The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data fetched from external web sources via the Brave Search API.
Ingestion points: scripts/brave_llm_context.py and scripts/brave_image_search.py retrieve and display content (snippets, page text, and captions) from the live web.
Boundary markers: The scripts use simple text headers (e.g., '--- Source N: ... ---') to separate results, but do not employ robust 'ignore instructions' delimiters for the extracted content.
Capability inventory: The skill utilizes Bash and Read tools to execute Python scripts and manage files, providing a path for potential follow-on actions if the agent obeys instructions found in search results.
Sanitization: Content is output with basic encoding handling (UTF-8/ASCII-safe), but there is no semantic filtering of instructions within the fetched data.
[EXTERNAL_DOWNLOADS]: The brave_llm_context.py script requires the requests Python library to be installed, which is a standard external dependency.

brave-search