The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill incorporates untrusted user input into prompts sent to an external API, creating a potential surface for indirect prompt injection.
Ingestion points: User research queries are accepted as command-line arguments in lookup.py and research_lookup.py.
Boundary markers: The query is wrapped in double quotes within the _format_research_prompt method. This provides minimal protection as input containing quotes could potentially break out of the intended prompt structure.
Capability inventory: The skill is configured to use the Bash tool and performs outbound network requests to the OpenRouter API (openrouter.ai) using the requests library. No sensitive file access was observed.
Sanitization: The skill does not perform explicit sanitization, escaping, or validation of user-provided queries before they are interpolated into the research prompt.

research-lookup