Runaway Execution and Cost Auditor

The problem

Automated agents and recursive workflows can easily trigger infinite loops, exponential fan-outs, or retry storms. Without hard execution boundaries, a single bug or edge case can result in massive API bills and exhausted infrastructure overnight.

What it does

• Analyzes execution graphs to identify unbounded recursion, loops, and parallel task explosions.
• Audits cost drivers across LLM tokens, third-party APIs, and compute resources.
• Evaluates the effectiveness of stop conditions, circuit breakers, and budget enforcement mechanisms.
• Simulates worst-case amplification scenarios to predict maximum potential spend.
• Provides a 100-point risk score and a prioritized remediation plan for safe production release.

Why this beats prompting it yourself

General prompts often miss the compounding effect of nested retries and concurrent child tasks. This skill uses a structured risk rubric and amplification failure catalog to find architectural vulnerabilities that simple code reviews overlook.

Use cases

• Reviewing autonomous agent loops before deploying to production environments.
• Auditing high-fan-out data processing pipelines for cost and queue safety.
• Validating budget controls and rate limits for multi-tenant AI applications.
• Assessing retry policies in distributed systems to prevent cascading failures.

Known limitations

This is a static audit tool. It cannot execute code, call live APIs, or provide guaranteed upper-bound cost certifications without explicit spend path data.