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

04 — QUALITY: Kit Builder

Pass threshold: 90 / 100 When to run: After producing any kit (Mode 1 or Mode 3), or after improving a kit (Mode 2). Run against the kit you just produced, not the deliverable the kit will eventually produce.

Completeness (20 points)

#	Check	Points
1	All files present with correct naming: `NN-[kit-name]-[file-type].md` (or `.html`). File count matches the kit type decision (6 standard, more if justified, fewer if lightweight)	4
2	Directory created at `content/frameworks/[kit-name]/`	2
3	Start-here (00) has: what this is, audience, format, operating modes, file inventory, "does NOT do" section	4
4	Context (01) has: inputs per mode, validation rules, priority hierarchy	3
5	Terminology (02) has: locked terms, forbidden terms, visual states (if applicable)	3
6	Golden example (03) exists as one of: fully populated deliverable, structural reference to a live example, or explicit placeholder awaiting first deployment	4

Quality Gate Design (15 points)

#	Check	Points
7	Quality gate format is justified (point-scored, pass/fail, checklist, or interactive HTML — see QC Format Decision Guide in 01-context)	3
8	Checks organized by category (not a flat list)	2
9	Every check is specific and testable — no vague language like "appropriate" or "good"	4
10	Common failure modes section present with: what goes wrong, what happens, how to fix	3
11	If point-scored: point total adds to 100 with explicit pass threshold. If pass/fail: blocking failures identified. If checklist: ship criteria defined.	3

Output Skill Quality (20 points)

#	Check	Points
12	Output skill restates scope and required inputs (standalone-readable)	3
13	Content rules are numbered and specific	3
14	For HTML outputs: reusable component templates provided. For markdown: section specifications with format requirements	3
15	Full template or structural skeleton provided (in output skill, not in golden example)	4
16	Delivery checklist at the end (pre-ship gate)	2
17	Production steps are in the right order — you can follow them top to bottom	2
18	External QC dependencies documented if applicable (copy-qc.md, sentence-editor.md, brand QC)	3

Self-Improvement Loop (15 points)

#	Check	Points
19	Start-here includes Mode 2 (Improve) as an operating mode	3
20	Mode 2 trigger is clear: QC failure, manual changes, or system suggestions	3
21	Quality gate has a "Common Failure Modes" section (even if empty initially)	3
22	Change log convention documented (HTML comment or markdown section)	2
23	Output skill includes "update the golden example" instruction for post-production changes	2
24	Start-here self-improvement loop references the three questions: Did I change anything? Did QC miss something? Should the kit do more?	2

Consistency with Vault Conventions (15 points)

#	Check	Points
25	File naming follows `NN-[kit-name]-[file-type].md` (or `.html` for golden examples and interactive QC)	3
26	Kit name is lowercase, hyphenated	2
27	Does not duplicate logic from existing kits — references them instead	3
28	Relationship to other kits documented in start-here	2
29	Terminology is consistent with other vault kits (same terms for same concepts)	3
30	Voice matches vault conventions: direct, specific, no jargon, no fluff	2

Accuracy (15 points)

#	Check	Points
31	Golden example content matches what the output skill would produce (or placeholder is justified)	4
32	Quality checks are actually testable against the golden example (no checks that the example would fail)	3
33	Context inputs match what the output skill actually needs to run	3
34	Terminology definitions match how terms are actually used across all kit files	3
35	Operating mode descriptions match what the output skill actually does	2

Common Failure Modes

Failure	What Happens	How to Fix
Template disguised as golden example	File 03 has `{{PLACEHOLDER}}` tags instead of real content	Replace with a fully populated deliverable. Templates go in the output skill, not file 03. Exception: placeholder with structural specs is valid for kits awaiting first deployment.
Vague quality checks	"Tone is appropriate" instead of specific, testable criteria	Rewrite: "Zero instances of [specific forbidden terms]. Every [section type] follows [specific pattern]."
Output skill can't stand alone	File 05 references inputs or terms defined only in other files without restating them	Add scope section and required inputs list to file 05. It should be runnable without reading 00-02 first.
Missing "does NOT do"	Start-here doesn't set boundaries, so the kit gets used for things it wasn't designed for	Add explicit scope boundaries. List 3-5 things that are adjacent but not this kit's job.
Orphan kit	Kit has no documented relationship to other kits	Add relationship section to start-here. Every kit either derives from, references, coordinates with, or extends at least one other kit.
QC without failure modes	Quality gate scores outputs but doesn't document lessons learned	Add "Common Failure Modes" section — even if initially empty. This is where Mode 2 improvements land.
One-mode kit	Start-here defines only creation, no improvement path	Add Mode 2 (Improve) at minimum. Every kit must have a path to get better over time.
Forced 6-file structure	Kit has 6 files when it needs 9 (missing consultant methodology, gap protocol) or 2 (bloated with unnecessary files)	Match file count to kit type. Read existing kits of similar complexity before deciding.
Wrong QC format	Kit uses 100-point scoring when pass/fail with blocking failures would be more appropriate (or vice versa)	Evaluate: are some errors disqualifying regardless of score? → blocking failures. Are quality dimensions relatively weighted? → point-scored. Is the output simple? → checklist.
Missing external QC dependencies	Kit produces copy-heavy output but doesn't reference copy-qc.md or voice checks	Check whether the output includes prose, client-facing language, or brand voice. If yes, add external QC pass to the quality gate and output skill.
Shallow pattern extraction	Kit builder read 1-2 reference kits instead of all kits of similar type before building	Read all kits in the vault before producing any new kit. What looks like a universal pattern from 2 examples may be a local decision.