The Agent Skills Directory

[SAFE]: The skill follows a defensive execution model that prioritizes read-only diagnostics and requires explicit user authorization before performing any file writes or 'minimum manual fixes'.
[SAFE]: The instructions contain clear restrictions against silent software installation (e.g., '不静默安装工具'), preventing the agent from installing unverified dependencies without user knowledge.
[SAFE]: No evidence of credential harvesting, unauthorized network communication, or code obfuscation was detected across the instructions or configuration files.
[INDIRECT_PROMPT_INJECTION]: The skill has a potential attack surface for indirect prompt injection as it processes code from a repository that may contain malicious instructions.
Ingestion points: Ingests data from local .liquid files and diagnostic output from the theme-check command.
Boundary markers: Uses a structured output template for reporting findings, though it does not define specific delimiters for separating file content from instructions.
Capability inventory: Includes execution of the theme-check CLI and file system write access for applying code fixes.
Sanitization: Relies on a human-in-the-loop requirement to review diffs and provide confirmation before any changes are applied, effectively mitigating automated injection risks.

codeguard-liquid