Builds evidence-only Kimball semantic layers for enterprise knowledge from APIs, CLIs, or tables/databases with an HTML decision workbench; emits OKF/YAML.
SkillScore87Maintained
Automatic
◆Data Engineer pack
Last verified 10/10/2026
What it changes
Passed 20 of 23 test cases with the skill, and 8 of 23 without it.
Without skill35%
With skill87%
Checks it passed that the model alone missed
Records review feedback in amendments.md.
Classifies POST /orders as a fact operation.
Derives concept IDs directly from interface paths.
and 14 more in the full report
12 fixed8 passed either way3 still failing0 broke
Pass-rate lift
+52 pp
Tokens
-35%
Agent turns
0%
Response time
-60%
Model gemini-3.6-flashMeasured 10/10/2026v220 of 23 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 12 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 12
Prompt
We are ready to emit artifacts for our user activity semantic layer without having specified any custom output preferences. Which bundle format should be produced by default?
What was checked, without and with the skill
Checked
Without skill
With skill
Selects Google OKF v0.1 bundle as the default output specification.
Missed
Passed
case-11 · gemini-3.6-flash · 10/10/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md49 lines8 files
Semantic KM Creator
Turn an interface contract—REST/OpenAPI, CLI, or table/DDL—into a governed semantic layer in four
strictly sequential phases. Advance only when the current phase meets its exit condition.
The interface is the sole source of truth. Never invent fields, grain, enums, or values. Stop on
missing evidence and ask one question at a time.
Workflow
Ingest (references/ingest.md) — Normalize interfaces into a complete operation inventory
and classify each operation as fact or dimension lookup with read/write status. Exit when every
operation is classified.
Review (references/review.md) — Decompose the model into atomic decisions. Show objective,
options, evidence, risks, dependencies, and tradeoffs, with fixed YAGNI guidance for facts,
dimensions, measures, and routing. Inject model JSON into assets/review-template.html; record
feedback in amendments.md and regenerate. Stop after the report. Exit only on approved:true
with no open or blocked decisions.
Emit — Default to a Google OKF v0.1 bundle using references/emit-okf.md; use repository
YAML only when explicitly requested, following references/emit-yaml.md. Both targets share
field semantics. Generate a thin root index for two or more bundles. Exit when every artifact
exists.
Verify (references/verify.md) — Run structural checks, semantic lint, and OKF hard
constraints when applicable. Report every pass and failure. Return non-mechanical failures to
Phase 2 with one blocking question.
Critical Rules
Evidence only — Unsupported fields, grain, enums, or values become TODO(verify) and an open
decision.
Grain first — Attach no dimensions or measures before grain is fixed.
One blocking ask — Stop only when an answer changes the model; ask one question with options.
Use the HTML review for bulk decisions.
Layer split — Shape, routing, and grain belong to the semantic layer. Values typed into
Verifies truthfulness, accuracy, and link integrity of content before publishing. Catches fabricated statistics, dead URLs, misattributed sources, and company claims that contradict the brand DNA.