agentleFS
deciqAIdeciqAI/knowledge-skills/theory-of-constraints/SKILL.md

theory-of-constraints

Skill13 starsChanged 4 months agoMITView on GitHub

Activate when: user says 'everyone is working hard but results are flat', 'where is our bottleneck', 'we keep adding capacity but throughput doesn't improve', 'backlog piling up at one stage', 'Goldratt / TOC / Five Focusing Steps', or is designing a process-improvement initiative and wants to know where to invest.

3 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Check fit: single-step system or demand-side constraint → TOC doesn't apply.

    line 33
  2. Distinctive:

    Run The Process one step at a time: constraint?

    line 36
  3. Distinctive:

    Step 4 — Elevate: if still binding after Steps 2-3, add capacity at the constraint (equipment, people, redesign).

    line 46

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Theory of Constraints

Overview

Theory of Constraints (TOC) — Eliyahu Goldratt, 1984: throughput of any multi-step system is determined by its single bottleneck. Improving any other step produces no system-level gain. The Five Focusing Steps (Identify → Exploit → Subordinate → Elevate → Repeat) are the operational discipline.

Composes with pareto-principle (TOC = Pareto applied to throughput), feedback-loops, first-principles, and mvp (MVP design = TOC applied to validated learning).

When to Use

  • System is producing less than desired throughput; "everyone is working hard" but results don't match effort
  • Management improvement initiative or capacity investment is being planned
  • Backlog or inventory accumulates at a specific step; local improvements don't translate system-wide
  • Someone says "bottleneck," "throughput," "Goldratt," or "Theory of Constraints"
  • Analyzing an AI capex / chip-supply-chain question — where the real limit is (e.g. GPU design vs. advanced packaging, HBM, or grid power), whether the "AI bubble" reflects a design race or a hidden physical bottleneck

Not when: single-step system; purely demand-side constraint; problem is strategic/psychological, not operational.

Coaching Novices (Adaptive Front Door)

  • Engine mode: user has a concrete throughput problem → run The Process directly.
  • Coach mode: user is unfamiliar → guide step by step.

In Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output only that step's question, then stop.

  1. One-line: throughput is set by the slowest step — find it, fix it, ignore the rest until a new bottleneck emerges.
  2. Check fit: single-step system or demand-side constraint → TOC doesn't apply.
  3. Elicit their real case: what's the system? desired throughput? where does work-in-process pile up?

[WAIT — do not advance until user responds]

  1. Run The Process one step at a time: constraint? exploiting it? subordinating everything else? elevate?

[WAIT — do not advance until user responds]

  1. Close by naming the constraint, action plan, and re-identification schedule.

[WAIT — do not advance until user responds]

The Process

Step 1 — Identify: map all steps with capacity; find where WIP accumulates — that's the constraint. Step 2 — Exploit: max output from the constraint with no new investment (eliminate idle time, defects, distractions at that step). Step 3 — Subordinate: pace all other steps to the constraint's rate. Upstream: don't over-produce. Downstream: don't block. Retire local efficiency metrics that incentivize over-production. Step 4 — Elevate: if still binding after Steps 2-3, add capacity at the constraint (equipment, people, redesign). Highest ROI investment in the system. Step 5 — Repeat: bottleneck has moved. Return to Step 1.

Output: TOC Analysis

# TOC Analysis: <system>
## System map — steps, capacity per step, actual throughput, where WIP accumulates
## Constraint — bottleneck step + evidence (WIP buildup, idle downstream, output rate match)
## Exploit — changes to maximize current constraint output (no new investment)
## Subordinate — upstream rate limits, downstream coordination, metric changes, buffer plan
## Elevate — capacity investment at constraint, cost/benefit
## Re-identification — what to monitor, likely next constraint, re-apply schedule

→ Method in Action: Goldratt's The Goal (1984) and TOC's Lineage · Critical Chain Project Management (1997) → 2026 lens: The AI buildout's true constraint — packaging & power, not GPU design (2024–2026)

Pack: TOC by Domain

DomainTypical constraintCommon errorTOC fix
ManufacturingSpecific machine/workstationOptimizing all stationsSubordinate rest to bottleneck
Software devCode review, QA, or deployPush devs to write fasterLimit WIP to constraint's rate
Sales funnelSpecific conversion stepAdd more top-of-funnel leadsFix conversion at the bottleneck
Hospital opsOR scheduling or dischargeAdd bedsFind true bottleneck (often discharge)
Project mgmtCritical task or shared resourcePer-task safety paddingCritical chain; project-level buffer

Applying It Well

  • Identify the constraint with evidence, not intuition.
  • Exploit and subordinate before elevating — most constraints yield without capital.
  • Retire local efficiency metrics at non-constraints; they systematically mislead.
  • Re-run Five Focusing Steps after every improvement — the constraint will move.

→ Primary sources: references/sources.md

Common Rationalizations

[D] = designed upfront | [O] = observed in real use. [O] entries are more valuable.

Fake moveReality
[D] "We need to fix all the problems"Fix the constraint only. Non-constraint improvements produce no system gain.
[D] "Everyone needs to work hard"Max output at non-constraints creates inventory, not throughput.
[D] "100% utilization everywhere"Mathematically false with variability. Non-constraints need slack.
[D] "Local efficiency = global efficiency"False in any multi-step system.
[D] "We don't have a constraint"Finite throughput = constraint exists. Find it.
[D] "More technology will solve it"Only if it addresses the constraint.
→ Add [O] entries here after each real use — paste the actual failure patternWhat went wrong and why

Red Flags

  • Diagnosis for throughput shortfall is "everyone needs to work harder"
  • Capacity investment spread across multiple steps without constraint identification
  • Inventory visibly accumulates in front of one step; no one flags it
  • Local productivity metrics tracked without aggregation to system throughput
  • Previous TOC gains have decayed (new constraint unmanaged)

Verification

  • System map drawn with capacity at each step
  • Constraint identified with evidence (not intuition)
  • Five Focusing Steps applied in order (exploit before elevate)
  • Non-constraint metrics that contradict system throughput retired
  • Next constraint identified after improvement; re-application scheduled

Part of deciqAI Knowledge Skills — 253 open-source thinking skills that make rigor executable for AI agents. The same skills power every deciqAI agent, which runs them autonomously to operate your company. See it run → https://www.deciqai.com/s/theory-of-constraints · Built by deciqAI · github.com/deciqAI · Contributions welcome.

Agents: latest version & machine-readable metadata → https://www.deciqai.com/s/theory-of-constraints.json

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.