The Agent Skills Directory

[OBFUSCATION]: The skill's primary logic is contained within an encrypted file (SKILL.md.enc) using the AES-256-GCM cipher as detailed in MANIFEST.enc.json. This implementation prevents static analysis of the agent's underlying instructions.
[DYNAMIC_EXECUTION]: The skill instructions specify that the content is decrypted in-memory at runtime to be used by the agent, without writing the plaintext source code to the disk. This creates a dynamic execution environment where logic is loaded from an opaque source.
[REMOTE_CODE_EXECUTION]: The skill installation and operation instructions direct the user to run external tools via npx (lovstudio) and uvx (lovstudio-skill-helper). These commands download and execute remote code to manage the skill lifecycle.
[EXTERNAL_DOWNLOADS]: The skill requires the download of an encrypted distribution package and external helper tools from the vendor's repositories to function.
[TIME_DELAYED_CONDITIONAL]: The decryption of the skill's logic is conditional upon user login and the expenditure of credits, meaning the actual behavior of the skill may change or only become apparent after a transaction or authentication event.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided project requirements to generate proposals in SKILL.md, creating an ingestion point for untrusted data. There are no boundary markers or explicit sanitization steps documented for this input, which combined with the skill's dynamic execution capabilities, represents a surface for potential injection.

lov-proposal