The Agent Skills Directory

[DATA_EXFILTRATION]: The skill's instructions in SKILL.md direct the agent to read personal information (including biography and social media handles) from ~/.claude/memory/personal-asset-index.json. While this data is used for personalizing designs, accessing files within the agent's internal memory directory is a sensitive operation that can lead to unintended exposure of session-related data.
[COMMAND_EXECUTION]: The skill frequently executes external commands for media processing and design validation. Scripts such as scripts/render-video.js, scripts/verify.py, and scripts/html2pptx.js utilize playwright to automate browser tasks, while ffmpeg is used for video encoding and audio mixing. These operations are consistent with the skill's primary objective of visual production.
[EXTERNAL_DOWNLOADS]: As part of the 'Core Asset Protocol', the skill uses curl to download logos and product images from user-specified brand websites. This automated retrieval of remote content is designed to facilitate high-fidelity design work by using official brand assets.
[PROMPT_INJECTION]: The skill reads and processes external HTML content from brand websites using curl and grep to extract design tokens like color codes and SVG elements. This creates an indirect prompt injection surface where a malicious website could attempt to influence agent behavior through its HTML source, although risks are mitigated by the specific extraction patterns used.

huashu-design