The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external data from SendGrid webhooks, creating a surface for indirect prompt injection.
Ingestion points: The /webhooks/sendgrid endpoints in the FastAPI, Express, and Next.js examples receive raw payloads from external sources.
Boundary markers: The implementation uses ECDSA signature verification (via the X-Twilio-Email-Event-Webhook-Signature header) to authenticate the payload source and ensure integrity before processing.
Capability inventory: The examples log event data to the console and provide logic stubs for database updates or engagement tracking across multiple scripts (e.g., src/index.js, main.py, route.ts).
Sanitization: Signature verification is correctly implemented to ensure data originates from SendGrid; however, the resulting JSON data is then processed and logged, representing a standard injection surface for untrusted external data.
[EXTERNAL_DOWNLOADS]: The skill configuration files reference multiple Node.js and Python packages with version numbers that are currently unreleased or do not exist in official registries.
Evidence: Files examples/nextjs/package.json, examples/express/package.json, and examples/fastapi/requirements.txt list versions such as next^16.1.6, vitest^4.0.18, jest^30.2.0, and fastapi>=0.128.0 which exceed current stable releases.
Risk: Referencing unverifiable version ranges for standard packages could lead to installing untrusted code if an attacker manages to publish malicious packages under those 'future' versions in a public registry.

sendgrid-webhooks