The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides a command to automatically install the jose package globally using npm install -g if the library is not found. While functional, this represents an external software dependency that is fetched and installed during runtime.
[INDIRECT_PROMPT_INJECTION]: The skill uses shell templates that interpolate user-supplied data, such as secrets and claims, directly into command strings (e.g., JWT_SECRET='user-provided-secret'). This pattern is susceptible to command injection if the user-provided content contains shell operators or single quotes designed to break out of the variable assignment.
Ingestion points: User-provided JWT secrets, algorithm choices, and payload claims processed in SKILL.md.
Boundary markers: Absent. The skill does not define specific delimiters or instructions for the agent to isolate or sanitize user input before command interpolation.
Capability inventory: The skill utilizes node -e (JavaScript execution), python3 -c (Python execution), and openssl (file system writes and cryptographic operations).
Sanitization: Absent. There is no guidance for the agent to escape shell-sensitive characters in the user's secrets or claim values.
[COMMAND_EXECUTION]: The skill relies on shell-based execution of script snippets via node and python3 flags, as well as complex openssl piping for token signing. These methods bypass standard file-based execution and increase the difficulty of auditing the final executed code.

jwt-encode