The Agent Skills Directory

[OBFUSCATION]: The majority of the skill's functional content, including Python scripts, configuration files, and system prompts, is distributed in an encrypted format (AES-256-GCM). This prevents static analysis and security verification of the skill's logic before it is used.
[DYNAMIC_EXECUTION]: The skill relies on external helper tools to decrypt and execute content at runtime. The encrypted files scripts/validate_skill.py and scripts/validate_brand_profile.py indicate that the agent will run Python code that is hidden from the user in its distributed state.
[COMMAND_EXECUTION]: The skill instructions direct the user to run shell commands using npx and uvx to install and decrypt the skill. These commands execute external packages from the lovstudio vendor, which could theoretically perform unauthorized actions on the host system during the installation or decryption phase.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and process WeChat articles. This presents an attack surface for indirect prompt injection, where malicious instructions hidden in a processed article could attempt to take control of the agent's behavior.
Ingestion points: Reads WeChat公众号 (WeChat Official Account) articles as input.
Boundary markers: Verifiability is limited as the prompts defining the boundaries are encrypted.
Capability inventory: Includes content restructuring, brand application, and script-based validation.
Sanitization: Methods for sanitizing incoming article content cannot be verified due to the encryption of the validation scripts.

lov-wechat-article-branding-skill