langfuse-cost-tuning
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill uses well-known Node.js packages (@langfuse/client, @langfuse/openai, @langfuse/tracing, and langfuse) for LLM observability and cost tracking, which is consistent with its stated purpose.
- [SAFE]: Network operations, specifically sending cost alerts to a Slack webhook, utilize user-provided environment variables (process.env.SLACK_WEBHOOK_URL) rather than hardcoded secrets, adhering to security best practices.
- [SAFE]: No malicious patterns such as prompt injection, obfuscation, or persistence mechanisms were detected. The instruction logic for model routing and cost reporting is transparent and aligns with the tool's intended utility.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests task descriptions to perform model routing. While this represents a minimal attack surface, the processing is limited to string matching and model selection with no execution of untrusted code or commands. Evidence: Ingestion points in SKILL.md (selectModel) and references/implementation.md (CostOptimizedModelSelector); boundary markers and sanitization are absent; capability inventory includes Langfuse API calls and model parameter selection.
Audit Metadata