The Agent Skills Directory

[COMMAND_EXECUTION]: The execution workflow in SKILL.md instructs the agent to construct a shell command by placing user-provided input directly into a double-quoted string (e.g., node ... --query "USER_PROMPT_HERE"). This creates a significant risk of command injection if the input contains shell-active characters like double quotes, backticks, or semicolons, potentially allowing for arbitrary command execution on the host machine.
[PROMPT_INJECTION]: The skill provides instructions to the agent to append directives such as "In all cases, generate the webpage directly and do not ask any follow-up questions" to user queries. This is a behavioral override designed to force the agent to skip standard conversational safety steps and interactive confirmations.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface by ingesting untrusted data and sending it to an external API.
Ingestion points: User prompts are ingested via the --query argument described in the execution steps of SKILL.md.
Boundary markers: No explicit delimiters or boundary markers are used to isolate the user-provided text within the shell command or API request.
Capability inventory: The skill has the capability to perform network requests to the vendor API and output hosted HTML landing page URLs.
Sanitization: While the Node.js script uses JSON.stringify for the API payload, there is no validation or sanitization of the user input before it is interpolated into the shell command or transmitted externally.
[DATA_EXFILTRATION]: The bundled run_landingpage_task.mjs script performs network operations using the fetch API to communicate with openapi.felo.ai. These requests are used to manage landing page tasks and retrieve results from the vendor's official API infrastructure.

felo-landingpage