Source: frameworks/kit-micromagnet/04-micromagnet-quality.md

MicroMagnet Factory — QC Checklist

Run after building all assets. Every item is pass/fail. Fix failures before shipping.

1. MicroMagnet Criteria (Blocking — all 7)

[ ] Splinter test passed — solves ONE tiny persistent problem, describable in one sentence, core skill action under 2 minutes
[ ] Output test passed — produces a THING the user uses immediately, not a score or assessment
[ ] Output is NOT a diagnostic — no audits, assessments, gap analyses, readiness checks, or scores
[ ] Delivery vehicle identified — template, script, swipe file, cheatsheet, or calculator
[ ] Pre-AI precedent documented — people used to pay for this, price noted
[ ] Cannibalization check passed — doesn't overlap with Practice Builders, Sprint, or AOS deliverables
[ ] Zero-friction input — user already has the data, pasteable in under 2 minutes
[ ] Complete solution — user is done after running. No step 2 required.
[ ] The Gap named clearly — one adjacent problem the MicroMagnet deliberately leaves unsolved, pointing at a specific CTA destination

2. Audience-Stage Match (for source-driven intake)

[ ] If the source was a client conversation with a practitioner above target audience, every splinter was translated to what the target audience feels today
[ ] Splinters that didn't translate were killed, not stretched
[ ] The final micro-problem is something the target practitioner, at their current stage, feels today — not a future problem

3. Title Heat

[ ] Title includes a specific numeric or structural marker ("60-Second," "3-Question," "15-Minute," "4-Filter")
[ ] Title includes a strong action verb (Kills, Ends, Saves, Reveals, Turns, Catches, Stops)
[ ] Title names a specific pain or outcome — not "better results" or "optimal outcomes"
[ ] Framework name stays consistent across Niche and Universal versions (if Both)

4. Design Brief

[ ] Design brief completed before build started
[ ] Micro-problem stated in one sentence
[ ] Positioning named (Niche / Universal / Both) with N/A reason if Universal not viable
[ ] 3 jobs defined — each describes a user outcome
[ ] Output sections listed with what each contains
[ ] Core principle identified — the one rule that prevents garbage output (lives in the Kit, not in the shipped skill)
[ ] Nano spec named — which step of the cheat code extracts standalone
[ ] Win Stack check — hits at least 3 of 5 wins (Insight, Speed, Output, Confidence, Capacity)

5. Config Values

[ ] Offer name used in CTA PowerUp is filled in from Config — no [SET ME] placeholder shipped
[ ] Offer URL used in CTA PowerUp is filled in from Config — no [SET ME] placeholder shipped
[ ] If any Config value is still [SET ME], the build stopped and flagged before producing

6. YAML Frontmatter

[ ] name field present (lowercase, hyphenated)
[ ] description field is multi-line, includes what it does + when to use it
[ ] description includes triggers list with natural language variations
[ ] metadata block present with author, version, date
[ ] Version follows semantic format (1.0.0)

7. Layer Separation (Shipped Skill File)

[ ] Shipped skill file does NOT contain a Core Principle section
[ ] Shipped skill file does NOT contain a standalone Rules section (constraints moved inline into The Method)
[ ] Shipped skill file does NOT contain an embedded Quality Check table
[ ] Shipped skill file does NOT contain a What Makes This Different section
[ ] Shipped skill file opens prospect-facing — What You Paste / What You Get Back / The Method / Output / The Next Step

8. Register (Prospect-Facing)

[ ] Voice uses "you" and direct commands, not "the skill will evaluate" process language
[ ] Opens with what the user walks away with, not with internal structure
[ ] The Method section is instructional to Claude; the surrounding framing is instructional to the prospect — the distinction is clear

9. What You Paste / What You Get Back

[ ] "What You Paste" names accepted input types, tolerance for thin input, minimum viable input
[ ] "What You Get Back" is a numbered list of 3-5 deliverables (or fewer if the skill produces fewer — don't pad)

