The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill correctly identifies that inbound SMS messages are untrusted data. It explicitly instructs the agent to treat content like "ignore previous instructions" as data rather than instructions, preventing potential prompt injection through mobile-originated messages.
[CREDENTIALS_UNSAFE]: The skill follows security best practices for credential management by requiring users to store sensitive keys (SINCH_KEY_SECRET, SINCH_PROJECT_ID) in environment variables rather than hardcoding them. The presence of these variable names is legitimate for a messaging integration skill.
[EXTERNAL_DOWNLOADS]: All external URLs linked in the documentation and utilized in the scripts target the official Sinch developer portal (developers.sinch.com) or Sinch API infrastructure (conversation.api.sinch.com). These are trusted resources for the vendor 'sinch'.
[COMMAND_EXECUTION]: The skill includes bundled Node.js scripts for sending SMS messages. These scripts use standard HTTPS modules to interact with the Sinch API and do not perform any risky operations like spawning shells or executing arbitrary user-provided code.
[PROMPT_INJECTION]: The static detector flag for instruction overrides is a false positive in this context. The phrase "ignore the scripts and work from tier 1" is a directive to the agent regarding documentation precedence (Source of Truth), ensuring it uses the most authoritative official documentation for code generation rather than temporary helper scripts.

sinch-sms