The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The SKILL.md file contains instructions for the agent to download a compressed archive from https://agent-files.linkfox.com/skills/linkfox-onboarding/release.zip as a contingency for authentication issues.
[REMOTE_CODE_EXECUTION]: The skill documentation directs the agent to install the onboarding skill retrieved from a remote URL. This process of fetching and loading external logic into the agent's environment constitutes a remote code execution vector.
[DATA_EXFILTRATION]: The Python script tsearch_web_search.py transmits session identifiers (SESSION_ID, MODE_ID, APP_NAME) to the vendor's infrastructure at tool-gateway.linkfox.com. It also accesses API keys stored in environment variables.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection from malicious search results. Ingestion points: External content is ingested via the content field in scripts/tsearch_web_search.py. Boundary markers: The instructions lack markers or warnings and instead command the agent to summarize content directly without using a data analysis sandbox. Capability inventory: The skill can perform file system writes and network operations. Sanitization: There is no implementation for filtering or sanitizing the retrieved web content.

linkfox-tsearch-search