Bottleneck-Skill: Theory of Constraints Diagnostic for Claude Code

Source: https://github.com/olelehmann1337/bottleneck-skill

Tweet: https://x.com/itsolelehmann/status/2049874169513590945?s=52

Found via: @telegram on 2026-05-03

Type: Claude Code skill / Prompt technique

Value Score: 4/5

Why not 5: It diagnoses the constraint but does not remove it — still requires Eli to execute the fix after the interview.

Depth: standard

Verdict: A well-engineered, immediately usable diagnostic prompt that belongs in Eli's Claude Code skills folder and should be run against both FJBC and Torah Media Agency this quarter.

What It Is

A single SKILL.md file for Claude Code that runs a 3-question Socratic business diagnostic based on Eliyahu Goldratt's Theory of Constraints. The skill forces the user to map their money-flow chain, identify where work piles up, and pass a numerical "lock-in test" before declaring a bottleneck real. It outputs a diagnosis card with the 5 focusing steps plus a 7-day experiment. The repo is new (31 stars, created 2026-04-29) by Ole Lehmann, who also built the well-regarded autoresearch-skill (789 stars).

GitHub Stats

The Whole Picture

Single technique. The content is one structured interview protocol packaged as a Claude skill. It has sequential steps (Q1 Chain → Q2 Pile-up → Q3 Lock-in Test) and a formatted output, but there is no feedback loop, no accumulating data layer, and no compounding mechanism — it is a high-quality one-shot diagnostic that can be re-run quarterly. The parts do not become more than the sum of their parts when run together; they are the single technique.

The Value

As parts

• The 3-question protocol → Run manually or via Claude Code against FJBC Website and Torah Media Agency to identify whether the real constraint is traffic, conversion, fulfillment, delivery capacity, or founder approval. The skill includes business-type-specific heuristics (e.g., "E-commerce: most often customer acquisition cost, fulfillment, or repeat purchase") that save time.
• The false-bottleneck pattern list → Add to OpenClaw's memory/topics/projects.md or AGENTS.md as a quarterly sanity check. The list ("We need more leads" → usually conversion is the bottleneck; "We need to hire" → founder hasn't built a system someone else can run; "We need better tools" → almost never the real constraint) is a concise heuristic set that matches OpenClaw's "judgment engine, not collection engine" thesis.
• The Lock-in Test question → Embed into Personal Brain Map as a decision filter: "If that step had infinite capacity starting tomorrow, and everything else stayed the same, would the outcome actually move within 90 days? By how much?" This is a transferable prioritization razor for any domain.
• Skill structure as template → The SKILL.md format (mandatory triggers, strong triggers, stated premise, step-by-step protocol with "do not paste all at once" pacing, diagnosis card output) is a reference for how Eli structures his own Claude Code skills (e.g., the gh-scout analysis skill or OpenClaw skills). Note the explicit trigger taxonomy and the pushback rules — these are higher-quality patterns than a simple instruction list.

Relevant Projects

• FJBC Website — E-commerce is explicitly covered by the skill's business-type heuristics. Eli is actively building newsletter + IG content; the skill would test whether the bottleneck is actually content volume or something downstream (conversion, repeat purchase, pricing).
• Torah Media Agency — The agency model is covered directly ("Agency: most often founder review/approval, sales pipeline, or delivery capacity"). The business plan describes a live GTM motion; this diagnostic would reveal whether sales pipeline or delivery capacity is the real constraint before scaling.
• OpenClaw — The skill can be dropped into ~/.claude/skills/ as-is. The structured Socratic interview pattern (probe, refuse lists, cross-check with live numbers, loop back on vague answers) is also a technique worth adapting for OpenClaw's conversational flows.
• Personal Brain Map — The false-bottleneck patterns are high-signal heuristics that map cleanly onto Eli's existing "shadow/scatter-fomo" and "decision framework gap" patterns. The Lock-in Test is a portable decision framework.

Already Doing This?

No existing project uses Theory of Constraints or structured bottleneck diagnosis. OpenClaw's morning-launchpad and eod-report do prioritization and anti-scatter scoring, but neither maps a revenue chain nor pressure-tests constraints with throughput math. The Torah Media Agency BUSINESS_PLAN.md describes the business model and pricing but contains no diagnostic for what's currently limiting growth. This skill fills a real gap.

How to Apply This

Parts to steal

1. Run the interview on FJBC this week. Start a Claude Code session with the skill active, answer the 3 questions for the jewelry business, and lock in one bottleneck. If the verdict is "repeat purchase" or "traffic conversion," that changes whether the current newsletter/IG work is the highest-leverage move.
2. Run the interview on Torah Media Agency before the next pitch. The operator-handoff deal is live; use the skill to confirm whether "sales pipeline" (closing sponsors) or "delivery capacity" (fulfilling for the first rabbi) is the true constraint. This determines whether Eli should be selling or building ops.
3. Add the false-bottleneck patterns to OpenClaw. Paste the symptom-disguised-as-constraint list into openclaw-config/memory/topics/projects.md or a new business-heuristics.md topic file so OpenClaw can surface them during quarterly planning conversations.
4. Copy the skill structure for future skills. The trigger taxonomy (mandatory vs. strong vs. do-not-trigger), the "ask one at a time" pacing rule, and the live-number cross-check are patterns Eli should reuse when writing SKILL.md files for gh-scout or OpenClaw.

Research leads (optional)

• Quarterly re-run cadence — The tweet suggests running this every quarter. Eli could add a recurring OpenClaw cron job (monthly or quarterly) that prompts him to re-run the bottleneck interview for each active business and logs the verdict to Obsidian. Not a "steal now" because it requires building a lightweight cron wrapper, but worth noting.

Bottom Line

Do not adopt whole — there is no whole system to adopt. Instead, install the skill into Claude Code, run it against FJBC and Torah Media Agency this week, and extract the false-bottleneck heuristics into OpenClaw's memory. The Lock-in Test question alone is worth the 10 minutes.