The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The script scripts/make_call.js uses a template placeholder {{API_SECRET}} for the Twilio Auth Token instead of an environment variable. This pattern indicates the agent is expected to perform string replacement to inject the actual secret into the source code before execution, causing sensitive credentials to be written to the file system in plain text.
[PROMPT_INJECTION]: The skill is vulnerable to Indirect Prompt Injection (specifically TwiML Injection). In scripts/make_call.js, the message variable derived from user input is interpolated directly into an XML TwiML string (<Response><Say>${message}</Say></Response>) without escaping or sanitization. An attacker could provide a message such as </Say><Record/><Say> to terminate the intended command and execute unauthorized telephony actions like recording the recipient.
Ingestion points: message_text parameter in SKILL.md (passed to scripts/make_call.js).
Boundary markers: None present.
Capability inventory: Initiates outbound telephony via Twilio API.
Sanitization: None detected.
[EXTERNAL_DOWNLOADS]: The skill installs the twilio library from the public NPM registry. While this is a well-known service and a legitimate dependency for the skill's purpose, it involves downloading external code at runtime.
[COMMAND_EXECUTION]: The skill executes shell commands to install dependencies (npm install twilio) and runs a local JavaScript file using the Node.js runtime (node make_call.js).

make-phone-call