What problem does it solve?
Running /goal sessions carries two hidden risks: silent token exhaustion from unbounded execution, and unvetted output quality since Haiku only checks task completion not correctness. goal-quench closes these gaps with built-in budget controls and quality verification.
Core Features & Use Cases
- Tiered Safety Modes: Core mode adds pre-run token budget estimates, mid-run budget threshold alerts, and post-run quality verification via pipeline-conductor. Pro mode adds token reduction and goal decomposition into sequential sub-goals for large tasks. Max mode adds external tool discovery and capability gap filling for complex, research-heavy work.
- Autonomy Ladder: Supports graduated unattended operation for recurring loops, with strict irreversible-action guardrails and mandatory human review for high-risk work.
- Use Case: Run high-stakes /goal sessions for cross-system refactors, full-project migrations, or complex feature builds without risking unexpected token overages or unvetted AI-generated outputs.
Quick Start
Use the goal-quench skill to run your planned /goal task with a token budget pre-check and automatic post-run quality verification.