Advisor Call Budget

•
19
•

Enforces hard call caps and drift checks when pairing cheap executor models with expensive advisor models to guarantee real cost savings.

Advisor Call Budget prevents cost overruns in multi-model architectures where a lightweight model executes tasks while consulting an expensive frontier model for guidance. Instead of relying on theoretical benchmark discounts, this skill introduces deterministic guardrails by establishing strict maximum advisor call limits, enforcing step-based drift checks to keep the executor on course, and programmatically calculating the effective price per token based on actual execution logs.

主な機能

01Configurable step intervals for executor drift detection
02Model-pair pricing audits using live token pricing data
03Automated effective cost computation from real call mixes
0419 GitHub stars
05Hard caps on advisor queries to prevent runaway API spend
06Deterministic verification script to validate budget enforcement

ユースケース

01Pairing small executor models with large reasoning models in multi-agent workflows
02Preventing silent cost creep when adopting benchmark-inspired orchestrator patterns
03Validating and enforcing actual cost savings during CI/CD test runs