# Skill: Assemble and install a reproducible AI workflow system using three principles, a prompt library, and a one-habit-per-week onboarding plan

> Lets an AI agent help a user move from luck-based AI use to a system that delivers stable, repeatable
> quality; applies when a user wants to stop getting inconsistent results and build the habits and
> assets that make good output the default, not the exception.

Source: sodigi·learn - neuroslop/sistema-raboty-s-ii · https://sodigi.io/learn/neuroslop/sistema-raboty-s-ii

## When to use
- When a user reports inconsistent AI results: great one day, mediocre the next, with no idea why.
- When a user has learned individual techniques but wants to assemble them into a daily workflow.
- When a user wants to make their AI process teachable or transferable to a team member.
- When a user wants to build a compounding asset (prompt library) rather than one-off "genius prompts" that are forgotten.

## Core rules
- Reproducible quality is a system of habits, not luck with a good generation. The difference between "lottery" and "system" is habits drilled until automatic.
- The three pillars of the system:
  - Principle 1 - Context on the way in, always: two minutes before every request to define audience, facts, goal, tone. When context-free requests start feeling physically uncomfortable, the habit has taken.
  - Principle 2 - Quality criterion in advance: before starting, define "good enough" for this specific task in writing. This protects against both slop (first coherent answer = done) and perfectionism (never done).
  - Principle 3 - Iterate to the standard, accept by the criterion: done when the result passes the criterion from Principle 2, not when you are tired of editing.
- The prompt library is a compounding asset: working prompts, benchmark examples, and checklists saved once and reused; every next similar task starts at 80% done.
- The personal system prompt (who I am, my product, my audience, my data, my tone, my no-go list, how we work) should be filled in once, dropped at the start of every session, and updated monthly.
- Install one habit per week over four weeks. Trying to install everything at once installs nothing.
- The system's goal is "I think faster and sharper with AI", not "AI does it for me". Removing yourself from the thinking produces slop regardless of technique.
- Two assets accumulate simultaneously: operator skill (transfers across models and professions) and reputation (audiences recognize consistent specificity in an era of free swill).

## Procedure
1. CONFIRM the three principles are understood before building the system. If any of the five operator error scenarios from the course are still active, address them first.
2. SET UP the prompt library storage (a notes app, document, or folder - medium does not matter) with three sections: working prompts, benchmark examples, checklists.
3. FILL IN the personal system prompt (see template below). This takes 30 minutes and pays back on every subsequent task.
4. INSTALL habits one per week in order:
   - Week 1: context on the way in. Rule: no request without two-minute context prep. Nothing else changes.
   - Week 2: criterion and checklist. Before each task write one sentence defining "good enough". After each result run the five-question review checklist.
   - Week 3: targeted iterations. Ban "redo it" and "make it better". Every edit must name spot + problem + direction. Track iteration count; norm is 2-4 to standard.
   - Week 4: library. Add the week's best prompt, one benchmark example, and your checklist to the library. Maintain it as you work.
5. After four weeks, run the daily calibration test each evening: did I make decisions faster and sharper today because AI cleared grunt work, or would I be unable to explain why the published content says what it says? First = leverage. Second = slop.
6. Update the system prompt monthly: new numbers, shifted emphasis, updated stop-cliche list.

## Ready-to-use prompt
```
[PERSONAL SYSTEM PROMPT - fill in once, paste at the start of every session]

Who I am: [role and field. Example: co-founder of a kids' party studio in Austin, I run the marketing myself].

My product: [what you sell, to whom, what makes it different. Be specific - no "unique solutions"].

My audience: [who these people are, what they know, what they are afraid of, what language they use].

My data: [3-7 key numbers and facts usable in texts: metrics, timelines, case studies, review quotes].

My tone: [2-3 lines. Example: conversational and warm, no exclamation marks and no corporate-speak, light humor without clowning].

My no-go list: [what never to do: cliche phrases you catch yourself using, promises without facts, invented statistics].

How we work: show me a plan first - write the full text only after I approve. The first answer counts as a draft. Make edits point by point without rewriting the parts that work.

[END OF SYSTEM PROMPT]

---
[TASK FOR THIS SESSION]
[Describe your specific task here, adding any new facts or constraints not covered in the system prompt above.]
```

## Pitfalls
- Do not try to install all four habits in one week. The one-per-week schedule exists because simultaneous habit installation reliably fails under deadline pressure.
- Do not treat the prompt library as an archive of old prompts; it is an active working asset. A prompt that produced a strong result must be saved immediately, before it is forgotten.
- Do not skip the monthly system prompt update. A system prompt with outdated numbers or last year's product framing actively produces weaker results.
- Do not define "good enough" after the fact, when tired. Fatigue votes for "good enough as is" every time; the criterion must be set before work starts.
- Do not use the system to remove yourself from the thinking. "AI did it for me" is the formula for slop. "I thought faster with AI" is the formula for leverage. The operator's context, position, and judgment must remain in every output.
- Do not keep the system only in memory; the moment it is not written down, a deadline will override it. Checklist on a sticky note, system prompt saved as a template, criteria in a table - anything tangible beats anything mental.