The Agent Skills Directory

[SAFE]: The skill facilitates cost monitoring and budget enforcement using official OpenRouter API endpoints and standard Python/TypeScript libraries.
[DATA_EXPOSURE_&_EXFILTRATION]: Sensitive credentials such as OPENROUTER_API_KEY, OPENROUTER_MGMT_KEY, and SLACK_WEBHOOK_URL are correctly handled via environment variables. The skill communicates with well-known services (OpenRouter and Slack) to perform its monitoring and alerting functions.
[INDIRECT_PROMPT_INJECTION]: The skill provides middleware to track costs of user prompts.
Ingestion points: Prompts are ingested by chat completion wrappers in references/request-level-controls.md and references/budget-tracking.md before being sent to the LLM provider.
Boundary markers: None are implemented in the provided cost-tracking examples.
Capability inventory: The skill possesses network capabilities (via requests) and file writing capabilities (to cost_tracking.json).
Sanitization: No input sanitization is performed on the prompt content as the skill acts as a cost-auditing layer.
[PERSISTENCE]: The skill suggests using standard system tools like cron to schedule balance checks, which is appropriate for its stated purpose as an alerting utility.

openrouter-cost-controls