The Agent Skills Directory

[PROMPT_INJECTION] (LOW): The skill is vulnerable to indirect prompt injection because it processes untrusted user-provided text and instructions to generate audio via an external API.\n
Ingestion points: SKILL.md and references/cli.md describe taking verbatim text and delivery instructions directly from users or temporary JSONL batch files.\n
Boundary markers: Absent. There are no delimiters or specific system instructions to isolate user-provided content from the synthesis engine's control parameters.\n
Capability inventory: The skill executes a Python CLI (scripts/text_to_speech.py), performs outbound network requests to the OpenAI API, and writes audio artifacts to the local file system.\n
Sanitization: Absent. No evidence of validation, escaping, or filtering of the input text is provided in the skill documentation.\n- [EXTERNAL_DOWNLOADS] (LOW): The skill requires the installation of the openai Python package from a remote registry.\n
Evidence: SKILL.md instructs users to install openai via uv or pip. The severity is downgraded to LOW/INFO because the package is maintained by a trusted organization (openai).\n- [COMMAND_EXECUTION] (SAFE): The skill runs a bundled CLI script to perform its primary tasks.\n
Evidence: references/cli.md provides usage examples for scripts/text_to_speech.py. This is standard for skill-based task automation. Note: the source of the Python script was not provided for detailed code-level audit.\n- [SAFE] (SAFE): The skill follows secure credential handling patterns.\n
Evidence: SKILL.md explicitly warns users against pasting the OPENAI_API_KEY into the chat and instead provides instructions for setting it as a local environment variable.

speech