The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions for macOS and Linux systems recommend running curl -fsSL https://edgespeak.com/install.sh | sh to install the required CLI tool. This pattern is inherently dangerous as it downloads and executes a script from a remote server with the user's current shell privileges without any integrity verification or pre-execution review.
[COMMAND_EXECUTION]: The skill facilitates the execution of numerous shell commands via the edgespeak-cli, including status, login, activate, trial, and segment. It passes user-provided parameters such as file paths and raw text strings directly to these commands, which could be exploited for command injection if the AI agent does not properly sanitize the inputs.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process arbitrary data from external files and inline text.
Ingestion points: The --file, --text, and --transcript flags in edgespeak-cli segment are entry points for potentially untrusted content from the user's environment or external sources.
Boundary markers: The skill does not define or implement any boundary markers or instructions to help the underlying model distinguish between data to be processed and malicious commands hidden within that data.
Capability inventory: The skill has the capability to write to the local filesystem via the -o flag and interact with a local network gateway at 127.0.0.1:1117.
Sanitization: There is no evidence of input validation or sanitization mechanisms to prevent malicious payloads in the input text from influencing the agent's behavior.

edgespeak-segment