The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (LOW): The skill instructs the user to install wrangler via npm and cloudflared via brew. While these are legitimate tools for the task, they are external dependencies installed at a global level.
[COMMAND_EXECUTION] (LOW): The skill is centered around executing shell commands to manage cloud infrastructure. The included scripts/check-usage.sh script automates the listing of deployments, databases, and buckets, which requires an authenticated environment.
[Indirect Prompt Injection] (LOW): The skill possesses a vulnerability surface for indirect prompt injection as it processes user-provided inputs into shell commands.
Ingestion points: User-provided project names, database names, and file paths in SKILL.md and check-usage.sh.
Boundary markers: Absent.
Capability inventory: Subprocess calls to wrangler and cloudflared are present throughout the skill.
Sanitization: Absent; the skill relies on the user to provide valid names.
[DATA_EXFILTRATION] (SAFE): No unauthorized data transmission patterns were found. The skill interacts with Cloudflare's official endpoints as expected for its stated purpose.
[PROMPT_INJECTION] (SAFE): The instructions are clear and do not contain patterns designed to bypass AI safety filters or hijack the agent's behavior.

cf-edge