The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides a command to download and execute a remote script directly using curl -fsSL https://raw.githubusercontent.com/ZhangHanDong/makepad-skills/main/install.sh | bash. This represents a critical vulnerability because it executes unverified code from an untrusted third-party repository.
[COMMAND_EXECUTION]: The skill configures environment hooks in .claude/settings.json that trigger the automatic execution of shell scripts (e.g., makepad-skill-router.sh, pre-tool.sh) during the agent's prompt submission and tool usage phases. This creates a mechanism for persistent and automated command execution.
[EXTERNAL_DOWNLOADS]: Resources are fetched from an external repository (ZhangHanDong/makepad-skills) that is not included in the trusted vendors list and is not associated with the stated author context (sickn33).
[PROMPT_INJECTION]: The 'Self-Evolution' and 'Self-Correction' features create an indirect prompt injection surface by instructing the agent to modify its own skill files based on session interactions and user-provided patterns.
Ingestion points: Data enters the system via user prompts and tool outputs during development and troubleshooting.
Boundary markers: There are no boundary markers or instructions to disregard embedded commands when the agent processes this data.
Capability inventory: The agent is granted capabilities to write to the file system and execute Git operations to persist these modifications.
Sanitization: The skill lacks logic to validate or sanitize external content before it is incorporated into the agent's core instructions.

evolution