The Agent Skills Directory

[SAFE]: The skill and its associated scripts communicate exclusively with LinkFox vendor domains (e.g., linkfox.com, tool-gateway.linkfox.com, api.linkfox.com, agent-api.linkfox.com) for trademark analysis, image hosting, and account onboarding.
[SAFE]: File system operations are limited to reading user-specified images for upload and writing session data/logs to a structured linkfox directory within the workspace or home directory, as described in the documentation. The scripts do not attempt to access sensitive system files.
[SAFE]: The onboarding process provides manual instructions for users to configure their own API keys in shell environment files (.zshrc, .bashrc), avoiding automated or stealthy persistence mechanisms.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data (image URLs and product titles).
Ingestion points: Untrusted data enters the agent context via imageUrl and productTitle parameters in SKILL.md and the scripts/ruiguan_trademark_graphic_detection.py script.
Boundary markers: The skill instructions explicitly direct the agent to warn users about high-risk matches (similarity > 0.8) and explain trademark statuses, providing contextual boundaries for processing results.
Capability inventory: Execution is limited to network requests to vendor APIs and file writes to the local linkfox directory.
Sanitization: The scripts use standard JSON serialization and HTTP libraries for data interpolation, minimizing the risk of command injection or prompt leakage during tool execution.

linkfox-ruiguan-trademark-graphic-detection