The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill relies on standard third-party libraries including openai and requests for Python, and openai for TypeScript/Node.js, which are fetched from official registries (PyPI and NPM).
[DATA_EXFILTRATION]: The skill performs network requests to the openrouter.ai API. This is a well-known service, and the data transmission (prompts and message histories) is necessary for the skill's primary function of model routing.
[INDIRECT_PROMPT_INJECTION]: The routing logic utilizes heuristic classification of user prompts to select appropriate model tiers. While this creates a surface where specific input patterns could influence the selection of more expensive model tiers (e.g., triggering 'code' or 'analysis' markers to force a premium model), this behavior is central to the skill's routing function and does not represent a malicious exploit.
Ingestion points: Prompt text and message histories are ingested by classification functions such as classify_complexity (SKILL.md) and detect_task_type (references/implementation.md, references/intelligent-model-selection.md).
Boundary markers: None; classification is performed directly on the raw input content.
Capability inventory: The skill utilizes network capabilities to send requests to the OpenRouter API via the OpenAI SDK and requests library.
Sanitization: No input sanitization or validation is implemented before the heuristic classification logic is applied.

openrouter-model-routing