The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data which could contain malicious instructions designed to influence the agent's behavior during the media generation process.
Ingestion points: User-supplied text for product descriptions, item advantages, and prices are interpolated into scripts and API payloads via templates like assets/templates/product-recommendation-script.md and references/script-templates.md.
Boundary markers: No boundary markers, delimiters, or explicit instructions to ignore embedded commands are present in the processing logic for user-provided variables.
Capability inventory: The skill possesses the capability to make network requests (requests.post/get in scripts/generate_audio.py and scripts/generate_image.py) and write files to the local filesystem (open().write() in multiple scripts).
Sanitization: There is no evidence of natural language filtering, escaping, or sanitization of the content before it is passed to script arguments or external APIs.
Multi-step chains: The skill is designed to consume output (prompts) from another skill (infinitetalk-shopping-avatar), creating a dependency chain where a compromise in the first skill could propagate to this one.
[DATA_EXFILTRATION]: The skill performs network operations that involve transmitting sensitive data to external endpoints.
Network operations: Scripts generate_audio.py and generate_image.py initiate POST requests to https://api.example.com. While example.com is a placeholder, the pattern demonstrates the transmission of an api_key in the Authorization header to a non-whitelisted domain.
[EXTERNAL_DOWNLOADS]: The skill dynamically downloads assets from the internet based on API responses.
Dynamic downloads: In scripts/generate_audio.py and scripts/generate_image.py, the code uses requests.get to download files from URLs provided by the API (audio_url and image_url) and saves them directly to the local file system without integrity verification.
[COMMAND_EXECUTION]: The skill's configuration executes a shell command at load time.
System dependency: The SKILL.md frontmatter contains a system dependency echo "Skill已加载", which executes a shell command to log status. While benign in this instance, it utilizes a powerful execution vector.

agentkit-multimedia-shopping