foundations-theory-of-constraints
Installation
SKILL.md
Theory of Constraints Foundations
When to Apply
Apply theory-of-constraints when:
- One bottleneck demonstrably gates total system throughput (the system has a constraint, not many)
- Roadmap or capacity-allocation under a hard limited resource (eng-weeks, GPU-hours, account-managers)
- Funnel debug where a single step blocks downstream conversion
- Policy constraint suspected (a rule, not a physical limit, is what's holding throughput)
- Subordination question — "should other steps slow down to match the bottleneck?"
- Post-AI adoption reassessment — when AI coding tools improve individual velocity but delivery metrics stay flat (DORA's throughput group: change lead time, deployment frequency, failed-deployment recovery time; instability group: change fail rate, deployment rework rate), re-run 5FS. DORA 2024 associated AI adoption with lower delivery throughput and lower stability; DORA 2025 (n≈5,000) found throughput positive and stability still negative, and replaced its four performance tiers with seven team archetypes. The sign of the throughput effect changed between years, so treat any single year as unstable. That the constraint moved to code review, verification or integration is this skill's own inference, not a DORA finding; verify it with 5FS
- LLM / agent-pipeline optimization — when end-to-end latency or task throughput of a multi-step AI pipeline is not meeting targets despite adding models or workers; the constraint is usually a specific stage (LLM decode, serialized tool execution, or a guardrail/eval step), not aggregate capacity — profile per stage before scaling