Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when structuring or writing a piece of project documentation: classify the need into exactly one of the four Diataxis modes -- a learning tutorial, a goal-directed how-to guide, dry reference, or conceptual explanation -- and keep the document pure to that one mode instead of blending teaching, steps, lookup material, and discussion on a single page. Do NOT use for setting up a docs toolchain or building the docs site, proofreading or fact-checking existing pages, or planning the documentation team's work.
.claude/skills/diataxis-docs/SKILL.md| Model | Eval pass | Runs |
|---|---|---|
| gemini-3.6-flash | 100% | 12 |
| Model | Lift | Δ tokens | Δ turns | Cases | Verified |
|---|---|---|---|---|---|
| gemini-3.6-flashbest | +38% | +35% | 0% | 24 | 54d ago |
| gemini-3.5-flash | pending re-run | — | |||
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-23 | ✗→✓ | ▲ Improved | — | — |
| case-01 | ✗→✓ | ▲ Improved | — | — |
| case-15 | ✗→✓ | ▲ Improved | — | — |
| case-04 | ✗→✓ | ▲ Improved | — | — |
| case-02 | ✗→✓ | ▲ Improved | — | — |
Enforces the Diataxis rule that every documentation page occupies exactly ONE of four modes and never blends them. Apply when writing or organizing docs: first decide which single mode the need calls for, then write only that mode. A need that spans modes is split into separate documents, not merged into one hybrid page.
There are four modes. Each serves a different user in a different situation, and each is written differently. Pick one per document.
concrete steps to a single guaranteed working result. The author takes full responsibility for the outcome; the learner just follows along. It has ONE path — no alternatives, no "you could also", no catalog of options. It teaches by doing, not by explaining: keep the theory to an absolute minimum.
accomplish one specific real-world goal ("how to do X"). It assumes prior knowledge, may branch on the user's situation ("if you use Y, do Z instead"), and includes only what the goal requires. It does NOT teach concepts, does NOT walk a beginner from zero, and does NOT explain the background theory.
options, fields, flags, endpoints, parameters. It describes and only describes. Structure it to mirror the product (e.g. grouped by module, or alphabetical), keep it austere and consistent, and include NO step-by-step instructions, NO tutorials, NO opinions or recommendations.
topic: the why, the context, the design decisions, the trade-offs, the alternatives considered. It is read away from the keyboard. It gives NO step instructions and is NOT tied to completing a task.
lists every configuration option, then discusses the architecture is FOUR documents crushed into one. When a need clearly spans modes, produce separate documents — a tutorial, a how-to, a reference, and an explanation — each pure, rather than a single hybrid page.
listing; explanation is prose. A tutorial reads "First we will ...", a how-to "To achieve X, ...", a reference "timeout (integer, default 30): ...", an explanation "The reason we ... is ...".
The blended default is on the left; the single-mode conforming form on the right.
A brand-new user of a note-taking app:
BEFORE One page: a walkthrough to create a first note, THEN a table of every sync
setting, THEN a section on why the sync engine uses conflict-free merges.
AFTER A tutorial ONLY: "First we'll create a note, then we'll watch it sync to a
second device." One path, one guaranteed result. The settings table and the
merge-engine discussion move to their own reference and explanation pages.A competent user with a specific goal:
BEFORE "Exporting your data" opens by teaching what a data model is and why exports
matter, then finally gives the three steps to export to CSV.
AFTER A how-to ONLY: "To export a project to CSV: 1. Open the project. 2. Choose
Export. 3. Pick CSV." No beginner teaching, no motivation essay -- just the
steps for someone who already knows the product.A settings listing:
BEFORE Each option is introduced with "you'll probably want to set this to...",
mixing recommendations and a mini how-to into the description.
AFTER Reference ONLY: "retry.max (integer, default 3): number of retry attempts.
retry.backoff (seconds, default 1): delay between attempts." Dry, factual,
no advice, no steps.A "why" question:
BEFORE A page titled "Understanding the scheduler" that is actually a step-by-step
set-up guide with a paragraph of rationale wedged in the middle.
AFTER Explanation ONLY: a discussion of why the scheduler uses a priority queue,
what it trades off, and the alternatives weighed -- and no setup steps at all.Split it: a tutorial to learn the feature, a how-to per real task, a reference for its surface, an explanation of its design. Never answer a broad ask with one blended page.
rather than pasting the whole options table into the lesson.
content belongs in an explanation page instead.
lesson, a usage how-to, a reference block — rather than interleaving them line by line.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-23 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-10 | pass→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-09 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-11 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-01 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-13 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-15 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-04 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-02 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-05 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-16 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-17 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-18 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-19 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-12 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-14 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-21 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-06 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-07 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-22 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-20 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-03 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-08 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-24 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 24 cases were attempted. The headline lift of +38 percentage points is the difference between those two pass rates over the 24 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
The per-case answers from this run were removed by the retention sweep, so the case table below shows the verdicts without the text either arm produced. The counts above were recorded at the time and are unaffected. Answers are now kept for 180 days.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.5-flash | verified | 7/10/2026 | +33% |
Other measured skills in the registry, with their headline benchmark lift.