The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied text by interpolating it directly into prompts sent to the external each::sense AI agent. This creates a vulnerability to indirect prompt injection, where instructions hidden within the input text (e.g., instructions to change voice style mid-sentence or override output formatting) could influence the agent's behavior.
Ingestion points: SKILL.md (specifically the prompt templates in the examples and the $TEXT variable in the batch script).
Boundary markers: None present; text is appended or interpolated without delimiters or 'ignore' instructions.
Capability inventory: Performs network operations via Bash(curl) to the eachlabs.run API.
Sanitization: No validation or escaping of input text is implemented before sending it to the synthesis service.
[COMMAND_EXECUTION]: The documentation provides multiple bash code blocks that use curl to make POST requests to the vendor's API. While these are provided as usage examples, they involve executing shell commands that interact with external network resources.
[EXTERNAL_DOWNLOADS]: The skill communicates with eachsense-agent.core.eachlabs.run and references documentation at eachlabs.ai. These domains are associated with the skill's primary functionality and its vendor, each::labs.

text-to-speech