The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted third-party data retrieved from TikTok, such as ad titles, advertiser names, and creative descriptions. This data is incorporated into the agent's context without sanitization or specific boundary delimiters, presenting a surface for indirect prompt injection.
Ingestion points: Payloads from the search and detail endpoints in scripts/chuhaijiang_ad_search.py, scripts/chuhaijiang_ad_detail.py, scripts/chuhaijiang_creative_search.py, and scripts/chuhaijiang_creative_detail.py.
Boundary markers: Absent; instructions do not mandate the use of delimiters for external content.
Capability inventory: Local filesystem writes for caching/logging in the linkfox/ directory and network requests to vendor APIs.
Sanitization: Absent; the display rules prioritize preserving raw returned values.
[COMMAND_EXECUTION]: The skill operates by executing Python scripts for all core functionalities. This requires the agent to have permission to run local scripts, which is the standard execution model for this skill.
[EXTERNAL_DOWNLOADS]: The scripts/onboarding.py utility references external Python packages requests, qrcode, and pillow. These are standard tools for API communication and QR code generation.

linkfox-chuhaijiang-tiktok-ads