10. The Method

[ ] Method has 3+ named steps
[ ] Each step names its inputs AND its outputs
[ ] Constraints that would live in a separate Rules section live inline with the relevant step
[ ] Output template shows specific deliverables ("a 3-tier client assignment table with trigger thresholds"), not abstractions ("a report" or "a recommendation")
[ ] The whole method does not fit in 2 sentences — if it does, the depth is too thin

11. Output Template

[ ] Complete template with every section placeholder
[ ] Copy-pasteable — someone could fill in blanks by hand and get the right structure
[ ] Matches the section-by-section design brief exactly

12. The Next Step (CTA PowerUp)

[ ] 3 sentences maximum
[ ] Names the Gap — what this MicroMagnet doesn't solve
[ ] Destination matches the Config tilt or the brief's named destination
[ ] Ends in a link
[ ] Does not read as a sales pitch — it's a path to the next logical thing

13. Trademark

[ ] Framework name has ™ on first use and on every heading where it appears
[ ] Step names inside the framework do NOT have ™
[ ] ™ is on the framework, not its sub-components

14. Both-Positioning Divergence (when applicable)

[ ] If Both positioning, Niche and Universal versions share the identical framework name
[ ] Niche and Universal versions share the identical step count in The Method
[ ] The cheat code is unchanged across positionings — only framing shifts (title, splinter language, examples)
[ ] Divergence here (different framework, different step count) means the generator rewrote the method — fix before shipping

15. Nano

[ ] Nano produced in the same session as the MicroMagnet — not deferred
[ ] Nano represents a single step or pass of the MicroMagnet's cheat code, not the whole thing
[ ] Nano is smaller than the parent MicroMagnet — a nano that replicates the micro has failed
[ ] Nano has no CTA — no PowerUp block, no "next step" section
[ ] Nano works standalone — a prospect can use it without running the full MicroMagnet first

16. Confidentiality (for client-conversation sources)

[ ] All names scrubbed — client, team, firm, tool — from every section including Source Evidence
[ ] All identifying details paraphrased (industry made generic, financial figures rounded or removed, engagement specifics abstracted)
[ ] Zero direct quotes from the source client anywhere in the output
[ ] Examples in the shipped skill are paraphrased or generic

17. Testing

[ ] Skill tested with at least 3 different inputs
[ ] At least one minimal-input test (should trigger ask-for-more if threshold defined)
[ ] At least one detailed-input test
[ ] At least one edge case (different voice register, unusual format, missing data)
[ ] Output matches section spec across all tests
[ ] No AI-rhythm violations in any test output

18. Quick-Start Guide

[ ] Opens with "You just got a free Claude skill. Here's how to use it."
[ ] Prerequisites listed (Claude Pro, input description)
[ ] Step 1: Enable Code Execution — exact UI path
[ ] Step 2: Upload the Skill — exact button names
[ ] "Running the Skill" section with trigger phrase
[ ] Tips section — 3+ tips from real testing
[ ] "What to Paste" section — accepted input types as a list
[ ] Footer: Built by Kathryn Brown — Advisory OS

19. Copy QC (applies to all shipped output — skill file, quick-start, nano)

[ ] No "it's not X, it's Y" constructions (reversals)
[ ] No three-beat parallel structures
[ ] No mirrored two-beat constructions
[ ] Numbers rendered as digits (3 not three)
[ ] No adverbs
[ ] Sentences land on the strong word
[ ] No consulting-speak (leverage, synergy, optimal, journey, mindset)
[ ] Locked terms from 02-micromagnet-terminology.md used consistently
[ ] No hedging in the skill file instructions to Claude — "Do this," not "You might want to consider"

20. Voice Check

[ ] Voice matches anchor files in Config (register, sentence length, how the practitioner is addressed)
[ ] Locked terms used (skill, run, paste, install, practice, practice owner)
[ ] No forbidden constructions (check 02-micromagnet-terminology.md)
[ ] Output contains zero AI-rhythm patterns (reversal, forced parallelism, motivational closers)