The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The environment setup script (setup_env.py) downloads the official installer for the uv package manager from a well-known service (astral.sh) and executes it using a shell command. This is a routine procedure for setting up the required development environment for the skill's features.
[COMMAND_EXECUTION]: The skill utilizes subprocess.run to automate various tasks, such as creating virtual environments, installing dependencies, and invoking ffmpeg for audio file processing in audio_utils.py. These operations are strictly scoped to the skill's intended functionality of environment management and exercise generation.
[EXTERNAL_DOWNLOADS]: The skill downloads and installs several reputable Python packages (edge-tts, genanki, pydub, kokoro) from standard registries to enable its speech synthesis and flashcard features. These dependencies originate from established projects and well-known services.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process user-provided content, such as written and spoken TOEFL answers, for automated evaluation, which creates a potential surface for indirect prompt injection.
Ingestion points: User-provided answers are ingested and processed by evaluation scripts like writing_grader.py and speaking_coach.py.
Boundary markers: The system provides the agent with explicit evaluation rubrics and feedback templates in references/scoring-rubrics.md to guide and constrain its assessment behavior.
Capability inventory: The skill is capable of local file operations within its workspace, generating audio assets, and producing markdown-based progress reports.
Sanitization: Evaluation prompts incorporate structured scoring criteria and specific linguistic instructions to ensure the agent focuses on performance analysis rather than potential adversarial instructions embedded in user content.

toefl