The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to ingest and process external webhook payloads (e.g., from Assistable or GoHighLevel) which serve as arguments for agent tool calls. This processing of untrusted data represents an indirect prompt injection surface.
Ingestion points: Webhook data located in body.args and request headers such as X-Subaccount-Id (File: guides/01-parse-webhook.md).
Boundary markers: The skill focuses on structured data parsing and does not explicitly define prompt-level delimiters to isolate untrusted data.
Capability inventory: Tool handlers wired by the skill are capable of performing stateful side effects, including writes to GoHighLevel contacts and data persistence in Supabase (Files: guides/03-tool-call-wiring.md, guides/04-routing-boundaries.md).
Sanitization: The skill significantly mitigates this surface by requiring HMAC signature verification and strict shape validation before any handler logic is executed (File: guides/02-validate-webhook.md).
[SAFE]: The skill incorporates strong credential safety protocols. Directive 5 (File: guides/00-principles.md) explicitly mandates that secrets such as API keys, signing secrets, and sub-account tokens must be stored in environment variables and never logged or committed to the repository.
[SAFE]: External dependencies and service references involve well-known technology platforms (GoHighLevel, Inngest, Supabase) and official documentation, which are handled appropriately for the skill's stated purpose.

assistable-sms-weapon