The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents how to handle external data sources (HTTP requests, email handlers, and message queues) which represent an inherent attack surface for indirect prompt injection if the agent processes untrusted data. However, the skill emphasizes security best practices such as parameterized SQL queries with D1 to prevent standard injection attacks.
Ingestion points: fetch() handlers (01-workers-fundamentals.md), email() handlers (01), and queue() consumers (07).
Boundary markers: Standard delimiters are used in code examples, though explicit instructions for the agent to ignore embedded instructions in data are not present.
Capability inventory: The skill facilitates command execution via the Wrangler CLI (wrangler deploy, wrangler d1 execute), network operations (fetch), and file system access (project scaffolding).
Sanitization: Code examples demonstrate the use of prepared statements (env.DB.prepare) for database interactions to prevent SQL injection.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of official tools and libraries from the Cloudflare ecosystem and well-known registries. These include wrangler, @cloudflare/workers-types, hono, and vitest. All downloads originate from trusted sources or well-known public registries (npm).
[COMMAND_EXECUTION]: The skill provides instructions for using the wrangler CLI to manage Cloudflare services, including deployment and database management. These commands are standard for the platform's operation.
[DYNAMIC_EXECUTION]: The skill includes a maintenance script scripts/check-updates.py that interacts with the npm registry and local file system to ensure the documentation remains current. The script uses standard Python libraries and performs safe integrity checks.

cloudflare-workers