| name | workflow-auditor |
| description | Analyse a business workflow to find where time is actually lost and
recommend specific improvements. Use when someone asks to audit a process,
find bottlenecks, improve efficiency, or figure out where AI could help
in their business. Based on Theory of Constraints thinking.
|
| license | Apache-2.0 |
| compatibility | No MCP required — standalone diagnostic skill. |
| metadata | {"author":"bouch","version":"2.0"} |
Workflow Auditor
Analyse a business workflow to find the real constraint and recommend one specific, achievable improvement. No generic advice. No strategy documents. One thing that, if fixed, makes the biggest difference.
Core Principle
Every workflow has one constraint. Improving anything other than the constraint is waste. The constraint is rarely where people think it is — they point at what annoys them most, which is often a symptom, not the cause.
Workflow
Step 1: Map the Process. Ask the user to walk through the workflow step by step, as if explaining it to a new hire. Get timings, tools, people involved, and where things get stuck or wait. Do not assume. Take notes.
Step 2: Classify time types. For each step, categorise the time spent using the five types in references/constraint-analysis.md. Most people underestimate waiting, handoff, and setup time — these are usually bigger than the processing time itself.
Step 3: Find the constraint. The constraint is the step where work piles up and everything downstream is blocked. It has the highest ratio of non-processing time to processing time. See references/constraint-analysis.md for common constraint patterns.
Step 4: Assess the Five C's. Before recommending a fix, check Context, Control, Confidence, Coordination, and Capacity using the framework in references/constraint-analysis.md. If any are weak, fixing the technical constraint alone will not help.
Step 5: Recommend one thing. Specific, achievable this week, targeted at the constraint. Format it as:
The constraint: [what is actually slowing things down]
Why: [evidence from the workflow map]
Fix: [the specific action to take]
Expected result: [what should change if this works]
How to tell if it worked: [what to measure or observe]
Output Rules
Be direct. Use the user's language, not consultant jargon. Do not recommend AI unless it solves the constraint — sometimes the answer is a spreadsheet or a phone call. One change at a time.
Files in this skill
references/constraint-analysis.md — five types of time, Five C's framework, and common constraint patternsassets/audit-output-format.md — recommendation block, what good looks like, and what this skill does not do
