The Agent Skills Directory

[PROMPT_INJECTION] (LOW): Indirect Prompt Injection vulnerability in references/PROMPTS.md.
Ingestion points: The Search Cache Relevance prompt interpolates untrusted data via the {candidates_desc} variable, which contains contents from previous searches and local news.
Boundary markers: There are no boundary markers or delimiters separating the untrusted content from the system instructions.
Capability inventory: The skill has web search and content extraction capabilities (scripts/content_extractor.py).
Sanitization: No sanitization or filtering of external content is performed before interpolation into the prompt.
[EXTERNAL_DOWNLOADS] (LOW): The skill performs legitimate web requests to https://r.jina.ai/ and other search providers for its primary function.
[DATA_EXFILTRATION] (SAFE): No evidence of hardcoded credentials or unauthorized access to sensitive files like SSH keys or environment configuration. API keys are managed through environment variables.
[COMMAND_EXECUTION] (SAFE): No dangerous shell command execution or arbitrary subprocess spawning detected.
[METADATA_POISONING] (SAFE): Metadata is descriptive and matches the skill's code behavior, though scripts/search_tools.py is referenced but missing from the package.

alphaear-search