The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted web data retrieved via the Perplexity API in ask.sh, research.sh, and fetch_url.sh. The ingestion of external, attacker-controllable content from the internet creates a surface for indirect prompt injection attacks, where malicious instructions embedded in web pages could influence the agent's behavior.\n
Ingestion points: Search results and raw page content are parsed and rendered by scripts/render_agent_response and scripts/render_search_response.\n
Boundary markers: The fetch_url.sh script includes system-level instructions for the model to quote content directly and avoid generating unverified information.\n
Capability inventory: The skill has the ability to make network calls, manage local cache files, and execute logic through shell and Python.\n
Sanitization: The skill uses Python's standard json library to safely handle and transform structured data from the API responses.\n- [DYNAMIC_EXECUTION]: The skill uses inline Python 3 scripts (embedded within shell scripts like common.sh, ask.sh, and research.sh) to perform logic that is difficult or unsafe to implement in pure shell, such as JSON parsing, configuration parsing, and date normalization. These scripts are limited in scope and utilize standard library functions.\n- [EXTERNAL_DOWNLOADS]: The skill makes network requests to the Perplexity API (api.perplexity.ai) to fulfill search and research queries. This is the core functionality of the skill and targets a well-known service.

perplexity-search