.md (AI-powered scraper examples).
Boundary markers: Absent in code examples; no explicit delimiters or instructions to ignore embedded commands are used when passing scraped content to LLMs.
Capability inventory: The skill possesses network access (fetch), database access (D1/KV), and local command execution capabilities via scripts/cloudflare_deploy.py.
Sanitization: The best practices section mentions URL validation and content sanitization, but the specific code examples do not implement these protections.
[COMMAND_EXECUTION] (SAFE): The scripts/cloudflare_deploy.py utility uses subprocess.run to execute the wrangler CLI. This is consistent with the primary purpose of the skill and is handled safely using static command lists.

devops