The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user prompts which are then passed to a local execution script and a remote API. There are no explicit boundary markers or sanitization steps to prevent instructions hidden within the prompt from being interpreted as commands by the agent.
Ingestion points: User prompts passed to scripts/generate.mjs via the --prompt argument as described in SKILL.md.
Boundary markers: Absent. The instructions do not specify the use of delimiters for user-provided data.
Capability inventory: The generate.mjs script performs network requests (fetch) and file system writes (fs.writeFileSync).
Sanitization: Absent. Input is used directly in the API request body and command arguments.
[COMMAND_EXECUTION]: The skill relies on the AI agent executing a local script (scripts/generate.mjs) with user-supplied strings. This creates a potential command injection surface if the execution environment uses a shell and does not properly escape the --prompt value.

freeimg